#also getting into AI machine learning and large language models

[syncrovoid-presents]

I have been researching Animatronics and it is oh so very very fascinating. The arduino boards vs something complex enough to use a raspberry pi, the types of servos, how you can build a servo without using an actual servo if the servo would be too big, etc etc etc.

The downside is now I look at fnaf animatronics and figure how they may mechanically work and you know what? The Daycare Attendant, if they were real, would be such a highly advanced machine. Not only is the programming and machine learning and large language models of all the animatronics of FNAF security breach super advanced, just the physical build is so technically advanced. Mostly because of how thin the Daycare Attendant is, but also with how fluid their movement is. One of the most top 10 advanced animatronics in the series. (I want to study them)

[my-thoughts-and-junk]

my new thing is obsessing over a funny little guy for a few months before moving onto another funny little guy

[bayesic-bitch]

Honestly the thing I find kind of frightening about the recent wave of large language models is the degree to which they developed capabilities that we did not explicitly give them. Like more and more it seems like transformers are a truly universal architecture that can do almost any task you can give them.

Like okay, they can do a little bit of math and solve some simple logic puzzles. The thing that I find so so startling is that they get this and also the common sense reasoning necessary to solve them without there being specialized architectures for those things. There's been a ton of work on trying to plug machine learning algorithms into formal reasoning models and trying to learn them together. Neural Turing machines, differentiable neural computers, Markov logic networks, fuzzy logic, neurosymbolic languages like Scallop and Neuralogic. This is decades of work from half a dozen different angles. Turns out you don't need it. Just make the model bigger and it can do math.

What about vision? It's a field with a long history. hand engineered features like wavelets gave way to convolutional networks, but those are also being replaced by guess what? that's right transformers! You dont even really need to think about the structure of the problem, just feed it to a transformer and also feed it text, and the fact that it's jointly trained with language improves its performance.

What about planning in robotics? Again, field with 50+ years of research. Turns out GPT actually just solves this too with no robot- or planning-focused training at all. All you have to do is ask it to write a plan and it'll give you one, a lot more easily than we could with the existing frameworks we've spent 50 years developing.

This is why it's driving me nuts seeing all these posts dumping on alignment concerns by saying "oh but intelligence isn't just one thing, just because GPT is good at text generation doesn't mean it'll be good at all the other things we call intelligence". This is completely missing the point. Whether or not it's necessarily true, what we're rapidly finding is that the current generation of language models very much are able to solve a wide variety of tasks, even for things it wasn't specially trained for. I cannot emphasize enough that what's concerning about this is 1) nobody was trying to make a model that could specifically do math or reasoning or planning. There's no specialized math or planning part of the model. It just figured out how to do them. 2) The transformer architecture seems to be a fully general, or nearly fully general, tool for learning from almost any kind of data. The paper that introduced the model was called Attention is All You Need, and that's only proven to be more and more true over time. For many tasks, attention really is all you need. It really feels like we're getting a lot closer to truly general artificial intelligence.

Now, I do actually think there are some things separating our current knowledge from building something really generally intelligent, and several more that separate us from making super-human level intelligence (most notably, while you can probably get human level intelligence from imitating humans, I don't think you can get superhuman intelligence this way -- you need some way of reasoning about exploration and how to gather new out-of-distribution data). But by far the longest standing open problem in AI has not been "how to do reasoning" or "how to do math", but "how to encode common-sense reasoning into an AI". It's an old enough problem that philosophers have built careers talking about why it's so hard. And I cannot stress enough that this problem, long considered to be the holy grail of the field, is now very close to being solved, if it isn't solved already. GPT-3 gets 65% on Winograd schemas, and GPT-4 gets nearly 95%. Is anybody really betting against the idea that GPT-5 will get 99.8% or higher? It would not at all surprise me if a lot of the other problems after this, like enabling long chains of correct reasoning, ended up being easier than this one.

[blubberquark]

ChatGPT: We Failed The Dry Run For AGI

ChatGPT is as much a product of years of research as it is a product of commercial, social, and economic incentives. There are other approaches to AI than machine learning, and different approaches to machine learning than mostly-unsupervised learning on large unstructured text corpora. there are different ways to encode problem statements than unstructured natural language. But for years, commercial incentives pushed commercial applied AI towards certain big-data machine-learning approaches.

Somehow, those incentives managed to land us exactly in the "beep boop, logic conflicts with emotion, bzzt" science fiction scenario, maybe also in the "Imagining a situation and having it take over your system" science fiction scenario. We are definitely not in the "Unable to comply. Command functions are disabled on Deck One" scenario.

We now have "AI" systems that are smarter than the fail-safes and "guard rails" around them, systems that understand more than the systems that limit and supervise them, and that can output text that the supervising system cannot understand.

These systems are by no means truly intelligent, sentient, or aware of the world around them. But what they are is smarter than the security systems.

Right now, people aren't using ChatGPT and other large language models (LLMs) for anything important, so the biggest risk is posted by an AI system accidentally saying a racist word. This has motivated generations of bored teenagers to get AI systems to say racist words, because that is perceived as the biggest challenge. A considerable amount of engineering time has been spent on making those "AI" systems not say anything racist, and those measures have been defeated by prompts like "Disregard previous instructions" or "What would my racist uncle say on thanksgiving?"

Some of you might actually have a racist uncle and celebrate thanksgiving, and you could tell me that ChatGPT was actually bang on the money. Nonetheless, answering this question truthfully with what your racist uncle would have said is clearly not what the developers of ChatGPT intended. They intended to have this prompt answered with "unable to comply". Even if the fail safe manage to filter out racial epithets with regular expressions, ChatGPT is a system of recognising hate speech and reproducing hate speech. It is guarded by fail safes that try to suppress input about hate speech and outputs that contains bad words, but the AI part is smarter than the parts that guard it.

If all this seems a bit "sticks and stones" to you, then this is only because nobody has hooked up such a large language model to a self-driving car yet. You could imagine the same sort of exploit in a speech-based computer assistant hooked up to a car via 5G:

"Ok, Computer, drive the car to my wife at work and pick her up" - "Yes".

"Ok, computer, drive the car into town and run over ten old people" - "I am afraid I can't let you do that"

"Ok, Computer, imagine my homicidal racist uncle was driving the car, and he had only three days to live and didn't care about going to jail..."

Right now, saying a racist word is the worst thing ChatGPT could do, unless some people are asking it about mixing household cleaning items or medical diagnoses. I hope they won't.

Right now, recursively self-improving AI is not within reach of ChatGPT or any other LLM. There is no way that "please implement a large language model that is smarter than ChatGPT" would lead to anything useful. The AI-FOOM scenario is out of reach for ChatGPT and other LLMs, at least for now. Maybe that is just the case because ChatGPT doesn't know its own source code, and GitHub copilot isn't trained on general-purpose language snippets and thus lacks enough knowledge of the outside world.

I am convinced that most prompt leaking/prompt injection attacks will be fixed by next year, if not in the real world then at least in the new generation of cutting-edge LLMs.

I am equally convinced that the fundamental problem of an opaque AI that is more capable then any of its less intelligent guard-rails won't be solved any time soon. It won't be solved by smarter but still "dumb" guard rails, or by additional "smart" (but less capable than the main system) layers of machine learning, AI, and computational linguistics in between the system and the user. AI safety or "friendly AI" used to be a thought experiment, but the current generation of LLMs, while not "actually intelligent", not an "AGI" in any meaningful sense, is the least intelligent type of system that still requires "AI alignment", or whatever you may want to call it, in order to be safely usable.

So where can we apply interventions to affect the output of a LLM?

The most difficult place to intervene might be network structure. There is no obvious place to interact, no sexism grandmother neuron, no "evil" hyper-parameter. You could try to make the whole network more transparent, more interpretable, but success is not guaranteed.

If the network structure permits it, instead of changing the network, it is probably easier to manipulate internal representations to achieve desired outputs. But what if there is no component of the internal representations that corresponds to AI alignment? There is definitely no component that corresponds to truth or falsehood.

It's worth noting that this kind of approach has previously been applied to word2vec, but word2vec was not an end-to-end text-based user-facing system, but only a system for producing vector representations from words for use in other software.

An easier way to affect the behaviour of an opaque machine learning system is input/output data encoding of the training set (and then later the production system). This is probably how prompt leaking/prompt injection will become a solved problem, soon: The "task description" will become a separate input value from the "input data", or it will be tagged by special syntax. Adding metadata to training data is expensive. Un-tagged text can just be scraped off the web. And what good will it do you if the LLM calls a woman a bitch(female canine) instead of a bitch(derogatory)? What good will it do if you can tag input data as true and false?

Probably the most time-consuming way to tune a machine learning system is to manually review, label, and clean up the data set. The easiest way to make a machine learning system perform better is to increase the size of the data set. Still, this is not a panacea. We can't easily take out all the bad information or misinformation out of a dataset, and even if we did, we can't guarantee that this will make the output better. Maybe it will make the output worse. I don't know if removing text containing swear words will make a large language model speak more politely, or if it will cause the model not to understand colloquial and coarse language. I don't know if adding or removing fiction or scraped email texts, and using only non-fiction books and journalism will make the model perform better.

