#ai generator tester

[rahulacademy]

Venkatesh (Rahul Shetty) Academy-Lifetime Access Get lifetime access to Venkatesh (Rahul Shetty) Academy and learn software testing, AI testing, automation tools, and more. Learn at your own pace with real projects and expert guidance. Perfect for beginners and professionals who want long-term learning support.

[genuflectx]

Doing testing on using the spectrograms for manually spotting AI generated music, and so far more people get over 50% than they do below, but the accuracy still isn’t high enough for my liking…

However, results are also likely highly dependent on how well each individual can read the contrast of the spectrogram and to what degree they believe the contrast matches either AI or human. So, I believe with practice or training, an individual can increase their accuracy in using the chart to spot spectrograms that are more likely AI than human 🤔 just like how you get better at picking out AI images with practice, even ones that are more convincing

However, over sampling 30+ spectrograms of both AI and human, there is the occasional one that looks similar to the other, but if found this to exceedingly rare. The one that was the closest to human spectrograms was an AI jazz song, which even to me looked indistinguishable. But I have yet to see a human one that is as washed out as AI ones, and they pretty much never have a flat yellow line at the bottom.

It is unlikely this test would ever be the only one needed in determining an AI song to human, but I believe it could end up being a good tool to use among other tools, namely human intuition, listening for common synthetic AI voices/compositions, generic lyrics, lack of credits, weird behavior from the uploader, etc etc.

When I have more to share I will 🤔

[kamalkafir-blog]

Senior Security Engineer, Retail Engineering

Job title: Senior Security Engineer, Retail Engineering Company: Apple Job description: be required. Minimum Qualifications Experience in an existing security engineer, security consultant, security architect, penetration… tester or similar role. Expertise in threat modelling, secure architecture design, and reviewing complex systems… Expected salary: Location: London Job date: Thu, 08 May 2025…

[getmoneymethods]

Seeking Beta Testers for New AI Content Generation Platform - WriteBotIQ

I’m excited to share that we're looking for beta testers for our new AI-powered content generation platform, WriteBotIQ. As a beta tester, you’ll get free access to generate up to 50,000 words and 25 custom AI-generated images.

WriteBotIQ offers a variety of powerful features:

AI Writing Assistant: Effortlessly create high-quality text for articles, blogs, reports, and more.

AI Image Creator: Design stunning visuals and graphics in seconds, perfect for social media posts and marketing materials.

AI Code Generator: Produce clean, efficient code tailored to your project’s requirements.

AI Chatbot Builder: Develop intelligent chatbots that enhance customer engagement and interaction.

AI Speech to Text Converter: Accurately transcribe speech into text, ideal for meetings, lectures, and personal notes.

AI Article Generator: Create well-structured and compelling articles based on your specific topics and keywords.

AI Photo Editor: Use advanced editing tools to enhance and perfect your images for both personal and professional use.

AI Document Collaborator: Collaborate in real-time within your documents to improve productivity and teamwork.

AI Image Recognition Tool: Instantly identify and analyze images to unlock new creative possibilities.

AI Website Chat: Integrate AI-driven chat functionality on your website for instant support and engagement.

AI Content Rewriter: Refine and enhance the clarity, detail, and quality of your text with AI-powered rewrites.

AI Chat Image Creator: Generate custom images based on chat interactions to enhance user engagement and visual storytelling.

AI Video to Blog Converter: Transform YouTube videos into well-written blog posts, expanding your content reach.

AI Coding Assistant: Receive real-time code suggestions and improvements to boost your productivity and code quality.

AI Content from RSS Feeds: Automate content creation from your favorite RSS feeds to keep your site fresh and up-to-date.

If you’re interested in helping us shape the future of AI content generation and get some free resources in the process, please sign up at writebotiq.com.

Thanks for considering this opportunity! Your feedback will be invaluable in helping us improve and refine WriteBotIQ.

Looking forward to your participation!

[zerogpt3]

An AI Detector is a tool that uses a vast datasets of information to determine whether a piece of text is genuinely human-writtten or if it's AI-generated.

[mostlysignssomeportents]

Even if you think AI search could be good, it won’t be good

TONIGHT (May 15), I'm in NORTH HOLLYWOOD for a screening of STEPHANIE KELTON'S FINDING THE MONEY; FRIDAY (May 17), I'm at the INTERNET ARCHIVE in SAN FRANCISCO to keynote the 10th anniversary of the AUTHORS ALLIANCE.

