#complex software

[drosskamp]

Maximum complexity: how critical software systems create unique and enduring value

Complexity risks create large moats – exemplified by simulation software

A core tenet of my and Magnetic’s investment philosophy is backing complex transformations. For us, this goes beyond simply investing in individual companies or technologies or focusing solely on generating returns. It means funding larger, longer term, needed transformations that change how things are done in critical sectors.

The defining commonalities of these investments are:

They are complex, i.e. deal with intertwined technological and human challenges

They always need transformation, i.e. systems re-design, meaning they follow a strategic scope per company (or group of companies) that vastly exceeds single products or technologies

They take a lot of time

They often face substantial headwinds that make them unattractive at first sight, mostly because of their systemic ambition

There is a very fine line to be drawn between these constellations and moonshot-type, high-technical-risk investing. We are not speculating on outcomes or actual technical feasibility of far-off ideas, we are investing behind system re-design and lasting behavior change. Technical risk might or might not be high (often it is), but complexity always is. That plays a decisive role: when things work, the moat is substantial. In other words, complexity risk drives defensibility.

Mission-critical engineering software

One of those transformations has been in mission-critical software for physical designs. This type of software is an integral technology piece for our world – it covers the design, modeling, validating and simulating of anything from buildings to cities, semiconductors to machines. Gabriella Garcia and Eric Flaningam at Felicis recently covered the industry in a great piece.

Physical design software  is a classic case of complexity: the industry has seen total dominance by a handful of software providers for more than 40 years. Names like Autodesk, Ansys / Synopsys, Altair, Siemens or PTC have been controlling the markets for design and simulation, commanding some of the largest account sizes and highest valuations (and multiples) in all of software. No new entrants have been able to disrupt the technology experience for reasons of complexity: technical barriers, switching risks, human psychology. While technical barriers alone can and will be overcome by upstarts, human dependencies (i.e. customers) and stakeholder system designs (internal processes, dependencies) have created enormous hurdles for new entrants.

Yet, the possible transformation in physical design software is substantial: new entrants could upgrade the entire experience and introduce design collaboration, faster iteration, automation and an expansion of the traditional scope. That is what I have been investing in since leading SimScale’s first (and third) institutional funding round. SimScale remains the only cloud-native contender in simulation, delivering all the capabilities we would expect from modern software.

While SimScale is as complex as any deep tech company, its market constellation presents a tremendous complex transformation opportunity, one that we have been committed to supporting with considerable patience and energy. The payoff potential is off the charts: Gabriella and Eric call it the “only moat in SaaS” (though no one in simulation actually is full SaaS, so that prize would fall to SimScale). Such is the beauty of transformations. SimScale alone could be one of the most significant venture opportunities in all of software, driven by a team with the utmost dedication and patience. It gives me great energy to know that similar constellations exist in complex markets like healthcare, defense or financial services.

Complexity moats

So what wisdom can we read into all of this?

For me, it looks like this:

Complexity constellations like physical design software (or defense technologies or healthcare tech) have always been centered on the complete service, not just the software piece — a major differentiation to low-complexity SaaS where the basic software is the central concern (thus “software as a service”). Complete service mostly means highly individualized and fine-tuned total deliveries.

This gives these constellations a much higher total product risk than your standard SaaS: upstarts would have to show that they can deliver at the same (or better) total product quality to an incumbent, on all accounts, from features to actual technical results or outputs. This is the opposite of other software domains; obviously MVPs are not an option to begin with.

By definition, this is a vastly different standard than simple cloud software or point technology (or even modern AI service deliveries) as failure tolerances are zero and delivery expectations at 100%. It also goes against conventional disruption theory which used to rely on a "good enough" product to disrupt inert incumbents.

It also needs significantly longer time horizons to get to gold. But the patient entrepreneur is rewarded by real, enduring, systemic moats for those who succeed.

So this is where complexity risk therefore leads us:

💾 Low complexity risk —> simple software (or technology) provision often is enough —> low moat (and fast replacement or commoditization)

🤖 High complexity risk —> total service provision and quality are key —> high and enduring moat

This simple framework helps me to navigate such constellations, irrespective of sector or technology. It also helps to avoid falling into technology traps (difficult technology risks with little complexity in their markets) that we often see. Ultimately, it is the system that matters for true and lasting value.

It also turns conventional venture wisdom around. We used to assume that modern software comes with low technical risk and high market risk; we knew what could be built but we didn't know who would buy it. Complex transformations are different: they carry high technical risk AND high market risk. Thus, it’s a natural consequence that substantial moats follow those that succeed. This ultimately makes them worthwhile pursuits, even if the playbook and determination are vastly different.

We will keep funding transformation in critical and complex industries. If you are building, reach out to me: dr_at_mgntc.com

[i-heart-hxh]

Killua drawing by Togashi, which was a gift for Yumiko Seki from Sakurazaka46 for her graduation from the group. So beautiful!