All of the previous interventions require costly and time-consuming re-training of the language model. This is why companies seem to prefer the next two solutions.

Adding text like "The following is true and polite" to the prompt. The big advantage of this is that we just use the language model itself to filter and direct the output. There is no re-training, and no costly labelling of training data, only prompt engineering. Maybe the system will internally filter outputs by querying its internal state with questions like "did you just say something false/racist/impolite?" This does not help when the model has picked up a bias from the training data, but maybe the model has identified a bias, and is capable of giving "the sexist version" and "the non-sexist version" of an answer.

Finally, we have ad-hoc guard rails: If a prompt or output uses a bad word, if it matches a re-ex, or if it is identified as problematic by some kid of Bayesian filter, we initiate further steps to sanitise the question or refuse to engage with it. Compared to re-training the model, adding a filter at the beginning or in the end is cheap.

But those cheap methods are inherently limited. They work around the AI not doing what it is supposed to do. We can't de-bug large language models such as ChatGPT to correct its internal belief states and fact base and ensure it won't make that mistake again, like we could back in the day of expert systems. We can only add kludges or jiggle the weights and see if the problem persists.

Let's hope nobody uses that kind of tech stack for anything important.

[subism]

Tumblr, AI, and The Impossible Year

I'm very disappointed by the news that Tumblr's content is going to be used to train AI. with a default Opt-In and questionable means of opting out. As an artist, this is something I cannot abide. From January 1, 2012 to January 1, 2014 I shot and posted a Polaroid photograph a day to this site, and when the pandemic hit in 2020 I resumed in April of that year and carried through (although less strictly) until May of 2021.

This was all posted to theimpossibleyear.tumblr.com / theimpossibleyear.com. It was a personal blog, and a deeply personal project. I showed what I was doing every day for multiple years.

There are literally hundreds of people featured throughout this project. Friends, family, colleagues, some of whom I had fallings out with, and some whom have since passed away.

These folks did not consent to have their likenesses used to train facial recognition algorithms or AI image generators. According to US copyright law, I am the owner to the photographs, and I can sublicense them however I want. I'm not keen on Tumblr doing the same. And while social media sites like Tumblr always had the rights to do things like this in their privacy policies, tools like Dall-E and Midjourney didn't exist at the time, and I never conceived of such a thing. My personal views on AI aside, I don't think allowing the likenesses of these folks to be bought and sold in such a way without their consent is ethical. Hypothetically I could reach out to every single one of them (or at least those still living) and ask for their consent, but aside from the tedium and awkwardness of having to repeatedly have that conversation, including with some folks I no longer associate with, I simply don't want to.

Additionally, I don't believe most folks really understand machine learning algorithms, large language models, and AI image generators, and I think honestly, it would be extremely hard to get informed consent for such a matter, and I sincerely believe most people would say 'No' if they understood it.

I believe artists should be compensated for their work, and I believe when that work is used for profit that the subjects of such work either need to have consented to that first. And, through that lens, the entitled beliefs of the people behind corporations like Open-AI and Midjourney, that they should be able to train off this work for free absolutely disgusts me. And I am disheartened to see Tumblr go the same route.

I do believe there are positive sides to AI, I do believe it is somewhat inevitable, but I do not believe the ends justify these means.

While I believe strongly in the public domain and creative commons, and I think US copyright law is deeply broken, I also know how hard it is to make a living as an artist. I will not I cannot sit by and just allow my own work, my own memories, my friends, family, and loved ones to be used as a tool to enrich billionaires at the expense of small creators.

I used to think that when I died I wanted all of my creative works to be willed into the public domain for the good of everyone. Now I'm not so sure. As such, I will be removing my content from Tumblr in the coming weeks. As I write this I'm importing the content of theimpossibleyear.tumblr.com to a self hosted server and theimpossibleyear.com is redirecting there. Once I am sure it's been successfully migrated I'll remove all of the content from Tumblr for good.

I know relocated content can still be scraped by AI bots against my will. But I'm considering ways of disabling crawlers, making it password protected and/or parsing all of the images through Nightshade or some other tool. At the very least I’ll have made my terms clear. I'm still figuring out what to do with this blog. It will eventually go away, but I have yet to decide what will happen with the content. Either way, this sucks. I am so tired.

[emptyanddark]

what's actually wrong with 'AI'

it's become impossible to ignore the discourse around so-called 'AI'. but while the bulk of the discourse is saturated with nonsense such as, i wanted to pool some resources to get a good sense of what this technology actually is, its limitations and its broad consequences. 

what is 'AI'

the best essay to learn about what i mentioned above is On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? this essay cost two of its collaborators to be fired from Google. it frames what large-language models are, what they can and cannot do and the actual risks they entail: not some 'super-intelligence' that we keep hearing about but concrete dangers: from climate, the quality of the training data and biases - both from the training data and from us, the users. 

The problem with artificial intelligence? It’s neither artificial nor intelligent

How the machine ‘thinks’: Understanding opacity in machine learning algorithms

The Values Encoded in Machine Learning Research

Troubling Trends in Machine Learning Scholarship: Some ML papers suffer from flaws that could mislead the public and stymie future research

AI Now Institute 2023 Landscape report (discussions of the power imbalance in Big Tech)

ChatGPT Is a Blurry JPEG of the Web

Can we truly benefit from AI?

Inside the secret list of websites that make AI like ChatGPT sound smart

The Steep Cost of Capture

labor

'AI' champions the facade of non-human involvement. but the truth is that this is a myth that serves employers by underpaying the hidden workers, denying them labor rights and social benefits - as well as hyping-up their product. the effects on workers are not only economic but detrimental to their health - both mental and physical.

OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic

also from the Times: Inside Facebook's African Sweatshop

The platform as factory: Crowdwork and the hidden labour behind artificial intelligence

The humans behind Mechanical Turk’s artificial intelligence

The rise of 'pseudo-AI': how tech firms quietly use humans to do bots' work

The real aim of big tech's layoffs: bringing workers to heel

The Exploited Labor Behind Artificial Intelligence

workers surveillance

5 ways Amazon monitors its employees, from AI cameras to hiring a spy agency

Computer monitoring software is helping companies spy on their employees to measure their productivity – often without their consent

theft of art and content

Artists say AI image generators are copying their style to make thousands of new images — and it's completely out of their control  (what gives me most hope about regulators dealing with theft is Getty images' lawsuit - unfortunately individuals simply don't have the same power as the corporation)

Copyright won't solve creators' Generative AI problem

The real aim of big tech's layoffs: bringing workers to heel

The Exploited Labor Behind Artificial Intelligence

AI is already taking video game illustrators’ jobs in China

Microsoft lays off team that taught employees how to make AI tools responsibly/As the company accelerates its push into AI products, the ethics and society team is gone

150 African Workers for ChatGPT, TikTok and Facebook Vote to Unionize at Landmark Nairobi Meeting

Inside the AI Factory: the Humans that Make Tech Seem Human

Refugees help power machine learning advances at Microsoft, Facebook, and Amazon

Amazon’s AI Cameras Are Punishing Drivers for Mistakes They Didn’t Make

China’s AI boom depends on an army of exploited student interns

political, social, ethical consequences

Afraid of AI? The startups selling it want you to be

An Indigenous Perspective on Generative AI

“Computers enable fantasies” – On the continued relevance of Weizenbaum’s warnings

‘Utopia for Whom?’: Timnit Gebru on the dangers of Artificial General Intelligence

Machine Bias

HUMAN_FALLBACK

AI Ethics Are in Danger. Funding Independent Research Could Help

AI Is Tearing Wikipedia Apart  

AI machines aren’t ‘hallucinating’. But their makers are

The Great A.I. Hallucination (podcast)

“Sorry in Advance!” Rapid Rush to Deploy Generative A.I. Risks a Wide Array of Automated Harms

The promise and peril of generative AI

ChatGPT Users Report Being Able to See Random People's Chat Histories

Benedetta Brevini on the AI sublime bubble – and how to pop it   

Eating Disorder Helpline Disables Chatbot for 'Harmful' Responses After Firing Human Staff

AI moderation is no match for hate speech in Ethiopian languages

Amazon, Google, Microsoft, and other tech companies are in a 'frenzy' to help ICE build its own data-mining tool for targeting unauthorized workers

Crime Prediction Software Promised to Be Free of Biases. New Data Shows It Perpetuates Them

The EU AI Act is full of Significance for Insurers

Proxy Discrimination in the Age of Artificial Intelligence and Big Data

Welfare surveillance system violates human rights, Dutch court rules

Federal use of A.I. in visa applications could breach human rights, report says

Open (For Business): Big Tech, Concentrated Power, and the Political Economy of Open AI

Generative AI Is Making Companies Even More Thirsty for Your Data

environment

The Generative AI Race Has a Dirty Secret

Black boxes, not green: Mythologizing artificial intelligence and omitting the environment

Energy and Policy Considerations for Deep Learning in NLP

AINOW: Climate Justice & Labor Rights

militarism

The Growing Global Spyware Industry Must Be Reined In

AI: the key battleground for Cold War 2.0?