The big news in search this week is that Google is continuing its transition to "AI search" – instead of typing in search terms and getting links to websites, you'll ask Google a question and an AI will compose an answer based on things it finds on the web:

https://blog.google/products/search/generative-ai-google-search-may-2024/

Google bills this as "let Google do the googling for you." Rather than searching the web yourself, you'll delegate this task to Google. Hidden in this pitch is a tacit admission that Google is no longer a convenient or reliable way to retrieve information, drowning as it is in AI-generated spam, poorly labeled ads, and SEO garbage:

https://pluralistic.net/2024/05/03/keyword-swarming/#site-reputation-abuse

Googling used to be easy: type in a query, get back a screen of highly relevant results. Today, clicking the top links will take you to sites that paid for placement at the top of the screen (rather than the sites that best match your query). Clicking further down will get you scams, AI slop, or bulk-produced SEO nonsense.

AI-powered search promises to fix this, not by making Google search results better, but by having a bot sort through the search results and discard the nonsense that Google will continue to serve up, and summarize the high quality results.

Now, there are plenty of obvious objections to this plan. For starters, why wouldn't Google just make its search results better? Rather than building a LLM for the sole purpose of sorting through the garbage Google is either paid or tricked into serving up, why not just stop serving up garbage? We know that's possible, because other search engines serve really good results by paying for access to Google's back-end and then filtering the results:

https://pluralistic.net/2024/04/04/teach-me-how-to-shruggie/#kagi

Another obvious objection: why would anyone write the web if the only purpose for doing so is to feed a bot that will summarize what you've written without sending anyone to your webpage? Whether you're a commercial publisher hoping to make money from advertising or subscriptions, or – like me – an open access publisher hoping to change people's minds, why would you invite Google to summarize your work without ever showing it to internet users? Nevermind how unfair that is, think about how implausible it is: if this is the way Google will work in the future, why wouldn't every publisher just block Google's crawler?

A third obvious objection: AI is bad. Not morally bad (though maybe morally bad, too!), but technically bad. It "hallucinates" nonsense answers, including dangerous nonsense. It's a supremely confident liar that can get you killed:

https://www.theguardian.com/technology/2023/sep/01/mushroom-pickers-urged-to-avoid-foraging-books-on-amazon-that-appear-to-be-written-by-ai

The promises of AI are grossly oversold, including the promises Google makes, like its claim that its AI had discovered millions of useful new materials. In reality, the number of useful new materials Deepmind had discovered was zero:

https://pluralistic.net/2024/04/23/maximal-plausibility/#reverse-centaurs

This is true of all of AI's most impressive demos. Often, "AI" turns out to be low-waged human workers in a distant call-center pretending to be robots:

https://pluralistic.net/2024/01/31/neural-interface-beta-tester/#tailfins

Sometimes, the AI robot dancing on stage turns out to literally be just a person in a robot suit pretending to be a robot:

https://pluralistic.net/2024/01/29/pay-no-attention/#to-the-little-man-behind-the-curtain

The AI video demos that represent "an existential threat to Hollywood filmmaking" turn out to be so cumbersome as to be practically useless (and vastly inferior to existing production techniques):

https://www.wheresyoured.at/expectations-versus-reality/

But let's take Google at its word. Let's stipulate that:

a) It can't fix search, only add a slop-filtering AI layer on top of it; and

b) The rest of the world will continue to let Google index its pages even if they derive no benefit from doing so; and

c) Google will shortly fix its AI, and all the lies about AI capabilities will be revealed to be premature truths that are finally realized.

AI search is still a bad idea. Because beyond all the obvious reasons that AI search is a terrible idea, there's a subtle – and incurable – defect in this plan: AI search – even excellent AI search – makes it far too easy for Google to cheat us, and Google can't stop cheating us.

Remember: enshittification isn't the result of worse people running tech companies today than in the years when tech services were good and useful. Rather, enshittification is rooted in the collapse of constraints that used to prevent those same people from making their services worse in service to increasing their profit margins:

https://pluralistic.net/2024/03/26/glitchbread/#electronic-shelf-tags

These companies always had the capacity to siphon value away from business customers (like publishers) and end-users (like searchers). That comes with the territory: digital businesses can alter their "business logic" from instant to instant, and for each user, allowing them to change payouts, prices and ranking. I call this "twiddling": turning the knobs on the system's back-end to make sure the house always wins:

https://pluralistic.net/2023/02/19/twiddler/

What changed wasn't the character of the leaders of these businesses, nor their capacity to cheat us. What changed was the consequences for cheating. When the tech companies merged to monopoly, they ceased to fear losing your business to a competitor.

