Vetch

What constitutes serious work and how seriously have you tried to do serious work with them? While those trying to claim a 30B dense model can match Opus 4.6 are engaging in either beyond over-excessive over-exaggeration or performing rather routine tasks, it's disingenuous in the other direction to claim the latest open 1T models are not useful for serious work. I find those making such claims have rarely spent more than a few minutes on halfhearted attempts and often on recently obsoleted models.

Openweight models turned a corner around kimi 2.6, deepseek v4 pro/flash, hy3 and mimo 2.5 pro. Similar to how closed LLMs turned a corner around gpt 5.2 and opus 4.5.

While they remain a step behind closed frontier models, for real world tasks ranging across functional reactive programming, distributed systems, mathematical modeling, to-the-millisecond highly optimized spatial data-structures, complex compute shaders and shader effects and non-trivial systems involving parser combinators and algebraic effect systems, I can say that open models have very recently gone from useless to productive. For my work, mimo v2.5 pro is hands down better than sonnet 4.6.

Interesting thought to consider in principle but fails because gorilla brains continued to evolve too, just along a different path. They're not snapshots of ancestral species locked in time.

This sounds very like Licklider's essay on Intelligence Amplification: Man Computer Symbiosis, from 1960:

Men will set the goals and supply the motivations, of course, at least in the early years. They will formulate hypotheses. They will ask questions. They will think of mechanisms, procedures, and models. They will remember that such-and-such a person did some possibly relevant work on a topic of interest back in 1947, or at any rate shortly after World War II, and they will have an idea in what journals it might have been published. In general, they will make approximate and fallible, but leading, contributions, and they will define criteria and serve as evaluators, judging the contributions of the equipment and guiding the general line of thought.

In addition, men will handle the very-low-probability situations when such situations do actually arise. (In current man-machine systems, that is one of the human operator's most important functions. The sum of the probabilities of very-low-probability alternatives is often much too large to neglect. ) Men will fill in the gaps, either in the problem solution or in the computer program, when the computer has no mode or routine that is applicable in a particular circumstance.

The information-processing equipment, for its part, will convert hypotheses into testable models and then test the models against data (which the human operator may designate roughly and identify as relevant when the computer presents them for his approval). The equipment will answer questions. It will simulate the mechanisms and models, carry out the procedures, and display the results to the operator. It will transform data, plot graphs ("cutting the cake" in whatever way the human operator specifies, or in several alternative ways if the human operator is not sure what he wants). The equipment will interpolate, extrapolate, and transform. It will convert static equations or logical statements into dynamic models so the human operator can examine their behavior. In general, it will carry out the routinizable, clerical operations that fill the intervals between decisions.

https://www.organism.earth/library/document/man-computer-sym...

The relu/if-then-else is in fact centrally important as it enables computations with complex control flow (or more exactly, conditional signal flow or gating) schemes (particularly as you add more layers).

This is essentially what any relu based neural network approximately looks like (smoother variants have replaced the original ramp function). AI, even LLMs, essentially reduce to a bunch of code like

    let v0 = 0
    let v1 = 0.40978399*(0.616*u + 0.291*v)
    let v2 = if 0 > v1 then 0 else v1

    let v3 = 0
    let v4 = 0.377928*(0.261*u + 0.468*v)
    let v5 = if 0 > v4 then 0 else v4...

I'm not sure that's the fully right mental model to use. They're not searching randomly with unbounded compute nor selecting from arbitrary strategies in this example. They are both using LLMs and likely the same ones, so will likely uncover overlapping possible solutions. Avoiding that depends on exploring more of the tail of the highly correlated to possibly identical distributions.

It's a subtle difference from what you said in that it's not like everything has to go right in a sequence for the defensive side, defenders just have to hope they committed enough into searching such that the offensive side has a significantly lowered chance of finding solutions they did not. Both the attackers and defenders are attacking a target program and sampling the same distribution for attacks, it's just that the defender is also iterating on patching any found exploits until their budget is exhausted.

Then the point still stands, this makes things even worse given that it's adding its own hallucinations on top, instead of simply relaying the content or idealistically, identifying issues in the reporting.

Being tall doesn't automatically make you good or dominant at basketball, you can even be too tall. Wemby might just be at that threshold, but the unusual thing about him is his dexterity despite his height; such maneuverability and flexibility is trainable. I hear he also spent the summer training, likely harder than most.

It's an artifact of post-training approach. Models like kimi k2 and gpt-oss do not utter such phrases and are quite happy to start sentences with "No" or something to the tune of "Wrong".

Diffusion also won't help the way you seem to think it will (that the outputs occur in a sequence is not relevant, what's relevant is the underlying computation class backing each token output, and there, diffusion as typically done does not improve on things. The argument is subtle but the key is that output dimension and iterations in diffusion do not scale arbitrarily large as a result of problem complexity).

You are right and the idea of LLMs as lossy compression has lots of problems in general (LLMs are a statistical model, a function approximating the data generating process).

Compression artifacts (which are deterministic distortions in reconstruction) are not the same as hallucinations (plausible samples from a generative model; even when greedy, this is still sampling from the conditional distribution). A better identification is with super-resolution. If we use a generative model, the result will be clearer than a normal blotchy resize but a lot of details about the image will have changed as the model provides its best guesses at what the missing information could have been. LLMs aren't meant to reconstruct a source even though we can attempt to sample their distribution for snippets that are reasonable facsimiles from the original data.

An LLM provides a way to compute the probability of given strings. Once paired with entropy coding, on-line learning on the target data allows us to arrive at the correct MDL based lossless compression view of LLMs.