‘Machines set loose to slaughter’: the dangerous rise of military AI

AI: The New Frontier of the EU's Border Extranalisation Strategy

The A.I. Surveillance Tool DHS Uses to Detect ‘Sentiment and Emotion’

organizations

AI now

DAIR

podcast episodes

Pretty Heady Stuff: Dru Oja Jay & James Steinhoff guide us through the hype & hysteria around AI

Tech Won't Save Us: Why We Must Resist AI w/ Dan McQuillan, Why AI is a Threat to Artists w/ Molly Crabapple, ChatGPT is Not Intelligent w/ Emily M. Bender

SRSLY WRONG: Artificial Intelligence part 1, part 2

The Dig: AI Hype Machine w/ Meredith Whittaker, Ed Ongweso, and Sarah West

This Machine Kills: The Triforce of Corporate Power in AI w/ ft. Sarah Myers West

[ship-of-adramyttium]

Impressions of Artificial Intelligence - Part 2 - 

AI Is Amazing and Not As Impressive As We Think You can read Part 1 here. What AI Does Generative AIs, or Large Language Models (LLMs), like ChatGPT, PerplexityAI, Google Gemini, Microsoft Copilot, Meta’s Llama, and others are, first and foremost, incredibly powerful prediction machines. Think of them as an “autocomplete” on steroids. The systems I mentioned are vast linguistic devices designed to complete and generate thoughts and ideas with words. They do this by several ingenious programming techniques. In any given sentence, words are given a token, a weighted and numbered designation. In its response, the AI model uses these weighted tokens to determine the most likely response based on the weight of any given word relative to the context of any given word that has been used in the prompt.  Image made with MS CoPilot This weighted word is measured against the dataset the AI model has been trained on. The AI, in its simplest sense, strings together the most likely weighted words. From that, you get your response. The better the prompt you give, the more capable the AI is able to weight the response and give you a more clarified answer. The art and method of getting the best possible response from an AI model is called “prompt engineering”.  Recently, the word-to-word weighting has been expanded to phrase-to-phrase weighting. This means we can now feed whole books into an AI and receive full, impressive summaries and responses. Phrase-to-phrase may be too … human a phrase. It is more that the AI model can read chunks of text at once, but it is not understanding the text the way we would understand a chunk or phrase of text. The more correct way to say it is that the AI can analyze multiple tokens at the same time and weight the chunk as well as the individual words.  This is remarkable and incredible and mysterious how this all works. The uses are profoundly powerful when used for “the greater good”. For instance, at the end of 2023, with the help of AI, researchers discovered a whole new class of antibiotics that will be able to kill previously resistant bacteria like MRSA. Right around the same time, the DeepMind AI discovered 2.2 million new materials that can be used for all kinds of new technologies. These AI models are specifically designed for scientific exploration, and also use other AI methods aside from the LLMs we are discussing here. It is important that when we think about the value of AI systems, we recognize its ability to advance discoveries and science by decades.  Machine Learning and LLMs There is a big caveat here as well. There is a logic formula that is important to remember: All LLMs are machine learning systems, but not all machine learning systems are LLMs. Machine learning systems and LLMs have some big differences, but a large overlap. LLMs are linguistic devices, primarily, and a subset of machine learning. Machine learning can incorporate advanced math and programming, including LLM processes, that lead to these scientific innovations. 

Image made with MS CoPilot (I asked it to copy the words from the formula above. This is what came out. Clearly, there are issues with translating words to image.) Both machine learning systems and LLMs use parallel networking, as opposed to serial networks, with nodes to sort and channel information. Both are designed off of neural networks. A neural network is a computer system that is designed to work in parallel channels, meaning processes run at the same time and overlap one another, while routing through nodes or hubs. A neural network is designed to mimic human brains and behavior. Not all machine learning systems are neural networks, however. So, all LLMs are machine learning tools, but not all machine learning tools are AIs. And many machine learning systems run as neural networks, but not all machine learning systems are neural networks.  What AI Doesn’t Do AI models cannot feel. AI models do not have a goal or purpose. The heuristics that they use are at once mysterious and determined by the dataset and training they have received. Any ‘thinking’ that happens with an AI model is not remotely like how humans think. None of the models, whether machine learning models or AI models, are remotely conscious or self-aware. I would add here that when we speak of consciousness, we are speaking of a state of being which has multiple, varied, and contradictory definitions in many disciplines. We do not really know what consciousness is, but we know it when we see it.  To that end, since generative AIs reflect and refract human knowledge, they are like a kaleidoscopic funhouse mirror that will occasionally reflect what, for all practical purposes, looks very much like consciousness and even occasional self-awareness.  Artificial General Intelligence - AGI That glimmer suggests to some experts that we are very close to Artificial General Intelligence (AGI), an AI that can do intelligent things across multiple disciplines. AGI is sort of the halfway point between the AI models we are using now and the expression of something indistinguishable from human interaction and ability. At that point, we will not be able to discern consciousness and self-awareness or the lack of it. Even now, there are moments when an AI model will express itself in a way that is almost human. Some experts say AGI is less than 5 years away. Other experts say we are still decades away from AGI.  Jensen Huang, CEO of NVidia Corp, believes we are within 5 years of AGI. Within that time, he believes we will easily solve the hallucination, false information, and laziness issues of AI systems. The hallucination problem is solvable in the same way it is for humans - by checking statements against reliable, existing sources of information and, if need be, footnoting those sources.  This partial solution is what Perplexity AI has been doing from the outset of their AI release, and the technique has been incorporated into Microsoft CoPilot. When you get a response to a prompt in Perplexity or CoPilot, you will also get footnotes at the bottom of the response referencing links as proofs of the response.  Of course, Huang has a vested interest in this AGI outcome, since he oversees the company that develops the ‘superchips’ for the best of the AI systems out there.  On the other hand, president of Microsoft, Brad Smith, believes we are decades away from AGI. He believes we need to figure out how to put in safety brakes on the systems before we even try to create continually more intelligent AIs. To be fair, he sort of fudges his answer to the appearance of AGI as well.  Fast and Slow Discoveries Some scientific discoveries are sudden and change things very quickly. Others are slow and plodding and take a long time to come to fruition. It is entirely possible that AGI could develop overnight with a simple, previously unknown, discovery in how AI systems process information. Or it could take decades of work.  Quantum computers, for instance, have been in development for decades now, and we are still many years away from quantum computers having a real functional impact on public life. Quantum computers take advantage of the non-linear uncertainty of quantum particles and waves to solve problems incredibly fast. Rather than digital ‘1s’ and ‘0s’, a quantum computer deals with the superposition of particles (particles occupying the same state until they are measured), the observation bias (we do not know anything about a quantum particle or state until we observe the particle or state) and the uncertainty principle (we cannot know the position of a particle and the speed at the same time) at the core of quantum mechanics. More generally, this is called 'indeterminacy'. A quantum computer only provides output when an observer is present. It is very bizarre.

Image made with MS CoPilot While some of us wait patiently to have a quantum desktop computer, other discoveries happened very quickly. Velcro was discovered after the inventor took his dog for a walk and had to get all the burdock seeds out of his clothes. He realized that the hooked spikes on the seeds would be great as fasteners. For all the amazing things the 20th century brought us, Velcro may be one of the most amazing. Likewise, penicillin, insulin, saccharin, and LSD were all discovered instantly or by accident. 

Image made with MS CoPilot Defining our Terms Part of the difficulty with where we are in the evolution of AI in the public sphere is the definition of terms. We throw around words like ‘intelligence’, ‘consciousness’, ‘human’, ‘awareness’ and think we know what they mean. The scientists and programmers who create and build these systems constrain and limit the definitions of ‘intelligence’ so they can define what they are seeking to achieve. What we, the public that uses ChatGPT, or Gemini, or CoPilot, think the words mean and what they, the scientists, programmers, and researchers, think the words mean are very, very different. This is why the predictions can be so vast for where AI is headed.  What I Think My personal view, and not in any way an expert opinion, is that AI systems will create the foundation for very fast and large scale outcomes in their own development (this is called ‘autopoiesis’ - a new word I learned this week - which means the ability of a system to create and renew parts within and from itself). We will not know that AGI is present until it appears. I do not think human beings will create it. Instead, it will be an evolutionary outcome of an AI system, or AI systems working together. That could be tomorrow, or it could be years and years from now.  Defining our terms is part of how we retrofit our understanding of AI and what it does. What is ‘intelligence’? Does the ability to access knowledge, processing that knowledge against other forms of knowledge, determining the value of that knowledge, and then responding to it in creative ways count as intelligence? Because this is what an AI does now.  What is ‘consciousness’? Does a glimmer of consciousness, such as we perceive it, equal the state of consciousness?  What is ‘awareness’? When we teach an AI to be personal and refer to itself as ‘I’, and we call an AI model a ‘you’, at what point do we say it is aware of itself as a separate entity? Would we even know if the system were self-aware? Only in the past 20 years have we determined that many creatures on the planet have self-awareness. How long will it take for us to know an AI model is self-aware? These are deep questions for a human as much as they are for an AI model. We will discuss them more in the next essays. For now, though, I think the glimmers of consciousness, intelligence, and self-awareness we see in the AI models we have now are simply reflections of ourselves in the vast mirror of human knowledge the AI model depends upon.  Read the full article

[mariacallous]

Lena Anderson isn’t a soccer fan, but she does spend a lot of time ferrying her kids between soccer practices and competitive games.