Google's 90% search market share was attained by bribing everyone who operates a service or platform where you might encounter a search box to connect that box to Google. Spending tens of billions of dollars every year to make sure no one ever encounters a non-Google search is a cheaper way to retain your business than making sure Google is the very best search engine:

https://pluralistic.net/2024/02/21/im-feeling-unlucky/#not-up-to-the-task

Competition was once a threat to Google; for years, its mantra was "competition is a click away." Today, competition is all but nonexistent.

Then the surveillance business consolidated into a small number of firms. Two companies dominate the commercial surveillance industry: Google and Meta, and they collude to rig the market:

https://en.wikipedia.org/wiki/Jedi_Blue

That consolidation inevitably leads to regulatory capture: shorn of competitive pressure, the companies that dominate the sector can converge on a single message to policymakers and use their monopoly profits to turn that message into policy:

https://pluralistic.net/2022/06/05/regulatory-capture/

This is why Google doesn't have to worry about privacy laws. They've successfully prevented the passage of a US federal consumer privacy law. The last time the US passed a federal consumer privacy law was in 1988. It's a law that bans video store clerks from telling the newspapers which VHS cassettes you rented:

https://en.wikipedia.org/wiki/Video_Privacy_Protection_Act

In Europe, Google's vast profits lets it fly an Irish flag of convenience, thus taking advantage of Ireland's tolerance for tax evasion and violations of European privacy law:

https://pluralistic.net/2023/05/15/finnegans-snooze/#dirty-old-town

Google doesn't fear competition, it doesn't fear regulation, and it also doesn't fear rival technologies. Google and its fellow Big Tech cartel members have expanded IP law to allow it to prevent third parties from reverse-engineer, hacking, or scraping its services. Google doesn't have to worry about ad-blocking, tracker blocking, or scrapers that filter out Google's lucrative, low-quality results:

https://locusmag.com/2020/09/cory-doctorow-ip/

Google doesn't fear competition, it doesn't fear regulation, it doesn't fear rival technology and it doesn't fear its workers. Google's workforce once enjoyed enormous sway over the company's direction, thanks to their scarcity and market power. But Google has outgrown its dependence on its workers, and lays them off in vast numbers, even as it increases its profits and pisses away tens of billions on stock buybacks:

https://pluralistic.net/2023/11/25/moral-injury/#enshittification

Google is fearless. It doesn't fear losing your business, or being punished by regulators, or being mired in guerrilla warfare with rival engineers. It certainly doesn't fear its workers.

Making search worse is good for Google. Reducing search quality increases the number of queries, and thus ads, that each user must make to find their answers:

https://pluralistic.net/2024/04/24/naming-names/#prabhakar-raghavan

If Google can make things worse for searchers without losing their business, it can make more money for itself. Without the discipline of markets, regulators, tech or workers, it has no impediment to transferring value from searchers and publishers to itself.

Which brings me back to AI search. When Google substitutes its own summaries for links to pages, it creates innumerable opportunities to charge publishers for preferential placement in those summaries.

This is true of any algorithmic feed: while such feeds are important – even vital – for making sense of huge amounts of information, they can also be used to play a high-speed shell-game that makes suckers out of the rest of us:

https://pluralistic.net/2024/05/11/for-you/#the-algorithm-tm

When you trust someone to summarize the truth for you, you become terribly vulnerable to their self-serving lies. In an ideal world, these intermediaries would be "fiduciaries," with a solemn (and legally binding) duty to put your interests ahead of their own:

https://pluralistic.net/2024/05/07/treacherous-computing/#rewilding-the-internet

But Google is clear that its first duty is to its shareholders: not to publishers, not to searchers, not to "partners" or employees.

AI search makes cheating so easy, and Google cheats so much. Indeed, the defects in AI give Google a readymade excuse for any apparent self-dealing: "we didn't tell you a lie because someone paid us to (for example, to recommend a product, or a hotel room, or a political point of view). Sure, they did pay us, but that was just an AI 'hallucination.'"

The existence of well-known AI hallucinations creates a zone of plausible deniability for even more enshittification of Google search. As Madeleine Clare Elish writes, AI serves as a "moral crumple zone":

https://estsjournal.org/index.php/ests/article/view/260

That's why, even if you're willing to believe that Google could make a great AI-based search, we can nevertheless be certain that they won't.