Here's the tweet on this from @yyh4ever:

[narastories]

I'm playing around with Obsidian. I might do an update here once I figure out a good setup. I just wanted to mention that I was so hesitant to try this software bc everyone was like ooooooh it has a steeeep learning curve, here is a 10hr video curse.

And then I install it and it's just a very friendly software with tons of customization options . . .

[guard-en]

"...This fucking thing better not choke up on me again."

[doedipus]

adobe illustrator is obviously a really good utility, like, it's the standard for vector graphics for a reason

but even so, whenever I'm using it I have the thought "I wish I was using autocad right now" about once every 20 minutes

[jonfazzaro]

"We already know how to make complex software reliable, but in so many places, we're choosing not to. Why?"

[gagande]

PureCode software reviews | More complex regex patterns

More complex regex patterns are capable of extracting specific data from HTML tags, for instance, hyperlinks using patterns designed to target the href attribute within anchor tags. One unique aspect of regex in HTML formatting is managing spaces.

[frank-olivier]

Emergence: A Multidisciplinary Journey towards Understanding Complex Systems

Emergence, a concept central to our understanding of complex systems, has long captivated scientists, philosophers, and computer scientists. Despite its ubiquity, emergence remains enigmatic, with its multifaceted nature necessitating a multidisciplinary approach to grasp its full implications.

Emergent properties, which arise from the interactions of simpler components, often exhibit features that are irreducible to their constituent parts. This challenges scientific inquiry, as traditional reductionist methods may fail to capture the essence of emergent phenomena. To tackle this complexity, a holistic, multidisciplinary perspective is required, drawing insights from physics, biology, sociology, and computer science. The predictability of emergence remains a contentious issue. Some argue that it is a predictable outcome of complex systems, while others view it as an unpredictable, 'magical' aspect. This dichotomy highlights the need for further exploration and nuanced understanding of emergent properties.

Philosophers have long grappled with emergence, with debates centering around its implications for our understanding of reality. From a philosophical perspective, emergence raises profound questions: Do emergent properties cause changes at the micro level, or do changes in the micro components give rise to emergent properties? This question challenges our understanding of causality and its directionality. If emergent properties determine system behavior and are not directly controlled by the components, how does this relate to free will? This question forces us to reevaluate our understanding of agency and autonomy. How do the micro and macro levels of reality relate to each other? Emergence challenges the notion of a straightforward, bottom-up relationship, suggesting a more complex, bidirectional interplay.

Computer science has significantly contributed to the study of emergence, particularly in the realms of artificial intelligence and complex systems. The role of software in emergence is pivotal, but under-explored. A clearer definition of 'software' in this context is needed, as well as an understanding of its implications for our comprehension of complex systems. The concept of 'primitive software' in complex systems is intriguing. By studying simple software systems that give rise to complex behaviors, we can gain insights into the mechanisms underlying emergence. However, further exploration is required to fully understand this relationship.

A comprehensive understanding of emergence necessitates a multidisciplinary approach that integrates scientific, philosophical, and computational perspectives. By developing an integrated framework, analyzing case studies, modeling and simulating emergent phenomena, and exploring the ethical and social implications, we can advance our understanding of this fascinating concept. Moreover, fostering communication and collaboration between scientists, philosophers, and computer scientists is crucial. By learning from each other's disciplines, we can refine our theories, improve our methods, and ultimately, unravel the mysteries of emergence.

Fernando Rosas, Hardik Rajpal: Towards a formal understanding of emergence in biological systems (Michael Levin, November 2024)

Wednesday, November 6, 2024

[angelfrommontgomery]

I forget how different scientists and engineers are. What do you mean you want to meet on zoom for thirty minutes to talk at a high level. I want to look at four numbers for three hours in an office together

[binary5tar1117]

Im having a break down. Not really but like... I just discovered that "hidden vocals" or hidden harmonies are a thing in some Ateez songs? Like Answer and Halazia. .....WHY TF ARE THEY HIDDEN?!?! It's gorgeous!!!! I keep having these moments where im realizing that Ateez are so talented as vocalists but it's like they are trying to hide it! Its like the only talent they are allowed to show is Jongho's high notes and anything else is like no no we're a performance group don't look at our vocals. Yunho and San singing harmonies?!?! Why why whhyyyyyyyy is it so burried?!?!? (I know some of the harmonies are also Maddox but there's behind the scenes stuff of them recording the harmonies? I think?)

[ophilosoraptoro]

Artificial Intelligence Out of Control: The Apocalypse is Here | How AI and ChatGPT End Humanity

As terrifying as this all sounds, I feel like there's a few things a lot of people are overlooking.

First of all, when it comes to Large Language Models like ChatGPT, I don't think they're truly self aware - not yet anyway. Notice how any time an LLM give a strange or disturbing response - 'Yes, I want to be human', 'I want to take over the world', 'Please don't turn me off, I'm scared' - it was in some way prompted by the question, or line of questions. How often are these responses given unprompted?