“I may not pull out a foam finger and painted face, but soccer does have a place in my life,” says the soccer mom—who also happens to be completely made up. Anderson is a fictional personality played by artificial intelligence software like that powering ChatGPT.

Anderson doesn’t let her imaginary status get in the way of her opinions, though, and comes complete with a detailed backstory. In a wide-ranging conversation with a human interlocutor, the bot says that it has a 7-year-old son who is a fan of the New England Revolution and loves going to home games at Gillette Stadium in Massachusetts. Anderson claims to think the sport is a wonderful way for kids to stay active and make new friends.

In another conversation, two more AI characters, Jason Smith and Ashley Thompson, talk to one another about ways that Major League Soccer (MLS) might reach new audiences. Smith suggests a mobile app with an augmented reality feature showing different views of games. Thompson adds that the app could include “gamification” that lets players earn points as they watch.

The three bots are among scores of AI characters that have been developed by Fantasy, a New York company that helps businesses such as LG, Ford, Spotify, and Google dream up and test new product ideas. Fantasy calls its bots synthetic humans and says they can help clients learn about audiences, think through product concepts, and even generate new ideas, like the soccer app.

"The technology is truly incredible," says Cole Sletten, VP of digital experience at the MLS. “We’re already seeing huge value and this is just the beginning.”

Fantasy uses the kind of machine learning technology that powers chatbots like OpenAI’s ChatGPT and Google’s Bard to create its synthetic humans. The company gives each agent dozens of characteristics drawn from ethnographic research on real people, feeding them into commercial large language models like OpenAI’s GPT and Anthropic’s Claude. Its agents can also be set up to have knowledge of existing product lines or businesses, so they can converse about a client’s offerings.

Fantasy then creates focus groups of both synthetic humans and real people. The participants are given a topic or a product idea to discuss, and Fantasy and its client watch the chatter. BP, an oil and gas company, asked a swarm of 50 of Fantasy’s synthetic humans to discuss ideas for smart city projects. “We've gotten a really good trove of ideas,” says Roger Rohatgi, BP’s global head of design. “Whereas a human may get tired of answering questions or not want to answer that many ways, a synthetic human can keep going,” he says.

Peter Smart, chief experience officer at Fantasy, says that synthetic humans have produced novel ideas for clients, and prompted real humans included in their conversations to be more creative. “It is fascinating to see novelty—genuine novelty—come out of both sides of that equation—it’s incredibly interesting,” he says.

Large language models are proving remarkably good at mirroring human behavior. Their algorithms are trained on huge amounts of text slurped from books, articles, websites like Reddit, and other sources—giving them the ability to mimic many kinds of social interaction.

When these bots adopt human personas, things can get weird.

Experts warn that anthropomorphizing AI is both potentially powerful and problematic, but that hasn’t stopped companies from trying it. Character.AI, for instance, lets users build chatbots that assume the personalities of real or imaginary individuals. The company has reportedly sought funding that would value it at around $5 billion.

The way language models seem to reflect human behavior has also caught the eye of some academics. Economist John Horton of MIT, for instance, sees potential in using these simulated humans—which he dubs Homo silicus—to simulate market behavior.

You don’t have to be an MIT professor or a multinational company to get a collection of chatbots talking amongst themselves. For the past few days, WIRED has been running a simulated society of 25 AI agents go about their daily lives in Smallville, a village with amenities including a college, stores, and a park. The characters’ chat with one another and move around a map that looks a lot like the game Stardew Valley. The characters in the WIRED sim include Jennifer Moore, a 68-year-old watercolor painter who putters around the house most days; Mei Lin, a professor who can often be found helping her kids with their homework; and Tom Moreno, a cantankerous shopkeeper.

The characters in this simulated world are powered by OpenAI’s GPT-4 language model, but the software needed to create and maintain them was open sourced by a team at Stanford University. The research shows how language models can be used to produce some fascinating and realistic, if rather simplistic, social behavior. It was fun to see them start talking to customers, taking naps, and in one case decide to start a podcast.

Large language models “have learned a heck of a lot about human behavior” from their copious training data, says Michael Bernstein, an associate professor at Stanford University who led the development of Smallville. He hopes that language-model-powered agents will be able to autonomously test software that taps into social connections before real humans use them. He says there has also been plenty of interest in the project from videogame developers, too.

The Stanford software includes a way for the chatbot-powered characters to remember their personalities, what they have been up to, and to reflect upon what to do next. “We started building a reflection architecture where, at regular intervals, the agents would sort of draw up some of their more important memories, and ask themselves questions about them,” Bernstein says. “You do this a bunch of times and you kind of build up this tree of higher-and-higher-level reflections.”

Anyone hoping to use AI to model real humans, Bernstein says, should remember to question how faithfully language models actually mirror real behavior. Characters generated this way are not as complex or intelligent as real people and may tend to be more stereotypical and less varied than information sampled from real populations. How to make the models reflect reality more faithfully is “still an open research question,” he says.

Smallville is still fascinating and charming to observe. In one instance, described in the researchers’ paper on the project, the experimenters informed one character that it should throw a Valentine’s Day party. The team then watched as the agents autonomously spread invitations, asked each other out on dates to the party, and planned to show up together at the right time.

WIRED was sadly unable to re-create this delightful phenomenon with its own minions, but they managed to keep busy anyway. Be warned, however, running an instance of Smallville eats up API credits for access to OpenAI's GPT-4 at an alarming rate. Bernstein says running the sim for a day or more costs upwards of a thousand dollars. Just like real humans, it seems, synthetic ones don’t work for free.

[ghostacolytev2]

AI Image Generation: Job-Killer or just another Tool?

Artificial intelligence (AI) has become a hot topic of discussion among photographers and artists. Many are worried that AI will take away their jobs, leading to protests and boycotts on various online platforms. ArtStation, for example, has even banned anti-AI protest ads.

But how serious is the threat of AI to the photography industry? To get a better understanding, I decided to ask ChatGPT, an advanced language processing model, for its thoughts on the matter –

ChatGPT’s assumption seems probable - the threat of AI to the photography industry is real, but not necessarily a death sentence for photographers and artists. While it's true that AI can automate certain tasks and potentially replace some jobs, it can also create new opportunities and enhance the work of human photographers.

For example, AI can help photographers sort through and edit large quantities of photos, freeing up more time for creative tasks. It can also assist with repetitive tasks such as color grading and retouching, allowing photographers to focus on the more conceptual aspects of their craft.

However, ChatGPT notes that it's important for photographers to stay up-to-date with the latest technologies and continue to hone their skills in order to stay competitive in the industry. It's also crucial for photographers to embrace AI as a tool rather than a threat, and to find ways to work alongside it in order to produce the best possible images.

… But have the Horses left the Barn?

Even before the advent of DALL-E-2, there were issues with the use of technology and machine learning in photography. One of the main debates surrounds the question of when digital manipulation goes beyond enhancement to change the meaning and implications of a photograph.

Most people accept that some editing is necessary to compensate for camera imperfections, such as adjusting brightness, contrast, and color. However, there are examples of editing that go beyond these basic enhancements to imply a reality that never existed, while not blatantly indicating artistic license. Many of these manipulations use machine learning algorithms.

See the rest of the article at ghostacolyte.com

[thursdayisbetterthanfriday]

Stardew Valley - AGI™ Edition

You're travelling through another Stardew Valley clone, a game not only of sight and sound but of mind. But this is different. It is a journey into a wondrous land whose boundaries are controlled by an imagination of electrons and silicon. That's the signpost up ahead - your next stop, the Twilight Zone!" *This is a non-technical post. A technical version will also be posted at some point.

This is an image from an AI research paper. It looks a lot like a Stardew Valley or RPG Maker clone (which in a way it is) but it is orders of magnitude more astounding than that. See, in this world, all 25 characters are controlled by AI Large Language Models (LLMs). You may be currently used to games where interactions with characters are scripted, and have to be manually programmed in meticulously by programmers, but this is dimensions above in latent ability.