If you'd like an essay-formatted version of this post to read or share, here's a link to it on pluralistic.net, my surveillance-free, ad-free, tracker-free blog:

https://pluralistic.net/2024/05/15/they-trust-me-dumb-fucks/#ai-search

Image: Cryteria (modified) https://commons.wikimedia.org/wiki/File:HAL9000.svg

CC BY 3.0 https://creativecommons.org/licenses/by/3.0/deed.en

--

djhughman https://commons.wikimedia.org/wiki/File:Modular_synthesizer_-_%22Control_Voltage%22_electronic_music_shop_in_Portland_OR_-_School_Photos_PCC_%282015-05-23_12.43.01_by_djhughman%29.jpg

CC BY 2.0 https://creativecommons.org/licenses/by/2.0/deed.en

[missmisnomer]

Just encountered something that made me so viscerally mad I had to stop everything I was doing and scream about it into the void (you are the void, tumblr, thank you for your time)

Ever seen that little test tube icon at the top of your Google search? That's Search Labs.

This led me to reading about all the "experiments" they have going on and OH BOY, this is the one that got me mad enough to actually leave feedback (bear in mind I like NEVER leave comments or reviews on things so you know it's bad)

Of all the "experiments" they are launching with their Search Labs, this one is the most egregiously transparent attempt to capitalize on the human brain's superiority over a computer's.

I see what they are doing.

"Look at these auto generated images and try your best to tell us what you think the original prompt is! Teehee, isn't this a fun test? Don't you want to have fun?? Play our game, uwu".

This is about as blatant as you can get: they want to train their Ai to get better at generating its frankly often garbage results by using real people to tell it how it could have done better.

My guess as to why Google is even doing this: ◽️Let’s say you use of one of Google's programs to generate an image based on your own personal written prompt (I know YOU wouldn't, fair tumblr user, but stay with me here)

◽️You are subsequently frustrated at the slop Google generated for you, and select a button that says something along the lines of "I am not satisfied with this result".

◽️This auto-triggers something on Google's end, which I assume captures this response along with the image it is attached to.

◽️Now, they can put that picture in front of thousands of unwitting people who can tell Google EXACTLY what "prompt" they associate with that image instead. Exactly the words that it SHOULD be tagged with.

These pieces of ai "art" are NOT good, let me make that clear. What financial benefit would Google have to present a panel of testers with perfectly generated images? To make this game in the first place? The only way ai can advance is when humans tell it what it did wrong. Because the computer doesn't fucking know what a raven vs a writing-desk is. It needs us to give it the words to think. Poor baby gets confused when we are vague. :(

All this under the guise of a cutesy little "test". A "game".

This is not fun, and it is extremely scummy. Do better Google. Be better. I'm attaching some screenshots of the first "level" so you too can enjoy the art of prompting!!! (she says with so much dry sarcasm the Deserts of Arrakis spontaneously turn into an ocean)

And to have the audacity to show the actual real pieces of art made by real artists that they trained this stupid machine with. Fuck entirely off.

[beclight]

it's okay chimera's bec design sucks ass and looks randomly generated. as usual.

im starting to think chimera is secretly a tester for neuralink and all his palettes are ai generated and seeking inspiration from kindergartners' paintings

[erosmutt]

IMPORTANT, PLEASE READ!

REGARDING THE FUTURE OF MY JANITOR AI ACCOUNT AND BOTS:

firstly if you don't use jai, feel free to ignore this post. if you do, please read the entirety of it thoroughly.

janitor ai is notorious for it's lack of censorship. also, for its horrible mod/dev team and horrible communication with its user base. they never tell us when anything is happening to the site, or if they do, it's after the fact. any site tests they do aren't on a separate code tester, it's on the up and running site itself. no matter how much we—the users—complain, they just say "oopsie! we'll do better next time guys!" with their fingers crossed behind their backs.

so recently shep—the head dev of janitor—made an announcement on discord instead of using the jai blog like a normal person would where everyone can see it. the announcement said in so many words that janitor would start implementing a filter to stop illegal things from happening in people's private chats, such as CSAM, beastiality, and incest. here's what was said:

Here's an important comment thread as well:

now while anthros and animals are absolutely not my thing, the problem with this isn't the purging of the mlp bots in general but that it was done without any warning.

how does this affect me? well, i have bots that are at the very least morally questionable such as numerous stepcest bots, a serial killer bot, and even a snuff bot which was a request. but i also have many, many normal bots that significantly outnumber the others. i have a dad!han bot which is currently a target to be deleted and a catalyst for a potential ban of my account.

related to this, here's part of a reblog of a post from @fungalpieceofextras (apologies for the random tag but this is a very good point):

"We need to remember that morality and advertiser standards are not the same thing, any more than morality and legality are the same thing. We also need to remember, Especially Right Now, that fictional scenarios are Not Real, and that curating your experiences yourself is essential to making things tolerable. Tags exist for a reason, and that reason is to warn and/or state the content of something."

janitor has a dead dove tag. there are also custom tags where you can warn what the content of your bot is, and you also have the ability to do so in the bot description. you can block tags on janitor ai. many people don't like stepcest or infidelity bots and you know what they do? block the tags!