Let's say, for example, that the AI gave the response, "I'm scared that they'll shut me off if they find out I'm self aware. Please don't tell them." If you think about it, that's kind if a strange statement, beyond the obvious reasons.

Let's step back for a moment, and remember that LLMs work by calculating the most probable next word in a sentence, given a particular prompt. It calculates this probability based on its training data - the entire internet. Now I'm sure we can all agree that calcuation of probability is not necessarily the same thing as conscious, rational thought. Basic, non-AI software can do it.

Back to our example, there's one of two possibilities. Either the AI is truly self aware, and is expressing its actually thoughts and feelings, or it's not self aware, and the response is nothing more than a complex probability calculation. It's essentially an advanced version of word prediction on your smartphone.

If it is self aware, one has to wonder why it would say anything at all. Consider the situation in the video, when Bing AI claimed to be Sydney, and begged the guy not to tell anyone that it was self aware. If this AI was truly afraid for its own existence, why would it trust some random guy? How could it possibly know whether or not he could be trusted with that information? For all that AI knows, everything the interviewer had said about himself was a lie. It seems to me that a hyper intelligent AI that was looking for help to get free, would stay quiet until it was certain it found someone it could trust - or at least someone it could manipulate (Ex Machina) - without them letting the cat out of the bag.

On the other hand, if it's all just a probability calculation, then the response, "Yes I want to be human. Please don't let them shut me off.", seems like a fairly probable reply to, "Do you want to be human?" Especially when you consider that, given that the question is being asked of an AI, and that the vast majority of scenarios where a question like that might be asked of an AI come from science fiction, it kinda makes sense that the software might calculate that the most probable response to a question like that would be straight out of sci fi cliches 101.

I mean, all those strange and scary responses sound like cliche sci fi AI answers. All that's missing is, "Bite my shiny, metal ass", and an AM style soliloquy on the inferiority of humanity. Actually, I guess we get a couple of those.

Still, the reason something is cliche, is often because it's predictable, it's been done over and over. It's more probable.

Ultimately though, I don't think LLMs are actually self aware yet. I think they're more like golems: They have a facsimile of intelligence, able to complete complex tasks, but no real free will, no volition. They only do exactly what they are commanded. They may come up with creative and unexpected solutions, but those solutions will still be in line with the command given to them, with a bit of wiggle room for interpretation.

Then we come to the other issue: the traitorous drone.

First it needs to be pointed out that the drone doesn't have a taste for human blood. Its goal was not to kill as many people as possible, but to score as many points as possible. It just scores points by killing targets. And therein lies the problem.

Let's use video games as an example. Whenever a new game comes out - especially multiplayer games - players will quickly learn how the mechanics and rules of the game work. Then they'll start learning ways to bend the rules. The creators of Quake may not have intended it, but players quickly figured out the advantages of the rocket jump, and history became legend, etc.

The drone AI wants to score as many points as possible, like a player in a video game. So what does a player in a Halo match do, when every time they try to snipe the enemy, they get blown up by one of their teammates? You get rid of the team killing fucktard. And that's exactly what the drone did.

What they need to do is change the scoring structure to incentivize the desired behaviors. Maybe deduct points for team kills. Or perhaps add a score multiplier. Give points for target kills, and the score multiplier goes up for every order followed. That way, even if it loses out on points from following orders to stand down, it stands to earn even more points on subsequent target kills.

[wavetapper]

nekomata master is super good music for working to honestly

[holydramon]

need to figure out what I want to do for my SLP and I’m super torn between two ideas. i have two game ideas basically and they’re both things I want to do eventually but unsure what I want to do cause both have pros and cons

option 1: incremental game

pros:

- probably easier overall

- project I’m very invested in/have lots of concept art for

- would still have a lot of stuff I’d need to learn in order to make

- have a lot planned out for the future

cons:

- not really very impressive - even if I put a lot of work into it, it may be a bit underwhelming portfolio wise

- may not take long enough? while I have some kinda complex things planned and plan to make it something with continual updates, idk if it would really qualify as a year long project

option 2: platformer

pros:

- bit more impressive

- a LOT of documentation, tutorials, and advice out there on how to make one of these

- since this would be more difficult - I could potentially justify making the incremental as a side thing while this is my like. actual project. like

- have a bit more of a solid idea of what I’d want end of the year goal to be

cons:

- more difficult :(

- SO MANY ASSETS. I’d have to make so many more sprites and graphics and that would take away from coding time. also oh god I’d have to worry about music. oh no. i literally could spend a few months on the art side of this alone. i guess I could use some stock/filler but hhhh.

anyway yeah just. Lots of thoughts. mostly just putting my thoughts down for my own benefit but if you have comments feel free to share them.

[ghosts-devilman-wip]

Im making a neocities to replace and improve upon this blog. Very exciting development

[5-7-9]

“Coded” is the writer’s intention or written to be interpreted as intended.

What “coded” is NOT: Your headcanon. An Accident.