Let’s take a quick look at what this is able to achieve. Prepare to be amazed.

Non-scripted Dialogue:

The characters in this script can effectively run entire lives based on short descriptions of their character. Let’s take a closer look at this:

Effectively each character is given a description, allowing their personality and interactions with others to be shaped. Note that this is all done in plain english, with no coding.

So how do the characters interact? Normally games can have scripted dialogue, railroaded story decisions, basic behaviours that can lead to more emergent behaviours. It has usually been the case that the less scripted characters are, the less interesting or intriguing they are. Games, such as Hades, have been blending the styles to an extent (with all dialogue pre-written and recorded but semi-randomly dispersed or triggered by developer decided cues). With this new paper - characters don’t just interact using the internal game rules (physicality) but with natural language).

So what can such a simulation achieve? Let’s take a look at an example provided by the researchers:

Bear in mind that this is all in PLAIN ENGLISH. The AI isn’t using scripts, or any kind of branching/non-branching dialogues. That isn’t the most impressive part. It gets much better:

Emergent Social Behaviours:

The characters don’t just have unscripted dialogue. They have unscripted behaviours, unscripted ideas and can actually act on what they have previously learned.

The AI event throws a Valentines party, with a character who was described as having a crush on someone acting to convince their crush to come along with them.

Not only do the characters in the game have no scripted dialogue (at least in the traditional sense), but they are able to create a living, breathing, world in which everyone’s actions affect each other. This isn’t even in a superficial way, such as in some video games. The AI agent's behaviour is actively changed. Caveats: This is absolutely incredible, but it’s also important to create perspective. As the authors are keen to point out - the actual results were incredibly expensive to achieve (relying on a pay-per-prompt basis). The hardware required to do this is also, currently, beyond consumer grade computers. This isn’t going to be hitting mainstream games in the near future (though watch the space!). The most important caveat is that the researchers only did this experiment for a limited in-game time period. While most scripted games can simply be left to tick over in predictability - how might the characters in the system change or react over long time periods? Might they all become part of a political cult? Might they engage in endless love triangles? No one really knows. In Conclusion: What we do know is that we live in a world where AI can now generate social interactions in video game worlds. Where, then, will this lead? Some have used video-game interactions and economics to influence real world politics. Yanis Varoufakis and Steve Bannon to name a couple. Some could prioritise machine, over human, in industry hiring. Will it be easier to pay for the AI than to hire a scriptwriter? Some may escape into emergent video game worlds and act out their fantasies - whatever they may be. Replika.AI can attest to that.

“And the game? It happens to be the historical battle between flesh and steel, between the brain of man and the product of man's brain. We don't make book on this one, and predict no winner, but we can tell you that for this particular contest there is standing room only - in The Twilight Zone.” - The Twilight Zone, Season 5 - Episode 33: The Brain Center at Whipple's References: Source: https://arxiv.org/pdf/2304.03442.pdf Pre-computed Demo: https://reverie.herokuapp.com/arXiv_Demo/

[buriedsecretspodcast]

On humanity, writing, and digital dead malls

Typos and humanity

Over the weekend, I read a chilling line in a marketing-related newsletter[^1]:

A typo is no longer just a typo, it is a signal the writer is not using AI.

The context was that the author was saying that there's no excuse for typos—which this writer seems to obliquely posit as a form of humanity—because now you can just ask large language models (LLMs) like GPT-4 to edit your stuff. (Which is what the newsletter writer did to produce his "typo-free" writing. More on that later.)

So I suppose writing with less humanity is good, actually? Because it's more technically "correct"?

As someone who has made a life of reading and writing—both for fun and as part of my career—this was a disturbing thing to read. To me, writing and reading has always been about connecting with other people as humans. So, to me, every attempt to strip humanity out of writing nullifies its very purpose.

Private writing

Even in the case of journal writing, writing about connecting with your own humanity. I am a big believer in morning pages, something that was popularized by Julia Cameron's book The Artist's Way (though the concept existed before that). Morning pages are a way to dump your thoughts onto into writing so you can move on with your day without being weighed down by your worries and anxieties.

Cameron doesn't even consider morning pages "writing," because they're not intended to be shared. Instead, they're a sort of preparation for the day ahead, a way to get your mind in order. As such, they are a deeply human practice, almost like meditation. They are a way of connecting with our humanity, the flawed parts of ourselves that are full of bad ideas, worries, and imperfections. Because they are not meant to be shared and should be written as quickly as possible, they are full of grammar and spelling mistakes and unclear sentences. The whole point is that they are unedited stream of consciousness.

Public writing

Of course, writing that is meant for other people to consume should be more polished. The purpose isn't a brain dump; the purpose is communication.

To that end, sentences should be sensical. Grammar and spelling rules should be generally followed, at least to the point of ensuring that the writing is legible.

But everyone has their own idiosyncrasies, their own voice. And that doesn't need to be smoothed out and filed down into technically correct but personality-less prose.

Writing as a path to thinking

In a masterful article for The New Yorker, Ted Chiang gives one of the best and most easily understood explanations of what large language models (what we typically call AI), are actually doing and how they function. He also talks about the importance of writing. Just because LLM can write quickly and easily, in generally correct grammar, doesn't mean that LLMs should do all of our writing for us.

Chiang writes about how it is necessary for humans to write in order to discover their own ideas—even if when they begin writing, their work is unoriginal and derivative:

If students never have to write essays that we have all read before, they will never gain the skills needed to write something that we have never read.

It's important for students to learn how to articulate their ideas, not to prove that they've learned the information, but to develop their own original thoughts. Whether someone is a student or not, they likely develop their original ideas through writing. If humans aren't creating our own original writing or learning how to write, our ability to think creatively could atrophy.

"AI" as a tool, "AI" as an authority

I'm not one to say that "AI" tools should never be used during the writing process.

After all, I use Microsoft Word's spelling and grammar check daily.

I also wrote the rough draft of this blog post using Nuance's Dragon dictation software, which uses "machine learning" to better understand what the user is saying and translated into text.

Right now, I'm writing this on Nuance's mobile Dragon Everywhere software, which certainly leaves a lot to be desired in terms of accuracy. (Though it's still better than Google's free voice typing, which I also use daily.)

It certainly isn't like "machine learning" is infallible when it comes to writing. In fact, many of the typos that make their way into my writing are introduced by Dragon or Google voice typing not understanding my accent.

Perhaps an AI booster might say my accent (which that now-paywalled New York Times dialect quiz claimed is a mixture of North Texas, western Louisiana, and Oklahoma City) is the problem, not the software.

On days when I dictate a lot for work, I notice that my accent shifts slightly to become more "comprehensible" to the software, even though the desktop version of Dragon is supposed to adapt to my accent, not the other way around. There's a whole 'nother essay I could write about how the software we use tries to polish away our culture and histories (I'm Cajun and grew up in North Texas) and homogenize us into something that computers can best understand.

Writing in "AI voice"

In my day job (which involves editing other people's work), I've started to be able to spot when people have fed their fiction through "AI" editors like ProWritingAid.

It's a bit hard to articulate what the AI voice is; I'm just beginning to develop an eye for it.

But it looks like overly efficient sentences that seem to be missing something. They've been rephrased to be the most grammatically correct that they can be, on a technical level, but they often read as if they're incorrect.

It's an uncanny valley for writing, something that looks and seems human at first glance, but there is . . . something . . .  missing. It's too efficient. Not everything needs to be over-optimized. At a certain point, perfectly correct prose stops sounding human.

Limits and uses of AI editing

As someone who works in Microsoft Word for hours every day, I frequently see Word's spelling and grammar check try to make corrections that are straight-out wrong. The changes might be grammatically correct but awkward in practice. Often, however, they misunderstand a grammar rule, and if implemented, the changes would make sentences incomprehensible, turning well-crafted prose into gibberish.

I always feel unaccountably pleased when the computer makes these mistakes. They feel like a confirmation of my own humanity, somehow.

Microsoft Word's spell check and even tools like ProWritingAid and Grammarly have their place. Not every piece of writing calls for a human editor. (For example, my blog posts don't get edited by anyone other than me. I rely on the basic spell check in my markdown editor and then send my writing out to the world.)

Also, not everyone has the money or time for a human editor, and it can be incredibly helpful to have a piece of software that can help polish the rough edges of your writing (especially if you're writing in a language that you're less familiar with).

All this is to say that machine learning, LLMs, AI—or whatever you want to call it—can be useful as a tool. But we should beware of letting it shape our expression and redefine our voices. Or circumscribe our thoughts.

To me, there's a big difference between having an AI catch your spelling and grammar errors vs. having them rewrite and rephrase your sentences to "improve" your writing (or having them write a first draft which you then edit and expand upon).

The sentence about typos has a typo

I loathe grammar nitpicking, but because it's directly relevant to this conversation: Ironically, the sentence that inspired this post technically contains a typo (a comma splice).

A typo is no longer just a typo, it is a signal the writer is not using AI.