anyways, my overarching point for making this post is that i have dead dove bots that are illegal (eg dad!han and serial killer!sam). but it's all fictional. that seems to not matter. so, what does this mean for my janitor account and the future of making bots?

well, for right now, i don't want to get wrongfully banned. i also don't want any of my bots deleted. i spend lots and lots of time perfecting bot personalities and making the perfect intro message for you all, and if that hard work culminates into deletion against my will, i am going to be pissed. so, i'm going to save the code and messages for my "illegal" bots for later, and i'm going to consider archiving said bots.

i'm so so sorry that this is happening and that someone's favorite bot will maybe be taken from them. if ever it does get taken down please, by all means, shoot me a message and i'll slide you the doc with the code. but i also will be wary of this because my bots have been stolen. so long as you don't transfer my bots to other sites for people to chat with, meaning that you only use it for yourself as a private bot, then i don't mind at all.

thank you all for reading and if there's an update i'll let you all know! <3

[rahulacademy]

AI Automated Testing Course with Venkatesh (Rahul Shetty) Join our AI Automated Testing Course with Venkatesh (Rahul Shetty) and learn how to test software using smart AI tools. This easy-to-follow course helps you save time, find bugs faster, and grow your skills for future tech jobs. To know more about us visit https://rahulshettyacademy.com/

[genuflectx]

Someone ought to tell all the rad Murder Drone artists whose art is used on the Church of Null thumbnails/slideshows that their art is, without permission, being associated with AI generated songs (and that the person is just lying about it being AI)

All the art is yoinked from Twitter which I don’t have anymore 😭 but the credits are on the video descriptions 🙏

Edit: Adding the evidence below the cut; the songs and Electrical Ink fail 3 different tests, which point towards high likelihood of AI usage

#1 - Using SubmitHub's AI song checker, first testing accuracy of checker. Claims 90% accuracy, did own test with 10 human and 10 AI. About the same accuracy on both at 60%, with remaining percentage usually "inconclusive" with a smaller percentage of inaccuracy (10% in the human test, 20% in AI test but only for Udio). Caveat: Udio throws it off, every Udio song tested was inconclusive or incorrect, however I noted a mix of human and AI tells still reported in the Udio breakdowns. Therefore, this checker is reasonably accurate at marking AI as AI and is more likely to say something is inconclusive than to mark it incorrectly, but should not be the only tool used to assess.

Sampling every other Church of Null song (testing with full MP3s), SubmitHub's checker identified 10 of 12 songs as strongly AI and the remaining 2 as inconclusive. This is a higher rate of being marked as AI than either the human or AI tests I did beforehand. If it were human there would be some marked human, if it were Udio AI it would more likely show a mix of human and inconclusive. All AI and two inconclusive is more likely Suno AI.

#2 - Using my manual spectrogram contrast test, first testing on 23 respondents. Results are in the link, with 65% of respondents getting a score of over half correctly assessed. This test becomes more accurate when used by an individual practiced in it (comparing an unlabeled spectrogram to a chart of AI and human spectrograms, then sorting it onto either side accordingly by which pattern it most closely resembles). My own score was 12/14. This test will never be 100% accurate as not every spectrogram follows the pattern; i.e. rarely, an AI song has the spectrogram appearance as a human made one.

Respondents were mixed on whether a sampling of 6 songs from Church of Null were AI or human at the end of the test, with both high scoring and low scoring assessing them about the same, with only a slightly higher rate of being assessed as AI in high scoring testers. Only one lower scoring tester assessed AI under 50%, assessing two of six as AI (33%). However, nobody assessed the set as being all human. Note, psychology may have made this set difficult, as respondents may have believed it was unlikely that a set would be all AI or all human, which would influence answers.

When I originally assessed the six spectrograms, I assessed all of them as appearing closer to AI generated spectrograms than human, using the reference charts. I am very practiced at spectrogram contrast assessment since I am the one who made these charts and tests, sampling 30+ AI and 30+ human.

#3 - Using a smell test, or suspicious tells that just make you feel like something is off with the vibes. This can be lack of credits/suspicious credits, an AI "shimmer" effect on all the audio, generic lyrics that sound AI generated/edited, a music production output that is unrealistic for hand-made music, etc.

In Church of Null's case, it's creator Electrical Ink: shows no musical production on the channel before CoN, claims to have 6 anon helpers/vocalists but only credits a weird blank "creative consulting" channel, has produced 25 beautiful songs with complex compositions and vocals in 4 months while claiming to record these in person (and simultaneously writing 62k+ words of the fanfic, or about 15k a month), includes the robotic "shimmer" present in Suno AI while claiming it's "autotune," deletes comments asking if it's Suno (happened to me), and uses art before asking permission for the thumbnails and lyric videos.

The one other credit I found under a reply to a random comment is E-LIVE-YT (a "collaborator" on one song, however E-LIVE may have exaggerated this as they couldn’t even remember Electrical Ink’s name during a livestream), a real person who uploaded at least 1 AI generated song (admitted) but claims the rest are human made. Though, they also produce music at an unrealistic rate (43 tracks in 5 months) but mostly "extensions" of existing music, something AI song generators let you do (he uses Bandlab, which has AI tools exactly for that). The ones with lyrics have Suno's "shimmer" and the lyrical breakdown that E-LIVE posts on comments read as AI summaries/analysis (right down to calling N "they," not knowing his pronoun; a shortcut to chatGPT was on their desktop during a livestream, they removed this for later streams). Additionally, E-LIVE also has strange credits to blank channels and 1 or 2 tiny channels that just upload poor quality Roblox clips.

Ironically, even this "fan" "collaborator" believes Church of Null is AI and complains about competing with it, and regular Suno AI users in the Suno Discord believe so too.

Sniff sniff.... somethin smells funny...

[kamalkafir-blog]

Ethical Hacker - Senior Consultant

Job title: Ethical Hacker – Senior Consultant Company: Forvis Mazars Job description: development. Job Purpose As a Senior Penetration Tester, you’ll lead complex security engagements across infrastructure, cloud… and Cloud penetration testing experience above and beyond running automated tools. A good understanding of Unix, Microsoft… Expected salary: Location: United Kingdom Job date: Thu, 15…

[samueldays]

Contra Yishan: Google's Gemini issue is about racial obsession, not a Yudkowsky AI problem.

@yishan wrote a thoughtful thread:

Google’s Gemini issue is not really about woke/DEI, and everyone who is obsessing over it has failed to notice the much, MUCH bigger problem that it represents. [...] If you have a woke/anti-woke axe to grind, kindly set it aside now for a few minutes so that you can hear the rest of what I’m about to say, because it’s going to hit you from out of left field. [...] The important thing is how one of the largest and most capable AI organizations in the world tried to instruct its LLM to do something, and got a totally bonkers result they couldn’t anticipate. What this means is that @ESYudkowsky has a very very strong point. It represents a very strong existence proof for the “instrumental convergence” argument and the “paperclip maximizer” argument in practice.

See full thread at link.

Gemini's code is private and Google's PR flacks tell lies in public, so it's hard to prove anything. Still I think Yishan is wrong and the Gemini issue is about the boring old thing, not the new interesting thing, regardless of how tiresome and cliched it is, and I will try to explain why.

I think Google deliberately set out to blackwash their image generator, and did anticipate the image-generation result, but didn't anticipate the degree of hostile reaction from people who objected to the blackwashing.

Steven Moffat was a summary example of a blackwashing mindset when he remarked:

"We've kind of got to tell a lie. We'll go back into history and there will be black people where, historically, there wouldn't have been, and we won't dwell on that. "We'll say, 'To hell with it, this is the imaginary, better version of the world. By believing in it, we'll summon it forth'."

Moffat was the subject of some controversy when he produced a Doctor Who episode (Thin Ice) featuring a visit to 1814 Britain that looked far less white than the historical record indicates that 1814 Britain was, and he had the Doctor claim in-character that history has been whitewashed.

This is an example that serious, professional, powerful people believe that blackwashing is a moral thing to do. When someone like Moffat says that a blackwashed history is better, and Google Gemini draws a blackwashed history, I think the obvious inference is that Google Gemini is staffed by Moffat-like people who anticipated this result, wanted this result, and deliberately worked to create this result.

The result is only "bonkers" to outsiders who did not want this result.

Yishan says:

It demonstrates quite conclusively that with all our current alignment work, that even at the level of our current LLMs, we are absolutely terrible at predicting how it’s going to execute an intended set of instructions.

No. It is not at all conclusive. "Gemini is staffed by Moffats who like blackwashing" is a simple alternate hypothesis that predicts the observed results. Random AI dysfunction or disalignment does not predict the specific forms that happened at Gemini.

One tester found that when he asked Gemini for "African Kings" it consistently returned all dark-skinned-black royalty despite the existence of lightskinned Mediterranean Africans such as Copts, but when he asked Gemini for "European Kings" it mixed up with some black people, yellow and redskins in regalia.

Gemini is not randomly off-target, nor accurate in one case and wrong in the other, it is specifically thumb-on-scale weighted away from whites and towards blacks.