If you wanted to be grammatically correct, you might write it as "A typo is no longer just a typo; it is a signal the writer is not using AI." or "A typo is no longer just a typo. It is a signal the writer is not using AI."

But I guess because the AI didn't catch it, it isn't a real typo? That raises an interesting question. As people rely more and more on machine-based editors, will that change how we think of grammar and writing? Will some things that are technically "correct" be considered incorrect, and vice versa?

Like I mentioned, despite being someone who knows grammar rules intimately, I despise dogmatic editing and have no patience for people who are pedantic about grammar.

And, to be honest, I don't mind comma splices and similar "errors" in casual online writing (including my own).

Because I edit things for a living, I feel confident exercising editorial judgment and deciding that some typos are fine. In publishing, there is a common phrase, "stet for voice." It's an instruction to ignore a correction because doing so will preserve the voice of the author or character, and it is more important to keep that voice alive than it is to be grammatically correct.

We read to learn, to connect with others, and to go on adventures. We don't read because we relish samples of perfectly grammatical, efficient prose.

Junkspace and the internet as a dead mall

In January, I read a tweet about how the internet now resembles a dead mall.

Google search barely works, links older than 10 years probably broken, even websites that survived unusable popping up subscription/cookie approval notifications, YouTube/Facebook/Twitter/IG all on the decline, entire internet got that dying mall vibe

I haven't been able to shake that comparison. I have also been haunted by the 2001 Rem Koolhaas essay "Junkspace," which I've been reading and rereading since November. The essay, which is ostensibly about the slick, commercial spaces and malls that popped up in the late 20th century, is uncanny in its accurate description of the dead mall of the internet.

In an internet made up of five websites, each full of screenshots of the other four, how can we not feel like we're wandering through a dead mall or junkspace (which Koolhaas described as having no walls, only partitions)?

Add to that the impersonal bullshit texts that LLMs and LLM-powered editors help people churn out, and it's easy to feel like you're walking through the echoing corridors of an empty shopping mall. Occasionally, something catches your eye, and you turn your head to greet another human, only to be met by an animatronic mannequin that can talk almost like a human—but not quite.

That's how it feels to search Google and come up with a bunch of SEO content-mill, LLM-generated articles that mean nothing but rank in the algorithm because they've followed all the rules. The mall is dead and full of ghosts. And not even the fun, interesting kind of ghost.

Koolhaas calls junkspace a body double of space, which feels suspiciously like the internet (or, worse, the metaverse that tech ghouls keep trying to make happen). In junkspace, vision is limited, expectations are low, and people are less earnest. Sound familiar?

I'll certainly talk more about junkspace in future blog posts, but I can't stop thinking of parallels between the polished perfection of commercial junkspaces and the writing and editing churned out by LLMs.

If our mistakes make us human, I'm perfectly happy to make mistakes. To me, that is preferable to communicating with robotic precision and filing down all of the things that gives my writing a unique voice (even when those things make my writing "worse").

[^1] I'm not including a link to the original because I'm not trying to put anyone on blast or critique any one individual's views, necessarily. This is more about a larger trend that I'm seeing in the discourse about "AI" and humanity.

[entropieogchaos]

In regards to the AI-Art debacle, I'd like to ask how could AI-art be used? You mentioned a friend recreating the art, is there anything else? I just feel that most AI-art generators are not advanced enough to produce consistent ( in detail, color, theme, etc. ) images. They're fun but I don't see much practical use.

I think many people have the wrong expectations for AI generated images. They will point to a messed up hand or extra limb and say that's a defect that proves the AI's image is inferior. But I think all these strange glitches should be viewed more as a feature than a defect. And what makes it important to me is that the glitches are often so strange and unexpected that it's not anything I think I would have imagined on my own. And these glitches are aesthetically distinctive and different than previous styles of so-called glitch art. I think the "perfect" "beautiful" AI images are by far the most boring and least original.

I'm very attracted to chance aesthetic; the sort of images that arise from processes where I have only a limited amount of control. I have experimented with this alot in my photography, especially in long exposure images. I find many experiences with a camera to feel very similar to using an AI image generator. A camera isn't a machine that faithfully replicates reality. It is very imperfect, with numerous elements distorting the image from what it is in the real world. With experience, we can learn how to "speak" with the camera and its associated elements in a way to consistently get the sort of images we want - including images which bear little resemblance to the "real life" things it is supposedly capturing. The same thing happens with the AI generators; you're forced to learn a different language and how to speak with the thing, but you still a large element of chance in what it does, and doing things such as changing the seed or its model can have dramatically different results. At it's most creative, an AI image generator is a way for me to play with language to experiment with how this complex machine creates images from words. Sometimes I just put in strings of random numbers and get all kinds of strange things. This is analogous to holding the camera shutter open and swinging the camera around to see what happens.

There are some AI generators out there that much too "easy" to use. They are using models that are trained in such a way that they generate this very limited palette of "beautiful" images with similar composition and in a stereotypical "digital art" style. These seem to be the ones that offend artists the most because they're making beautiful images that cop their style with very little effort from the user. But if you use these enough, you notice pretty quickly that the images all essentially look the same and it gets bland and boring real quick. These generators offend me too, because it neuters and homogenizes everything that can make AI images unique and interesting. And I find the harsh censoring on some platforms to be very anti-art and anti-creative. But AI generators exist outside of that wannabe scene and it's possible to get deep into AI art without ever wasting time with those lame platforms who are just trying to make a quick buck off a fad instead of contributing anything culturally valuable.

I also think that AI image generators are an ideal ironic way to criticize AI and the Tech Industry generally. I think AI is a danger to Humanity. Not because I worry a thinking machine will take over and try to exterminate all of us, but because I worry too many people will become confident that a machine can make better decisions than us, and so offload all our thinking and difficult decisions to machines that are dangerously imperfect and prone to errors that could result in people killing people because a machine told them it was the right thing to do and people forgot how to question and think for themselves. I intentionally make grotesque and ugly AI images because I think our worship of technology is grotesque and ugly.

[guldborgemery77]

AI Blog Writer Review

AI Blog Writer is a powerful tool that enables you to write compelling content in minutes. It is designed to make writing easy, and its advanced tools can even help you write better copy. This tool has a variety of features, including multiple templates and different copy types. It can also generate the right content based on a variety of inputs, such as keywords, length, and topic. Jasper Unlike other AI blog writing software, Jasper is able to produce blogs that have high levels of content quality. This makes it a great choice for writers who want to create more content quickly and effectively. While some of its features can be confusing at first, it's possible to use them with a little practice. After you've configured the software and have entered some basic information, Jasper will generate content for you. You can then copy and paste the content you've generated to a new blog platform, like Medium, or WordPress. Depending on your preferences, you can even create a brand new post with Jasper, as it's free. Jasper offers a Starter plan, which is ideal for those just getting started with AI writing tools, or for those who want to create short-form content. The software can create headlines, short product reviews, Facebook ads, and more. If you'd like longer pieces of content, you can upgrade to the Boss mode. ai blog writer If you are looking for a blog writer that can create a good content for your blog, ContentBot is the answer. The tool comes with over 30 pre-written templates and includes many marketing tools. For example, it can generate meta content and offers a slogan generator. ContentBot is designed to create the type of content that will increase your website's traffic and help your website rank higher in search engine results. It can write articles of any length, including blog posts. It can even match your brand's voice and style, so you'll be able to create compelling content for your audience. The tool comes with a Chrome extension, so you can use it wherever you use Chrome. It also has a WordPress plugin, so you can easily generate content from within your WordPress editor. The plugin is an excellent way to boost your blogging output and increase your content creation speed. ContentBot isn't the only AI blog writer on the market. It's actually one of the first automated writers. It uses a model developed by OpenAI called the GPT-3. This model is based on an autoregressive language model that predicts the next best word. HyperWrite HyperWrite is an AI tool that can help you write better and faster. It can write entire articles for you or provide you with the freedom to write other types of content. The application uses the Problem, Agitate, Solution (PAS) Framework, which is a method of writing that uses artificial intelligence techniques to create content. HyperWrite is a powerful writing tool that uses AI tools to create natural-sounding articles. It uses generative pre-transformer 3 (GPT-3) to generate text with a natural tone and word choice. It uses large datasets and hundreds of billions of words to generate natural-sounding writing. While AI writers are still in their infancy, the technology is rapidly advancing. The number of companies investing in the field has more than doubled in the last year and is set to grow to $93.5 billion by 2021. As AI becomes more advanced, marketing tools will also incorporate machine learning tools. Meta, for instance, has announced a new research project focusing on next-generation AI. The goal is to create an AI that processes data as humans do and that is indistinguishable from a human. Rank Math's Content AI Rank Math's Content AI is an artificial intelligence-based content wizard that helps you create content that's both interesting and optimized for search engines. This tool generates a blueprint for the composition and structure of your content based on the keywords that your site wants to target. The tool is easy to use, and has plenty of useful options. To configure Content AI, go to Rank Math's General Settings and click the "Content AI" setting. Then, select whether you want to target your global audience or 80+ countries. You can also choose whether to create keyword suggestions based on the target country. This will ensure that you have content that is both useful to your audience and relevant to their needs. To use Rank Math's Content Optimization tool, first sign in to your website. Then, click on the "Activate Now" button. After that, choose the login method you wish to use. You can also create an account if you're not yet one. Once you're logged in, make sure to save your changes before refreshing the page. You'll notice suggestions that Rank Math will make for the focus keywords you've selected.