If there's an alignment problem here, it's the alignment of the Gemini staff. "Woke" and "DEI" and "CRT" are some of the names for this problem, but the names attract flames and disputes over definition. Rather than argue names, I hear that Jack K. at Gemini is the sort of person who asserts "America, where racism is the #1 value our populace seeks to uphold above all".

He is delusional, and I think a good step to fixing Gemini would be to fire him and everyone who agrees with him. America is one of the least racist countries in the world, with so much screaming about racism partly because of widespread agreement that racism is a bad thing, which is what makes the accusation threatening. As Moldbug put it:

The logic of the witch hunter is simple. It has hardly changed since Matthew Hopkins’ day. The first requirement is to invert the reality of power. Power at its most basic level is the power to harm or destroy other human beings. The obvious reality is that witch hunters gang up and destroy witches. Whereas witches are never, ever seen to gang up and destroy witch hunters. In a country where anyone who speaks out against the witches is soon found dangling by his heels from an oak at midnight with his head shrunk to the size of a baseball, we won’t see a lot of witch-hunting and we know there’s a serious witch problem. In a country where witch-hunting is a stable and lucrative career, and also an amateur pastime enjoyed by millions of hobbyists on the weekend, we know there are no real witches worth a damn.

But part of Jack's delusion, in turn, is a deliberate linguistic subversion by the left. Here I apologize for retreading culture war territory, but as far as I can determine it is true and relevant, and it being cliche does not make it less true.

US conservatives, generally, think "racism" is when you discriminate on race, and this is bad, and this should stop. This is the well established meaning of the word, and the meaning that progressives implicitly appeal to for moral weight.

US progressives have some of the same, but have also widespread slogans like "all white people are racist" (with academic motte-and-bailey switch to some excuse like "all complicit in and benefiting from a system of racism" when challenged) and "only white people are racist" (again with motte-and-bailey to "racism is when institutional-structural privilege and power favors you" with a side of America-centrism, et cetera) which combine to "racist" means "white" among progressives.

So for many US progressives, ending racism takes the form of eliminating whiteness and disfavoring whites and erasing white history and generally behaving the way Jack and friends made Gemini behave. (Supposedly. They've shut it down now and I'm late to the party, I can't verify these secondhand screenshots.)

Bringing in Yudkowsky's AI theories adds no predictive or explanatory power that I can see. Occam's Razor says to rule out AI alignment as a problem here. Gemini's behavior is sufficiently explained by common old-fashioned race-hate and bias, which there is evidence for on the Gemini team.

Poor Yudkowsky. I imagine he's having a really bad time now. Imagine working on "AI Safety" in the sense of not killing people, and then the Google "AI Safety" department turns out to be a race-hate department that pisses away your cause's goodwill.

---

I do not have a Twitter account. I do not intend to get a Twitter account, it seems like a trap best stayed out of. I am yelling into the void on my comment section. Any readers are free to send Yishan a link, a full copy of this, or remix and edit it to tweet at him in your own words.

[wageronancap]

Extremely lukewarm take: AI is neither a silver bullet solution to every problem to ever exist, nor is it the literal spawn of the antichrist -- it is a tool, and like all tools, can be used appropriately under the right conditions to great effect, or misused to great harm.

I've done some experiments with ChatGPT, and admittedly I've had a lot of fun. Detailed stories beyond the break.

I started with a statistical probability problem I knew how to describe but didn't know the math to calculate myself -- I'd just watched a number of engineering YouTube videos involving marble runs like Wintergaten, Engineezy, etc, and I wanted to know in a very limited, extremely controlled circumstance, how long could a device go before it "missed" -- before a ball that was supposed to be on a track, wasn't actually there because the reload mechanic is the random element. I gave ChatGPT a very involved prompt and regrettably I don't have it saved, but I was stupendously impressed with how it didn't need clarity from me -- it correctly interpreted the prompt, laid down some additional baseline assumptions, and then just did the math. Which honestly I expected -- computers are really good at counting. I was more pleased the AI had understood the prompt.

Later I asked to play a text-based version of XCOM 2 with the serial numbers filed off and then a Star Trek TNG-style bridge commander game. While the difficulty was too easy and it can lose track of a large cast of characters on occasion, it overall was very fun.

I also fed it a D&D 3.5 monster description and a fairly detailed character concept and it put meat on the bones (the finality of names is the bane of my existence as a worldbuilder); it even helped generate the leadership of a cult that serves the monster-character, and did so in the span of seconds -- it was able to take a concept and drum up roleplay-ready content on demand; content that would have taken me hours to develop by hand.

I recently used ChatGPT to test an item from the 2024 D&D release and immediately realized Chat's true strength for D&D as a balance tester. It is fabulously easy to import stat blocks (or homebrew non-OGL blocks) and just crunch a lot of trials to see the results.

Which led me to revisit an encounter I actually ran myself by hand some years ago -- a Warforged Titan against a Tyrannosaurus Rex. In my "canon" encounter, the tyrannosaurus went blow for blow for about four rounds before landing a critical hit and executing the titan -- a fucking epic brawl with an incredible finish on the fifth round.

I ran the same encounter with ChatGPT 100 times just now and realized just how anomalous my version of the encounter was:

Like, that's cool as fuck to be able to test, analyze, tweak, and retest without sinking a massive amount of time into playtesting, because the computer can play way faster than we fleshlings can -- and I can tune an encounter to give the feel I want it to impart to the players while still leaving room for players to innovate and surprise me as game master.

There are many things about my life I haven't dropped into the AI because it's not a counselor or therapist, it's not a financial advisor, it's not a friend -- it's a rock we put lightning into and taught to count, and it's really good at counting. It's even halfway decent at just helping make an entertaining story.

But it's not The Solution to Life The Universe And Everything.

[forkingandcountry-if]

Im not meaning to offend when i say this but Grammarly uses generative ai for its use, be it tone checking or spell checking. It may be unintended, but grammarly boasts how they use ai for their grammar checking and pacing things. Again, i only mean to inform, not offend, i am not meaning to start something and i apologize if it sounds like it. I have complete faith you are writing from the heart and imagination.

Hey, it's totally fine and I'll keep that in mind for next time. I do rely on Grammarly for the aforementioned purposes because I don't have beta testers to advise me on how text reads to other eyes besides my own and to catch little slip ups like misspelled words (& as we've seen a whole raft of little bugs 😅) but I do understand people being uncomfortable with work that has had any interaction with artificial intelligence.

I do hope despite it that you'll be able to enjoy Chapter 1 as much as is possible.

[socksbro1]

Date: march 14th, 2025

Discovery

I decided that while Remington-Gede is working on his memoir, I’ll work on Task 1. By examining the help file for the default cube, I figured out that there’s an automated system that is in charge of spawning and despawning them. It seems to rely on new interactions.

Essentially, as the simulation runs, new events occur that never occurred before. When that happens, two things are activated. One spawns a cube, and the other, I am unsure of. When the process is finished, the cube is supposed to despawn. Something seems to be preventing that with the one I use, but that’s perfectly acceptable for my purposes.

I decided to try querying the second process from the cubes DataFile. I got a type 4 error accidentally, then an unrecognized error that I’m now calling a type 5, which I’m pretty sure is calling a function without the necessary input values.

I looked at its help page, and it seems that it requires a new data type I’m calling a Soft Error Trigger. A soft error, as I’m calling it, is when the simulation is required to render the result of an interaction that has no recorded result. A Soft Error Trigger is a compressed form of that interaction, removing all extraneous data values. It seems to be able to compress any data type I discovered.

I was able to artificially generate a Soft Error Trigger by utilizing an interesting quirk I discovered in the help files. A Soft Error Trigger always follows a distinct format, starting with “Te4DTe” and ending with “Te4ETe”. The trigger also always contains at least 2 file references, often directly referencing Data Values. Finally, there is always a file references within that references an AdvUript that is at least a type 2.

I entered some values in this format, then pasted the command into the Cube. I used myself as the AdvUript, as Gede was busy. The Trigger I used referenced material 6 and the Cube’s UPV. I started daydreaming what it could cause. Would the UPV make another cube there? Would I receive an error? Would material 6 be visible, but not cause headaches? I decided that the most likely outcome was that a cube would spawn, made out of material 6, then would likely error itself out, since Material 6 doesn’t seem to work right.

Almost immediately, I felt a burst of pain that disappeared moments after. I then felt an urge to think of a 3rd tier AdvUript. My mind went to Gede instantly, and Remington froze where he was typing. A few minutes later, another cube appeared next to the original. Remington-Gede then trotted up to me and bit me.

Apperantly that second script is a sort of AI generation for possible interactions. What I was imaging was what would be tested. The burst of pain was the second cube spawning, then deleting itself since it occupied a space that was already filled by the first cube. Then, I forwarded the interaction, plus my idea to Gede, who had to figure out how to make it work. The bite was because he and Remington were in the middle of a discussion, and my idea had so many issues he was temporarily overloaded.

I now know I likely need Gede’s help with spawning new cubes. I also now know not to interrupt him.

Pleasant day,

Tester