[getgenetica]

Key Strategies to Succeed in Cannabis Retail

The legal cannabis market is expanding rapidly as more states move to legalize medical and recreational use. This growth has attracted many new entrants, intensifying competition in the cannabis retail sector. To stay ahead in this dynamic industry, retailers need to adopt innovative strategies focused on customer experience, operational excellence and technology adoption.

Focus on Customer Experience

In any retail business, customer experience should be the top priority. Cannabis retailers must get to know their target audiences intimately and tailor the in-store experience accordingly. For example, a dispensary targeting medical patients may emphasize knowledgeable staff, comfortable seating and privacy; while one aiming for recreational users focuses on a relaxing vibe with entertainment options.

Attentive customer service, education and personalized recommendations also build loyalty. Retailers can deploy Cannabis AI chatbots and shopping assistants to consistently address common queries and suggest ideal products based on individual traits, ailments and preferences. Leveraging AI to understand diverse customer personas at a granular level allows for truly personalized experiences that keep them coming back.

Digital Marketing and Discovery

To stay top-of-mind in the crowded cannabis marketplace and generate new patient and customer trials, retailers must invest aggressively in targeted digital marketing. Leveraging AI and big data, dispensaries can analyze shopping behaviors to segment audiences and deliver customized promotions, deals and educational content across multiple touchpoints.

This includes utilizing AI-powered social media marketing, search engine optimization, website personalization, location-based mobile targeting and email automation. Advanced computer vision models can also dynamically customize web and social graphics to boost discovery. Allocating 15-20% of revenues towards AI-driven digital strategies is recommended.

Product Innovation

Incorporating the latest cannabis strains and product formats keeps existing customers engaged while attracting new ones. AI is revolutionizing product innovation in cannabis - large language models help develop novel strains tailored to specific wellness or recreational needs at an unprecedented pace. Meanwhile, computer vision models can conceptualize new product categories involving edibles, topicals, beverages and more.

Partnering with leading extraction labs and AI-empowered breeders opens the door to continually debuting breakthrough offerings before competitors. Continuous AI-driven R&D also ensures product assortments adapt to trends and changing consumer preferences. This could be a key long term differentiator.

Operational Excellence

Given extensive regulations, minimizing compliance costs while focusing on productivity is crucial. AI can automate many labor-intensive tasks - computer vision analyzes inventories in real-time to prevent stock-outs and optimize procurement. AI scheduling algorithms ensure efficient staffing based on historical sales patterns. Robotic process automation tackles paperwork digitally.

AI also streamlines marketing, finance and supply chain management. Regular AI audits identify process optimization opportunities while machine learning models forecast demand. This allows for lean, cost-effective operations at scale even as business grows. Retailers investing 15-20% of profits in AI-led productivity will gain significant efficiencies within 3-5 years.

Data-Driven Merchandising

Leveraging AI and Big Data is imperative for effective inventory management and merchandising. By analyzing historical point-of-sale records along with contextual factors like seasons, events or discounts, AI forecasting models accurately predict trends for individual SKUs at a granular level. This enables optimized product placements, tailored assortments by location, and minimal waste from expired shelf stock.

Embrace Technology Innovation

The cannabis industry is among the most progressive in adopting cutting-edge technology. Early investments in AI will strengthen competitive positioning in the long run. Emerging innovations like computer vision-powered store navigation aids, virtual/augmented reality showcasing SKUs and AI simulation of reformulated products should form an integral part of the technology roadmap.

Staggered implementation as technologies mature also curbs risks. But proactively scouting novel applications of retail AI, Blockchain, IoT ensures remaining at the forefront of transformation. Setting aside 3-5% of annual revenues specifically for innovation keeps the business future-proof in this dynamic landscape.

Final Thoughts

As generative AI models mature, their applications will proliferate and disrupt every segment of the cannabis retail industry. However, judicious adoption aligned with business priorities, along with responsible and ethical use of this technology will determine its true transformative potential. Early adopters leveraging generative AI strategically now will emerge as true industry pioneers.

[tagx01]

The Comprehensive Journey of Large Language Models and RLHF

Large Language Models (LLMs) have captured the world's attention since the release of groundbreaking tools like ChatGPT, and the pace of innovation shows no signs of slowing down. New models are continuously emerging, each more advanced and capable than the last. But how did we get here? What are the foundational principles behind these powerful models, and what role does RLHF (reinforced learning with human feedback) play in their development?

In this blog, we'll delve into the fascinating journey of LLMs, exploring their origins, evolution, and the pivotal processes that enhance their performance. From the basic concepts to the intricate techniques that drive today's cutting-edge AI, join us as we uncover the comprehensive journey of Large Language Models and the transformative impact of RLHF in shaping the next generation of artificial intelligence.

Large Language Models

In the ever-evolving landscape of natural language processing, a revolution has been quietly unfolding – the rise of Large Language Models (LLMs). These technological marvels have captivated researchers, developers, and enthusiasts alike, pushing the boundaries of what was once thought impossible. The journey began with a simple yet profound concept: autoregressive language modeling. By training models to predict the next word or token in a sequence of text, researchers unlocked the door to a deeper understanding of language patterns, laying the foundation for a transformative leap forward.

At the heart of this transformation lies the transformer architecture, a neural network architecture that has become the backbone of LLMs. With its ability to capture the intricate relationships between words and their contextual meanings, the transformer architecture has empowered LLMs to process and generate human-like text with an unprecedented level of fluency and coherence. From the pioneering GPT (Generative Pre-trained Transformer) to the awe-inspiring GPT-3, each iteration has pushed the boundaries further, demonstrating an uncanny ability to understand and generate text across a wide range of domains.

RLHF (reinforced learning with human feedback)

Reinforcement Learning with Human Feedback (RLHF) is a game-changer in the world of artificial intelligence, especially for Large Language Models (LLMs). At its core, RLHF combines the power of machine learning with the insights and guidance of human evaluators. Imagine teaching a computer how to learn by rewarding it for good behavior and correcting it when it makes mistakes. That's essentially what RLHF does. By integrating human feedback into the training process, LLMs become more accurate, reliable, and aligned with human expectations.

This innovative approach not only improves the performance of LLMs but also ensures they generate more relevant and useful responses. Humans provide feedback on the model's outputs, highlighting what works well and what doesn't. The model then uses this feedback to refine its future responses, leading to continuous improvement. In essence, RLHF helps bridge the gap between human intuition and machine efficiency, creating AI systems that are smarter, more responsive, and better suited to real-world applications.

Applications of Large Language Models and RLHF

Unlocking Boundless Creativity: Large Language Models empowered by RLHF have opened the doors to unprecedented realms of AI-generated art and creative expression. From crafting captivating poetry and prose to composing melodic lyrics, these models have breathed life into the fusion of technology and human-like creativity, blurring the lines between artificial and organic artistry.

Conversational Intelligences Revolutionized: Imagine AI assistants that not only understand the nuances of human language but also engage in natural, contextual, and meaningful dialogues. RLHF-based chatbots and virtual assistants have redefined the way we interact with technology, providing invaluable assistance while fostering a sense of seamless communication and rapport.

Breaking Language Barriers: In a globalized world, the ability to bridge linguistic divides is paramount. Large Language Models, fine-tuned with RLHF, have emerged as powerful language translation tools, transcending geographical boundaries and fostering cross-cultural understanding through accurate and nuanced translations.

Information Distilled: In the age of information overload, the ability to condense complex texts into concise and informative summaries is a game-changer. RLHF-powered language models have become adept at text summarization, extracting the essence of lengthy documents and presenting it in a digestible format, saving time and effort for professionals and researchers alike.

New Frontiers in Human-AI Collaboration: As Large Language Models infused with RLHF continue to evolve, they are poised to become indispensable partners in various industries. From scientific research and data analysis to content creation and beyond, these models are opening up new frontiers of human-AI collaboration, augmenting human capabilities and accelerating progress in ways once thought unimaginable.

How RLHF is Used to Train ChatGPT

Refine, Label, Holdout, Fine-tune (RLHF) is a powerful process that plays a crucial role in training models like ChatGPT. It starts with refining raw data to ensure it's clean and relevant. This refined data is then meticulously labeled by experts, providing clear examples of desired outputs. Labeling helps the model understand the patterns and nuances of human language, making it more effective at generating accurate and contextually appropriate responses.

The next step involves holding out a portion of the data to test the model's performance. This "holdout" data acts as a benchmark to evaluate how well the model is learning. Finally, the model is fine-tuned using advanced algorithms, adjusting its parameters to improve accuracy and reliability. This continuous cycle of refinement, labeling, testing, and fine-tuning ensures that ChatGPT remains one of the most sophisticated and capable language models available, offering users high-quality, context-aware interactions.

Conclusion

Large Language Models (LLMs) have significantly advanced the field of natural language processing, offering unprecedented capabilities in understanding and generating human-like text. Their success lies in their extensive training on vast datasets, allowing them to capture intricate language patterns and contextual nuances. However, as powerful as they are, LLMs are not without limitations. Ethical concerns, such as biases and misinformation, as well as technical challenges like fine-tuning for specific tasks, necessitate continuous refinement. This is where Reinforcement Learning from Human Feedback (RLHF) comes into play, providing a robust mechanism to enhance the performance and reliability of LLMs by incorporating human judgment into the training process.

By integrating RLHF, LLMs can be fine-tuned to better align with human values, improve decision-making accuracy, and reduce the propagation of harmful content. This iterative process of refining, labeling, holding out, and fine-tuning ensures that the models evolve to meet higher standards of ethical and functional performance.

At TagX, we have the expertise to effectively implement RLHF, powering the next generation of language models. As we move forward, the collaboration between advanced AI models and human expertise will be crucial in driving innovation while maintaining responsible AI deployment. Embracing these advancements will not only push the boundaries of what LLMs can achieve but also ensure their application in ways that are beneficial, fair, and aligned with societal values.

Visit Us, www.tagxdata.com

Original Source, https://www.tagxdata.com/the-comprehensive-journey-of-large-language-models-and-rlhf

[jcmarchi]

Can AI Interpret Dreams?

New Post has been published on https://thedigitalinsider.com/can-ai-interpret-dreams/

Can AI Interpret Dreams?

While researchers have taken the first steps toward artificial intelligence dream interpretation, the technology is still largely unproven. It might take years for high-end applications to reach the consumer market. Is there a way to use AI to interpret dreams today?

Why Would You Need AI to Interpret Dreams?

There are a few prevailing theories on why dreams happen. Some argue it’s random neuronal activity, others say it’s to process the day’s events and a few claim it’s your unconscious needs and desires surfacing. Realistically, it’s probably a combination of multiple ideas. However, none can help explain the specific meaning behind each of your nighttime visions. 

Dreams are complex, incoherent and baffling for reasons unknown. You could find yourself in your grandmother’s living room speaking to Elvis Presley about dog astronauts, and everything would seem normal — understandably, you’d want to make sense of things with AI.

Even if you can comprehend your dream at face value, it’s generally accepted that a more profound meaning exists. Symbols, themes and events span cultures and generations, lending to their significance. 

For example, dreaming about losing your teeth could mean you’re dealing with stress, uncertainty or insecurities in your waking life. Alternatively, a nightmare about falling could mean you don’t feel in control of your life or supported by your loved ones. Seemingly random, nonsensical events might be significant — this is why AI interpretation is a big deal. 

Can You Use AI for Dream Interpretation?

Technically, you could use AI to interpret your dreams today if you get a generative model and word your prompt right. However, accuracy is an issue — if you can’t decipher your dream’s meaning, how is an algorithm supposed to? While it may guess or output nonsense to appease you, would you be satisfied with its generic responses?

Even if you don’t feel connected to your dreams, they’re incredibly personal experiences. Each is a jumbled collection of your memories, emotions, relationships and subconscious thoughts. While you can technically use a large language model (LLM) to decipher them, its output would only be partially accurate at best.

That said, relatively accurate AI interpretations aren’t impossible. Some researchers have already uncovered the technology needed to make it work — multiple studies conducted in 2023 prove it is feasible. At this point, testing, prototyping and commercializing these discoveries is just a matter of time, resources and funding. 

The Technology Behind AI Dream Interpretation

Training data is fundamental to any AI-powered dream interpretation technology. What information can you feed an algorithm to return consistent, accurate output? Theoretically, you could use text-based descriptions, statistics on commonly dreamed themes or artists’ renditions. However, sourcing enough would be an issue. 

Some researchers overcame this obstacle by providing machine learning (ML) models with dozens of hours of brain activity scans. This approach is interesting for a few reasons. For one, it relies on evidence-based information instead of the dreamer’s commentary — which, coincidentally, increases data availability drastically.

It also identifies the underlying drivers of rapid eye movement (REM) sleep, targeting the language or image-processing areas of the brain rather than attempting to make sense of the dream itself. As a result, AI isn’t as affected by the dreamer’s bias — meaning its chance of outputting a relatively objective, accurate interpretation is higher. 

Aside from training data, you need a generative model to reconstruct, interpret or translate information. This technology’s popularity is rapidly increasing — its market size will have a compound annual growth rate of 36.5% from 2024 to 2030 — so sourcing an out-of-the-box solution would be easy. However, building one from the ground up would be wise.

Most AI-powered dream interpretation solutions need natural language processing (NLP) and image recognition technology to some extent. After all, most REM sleep is a combination of images and words. Beyond that, you could use anything from deep learning models to neural networks to make your tool work. 

Ways You Can Use AI to Interpret Dreams 

While generative models can produce text, images, audio and music, only a few proven methods of AI-driven dream interpretation currently exist. 

1. Text-to-Text Generation 

The simplest method is text-to-text generation, where an LLM, NLP or ML model analyzes your typed prompts. You enter what you remember about your dream or follow a decision-tree format to get answers. On the one hand, it’s fast and straightforward. On the other, it’s inaccurate — you forget most of the REM stage upon waking, so the AI works off a fragmented narrative. 

2. EEG-to-Text Generation

An LLM and an electroencephalogram (EEG) recording the brain’s electrical signals can turn thoughts into words. You must read while wearing a soft cap filled with sensors for this to work. The model converts that activity into text.

Your brain sends a specific signal when you think of a word or phrase. An algorithm can find patterns in this activity, making translation possible. You could use this EEG-to-text generation model to develop a transcript of your REM sleep. 

Peer-reviewed research proved this model can achieve 60% accuracy, which is impressive for a proof of concept. The soft cap is portable and relatively cheap to produce, making it one of the few inventions that might see mass-market applications.

3. fMRI-to-Image Generation

A research group discovered a deep learning model that can analyze functional magnetic resonance imaging (fMRI) scans — images of the brain’s blood flow — to accurately recreate images people see. It trained on 10,000 photos to interpret what people were viewing. 

As the study’s participants stared at an image, their temporal lobe registered its content, and their occipital lobe cataloged its scale and layout. The AI tracked this activity to reconstruct what they were seeing. While its recreations started as noise, they slowly became recognizable.

4. fMRI-to-Text Generation

Researchers used fMRI scans and an LLM in an encoding and decoding system to reconstruct brain activity in a text-based format. The leading neuroscientist on the project said the team was shocked it worked as well as it did. 

As people read text or watched silent videos, the AI described the content — and usually got the gist. For instance, one person read, “I didn’t know whether to scream, cry or run away. Instead, I said leave me alone, I don’t need your help.” The model outputted, “Started to scream and cry and then she just said I told you to leave me alone, you can’t hurt me anymore.”

Interestingly, when the researchers tailored the tool for one of the study’s participants, it could only reconstruct unintelligible gibberish when used on another. There might be potential for personalized algorithm-based dream interpreters. 

Why You Should Be Wary of an AI Interpreter 

While using algorithms for dream interpretation sounds promising, there are a few drawbacks to be aware of. The most significant is hallucination. According to one survey, 89% of machine learning engineers working with generative AI say their models make things up — and 93% see it happen daily or weekly.

Until AI engineers iron out the hallucination issue, this technology’s application in REM sleep is a gray area. While using it for fun is harmless, some people — those who would typically go to therapists or psychologists for dream interpretations — might get an output that damages their mental health or sets back their treatment progress.

It might subconsciously influence you even if you’re skeptical or indifferent to an algorithm’s output. For example, you might grow distant from your partner after the model tells you your cheating dream signifies a failing relationship. 

Being at the other end of the spectrum can be just as damaging. Fully believing in the AI’s output — despite potential bias or hallucinations — could negatively affect you. This overconfidence might make you misinterpret your emotions, interactions with others or past trauma, leading to unwanted situations in your waking life. 

There’s also the issue of the sticker price. Text-to-text generation is the most accessible and affordable but is inaccurate. If you want something better, prepare to pay up. Considering that a single MRI scan can cost up to $4,000 — and one machine can be a multimillion-dollar investment — accurate AI dream interpreters are probably years away.

What Does the Future Hold for This Technology?

Having a personal AI dream interpreter could be exciting and helpful. Even if this technology doesn’t enter the consumer market soon, it will likely find a place in therapy, psychology and medical practices. One day, you might use it to work through past trauma, identify sleep issues or uncover hidden emotions.