Unless you're also writing your own graphics and game engine from scratch, if you're making a truly novel and balanced game, then it should not be possible to crank out code with AI. When working in engines, the bulk of the work is usually in gameplay programming so the fact that its code is so predictable should be concerning (unless the programming is effectively in natural language). Not spending most of your time testing introduced mechanics, re-balancing and iterating should be triggering alarm bells. If you're working on an RPG, narrative design, reactivity and writing will eat up most of your time.

In the case you're working as part of team large enough to have dedicated programmers, the majority of the roles will usually be in content creation, design and QA.

Why would the proportion of high quality games increase? The number yes, but I expect not the proportion. Lowering the entry barrier means more people who have spent less time honing their skills can release something that's lacking in polish, narrative design, fun mechanics and balance. Among new entrants, they should number more than those already able to make a fun game. Not a value judgement, just an observation.

Think of the negative reputation the Unity engine gained among gamers, even though a lot of excellent games and even performant games (DSP) have been made with it.

More competitors does also raise the bar required for novelty, so it is possible that standards are also rising in parallel.

But why isn't this merely papering over a more fundamental issue with how these models are "aligned"? LLMs are, for example, not inherently sycophantic. kimi k2 and o3 are not, and Sydney, mentioned in the blog post, was most decidedly not.

In my experience, the issue of sycophancy has been longest in the Anthropic models, so it might be most deeply rooted for them. It's only recently, perhaps with the introduction of user A/B preference tests such as by lmarena and the providers themselves has this become a major issue for most other LLMs.

Thinking that simple actions like adding an anti-evil vector to the residual stream to improve behavior sounds naively dangerous. It would not surprise me if unexpected and unwanted downstream effects resulted from this; which a future paper will address too. Not unlike what happened with tuning for user preference.

The brain is certainly vastly more energy efficient at inference than LLMs on GPUs. But it looks like you're trying to make a different argument, that an LLM can spend less energy than a human to complete a given task. Unfortunately, you have not made that argument and I won't be reading unverified LLM output that might contain hallucinated steps or claims.

V3/R1 scale models as a baseline, one can produce 720,000 tokens

On what hardware? At how many tokens per second? But most importantly, at what quality? I can use a PRNG to generate 7 billion tokens at a fraction of the energy use of an LLM but those tokens are not going to be particularly interesting. Simply counting how many tokens can be generated in a given time frame is still not a like for like comparison. To be complete, the cost required to match human level quality, if possible, also needs accounting for.

Deeply thinking humans expend up to a a third of their total energy on the brain

Where did you get this from? A 70B LLM? It's wrong or at best, does not make sense. The brain barely spends any more energy above its baseline when thinking hard (often not much more than 5%). This is because most of its energy use is spent on things like up-keep and maintaining resting membrane potential. Ongoing "Background activity" like the DMN also means the brain is always actively computing something interesting.

That math is for random projections? Note that JL lemma is a worst case guarantee and in practice, there's a lot more distortion tolerance than the given bounds would suggest. Concepts tend to live in a space of much lower intrinsic dimensionality than the data's and we often care more about neighbor and rank information than precise pair-wise distances.

Also, JL is only a part of the story for the transformers.

Compounding with learn and iterate, humans also build abstractions which significantly shorten the number of steps required. These are more expressive programming languages, compilers and toolchains. We also build engines, libraries, DSLs and invent appropriate data-structures to simplify the landscape or reuse existing work. Besides abstractions, we build tools like better type systems, error testing and borrow checkers to help eliminate certain classes of errors. Finally, after all is said and done, we still have QA teams and major bugs.

he's inability to see its application to modern compute held the field back by years.

I find Schmidhuber's claim on GANs to be tenuous at best, but his claim to have anticipated modern LLMs is very strong, especially if we are going to be awarding nobel prizes for Boltzmann Machines. In https://people.idsia.ch/%7Ejuergen/FKI-147-91ocr.pdf, he really does concretely describe a model that unambiguously anticipated modern attention (technically, either an early form of hypernetworks or a more general form of linear attention, depending on which of its proposed update rules you use).

I also strongly disagree with the idea that his inability to practically apply his ideas held anything back. In the first place, it is uncommon for a discoverer or inventor to immediately grasp all the implications of and applications of their work. Secondly, the key limiter was parallel processing power; it's not a coincidence ANNs took off around the same time GPUs were transitioning away from fixed function pipelines (and Schmidhuber's lab were pioneers there too).

In the interim, when most derided Neural networks, his lab was one of the few that kept research on Neural networks and their application to sequence learning going. Without their contributions, I'm confident Transformers would have happened later.

It's clear to me no one read his early paper's when developing GANs

This is likely true.

self-supervision/transformers.

This is not true. Transformers came after lots of research on sequence learners, meta-learning, generalizing RNNs and adaptive alignment. For example, Alex Graves' work on sequence transduction with RNNs eventually led to the direct precursor of modern attention. Graves' work was itself influenced by work with and by Schmidhuber.

The non-o-series models from OpenAI and non-Opus (although I have not tried the latest, so it's possible that it too joins them) from Anthropic are cloyingly sycophantic, with every other sentence of yours containing a brilliant and fascinating insight.

It's possible that someone already on the verge of a break or otherwise in a fragile state of mind asking for help with their theories could end up with an LLM telling them how incredibly groundbreaking their insights are, perhaps pushing them quicker, deeper more unmoored in the direction they were already headed.

This contains a common misstep (or misgeneralization of an analogy) among those who are much more familiar with computers than with the brain. The brain is not digital and concepts like frames per second and resolution don't make much sense for vision. First, there aren't frames, neuron activity is asynchronous with changes to sensory neuron firing rate responding to changes in the environment or according to saliency.

Between the non-uniformity of receptor density (eg fovea vs peripheral vision but this is general across all senses), dynamic receptor fields and the fact that information is encoded in terms of spike rate and timing patterns across neural populations, the idea of pixels in some bitmap at some resolution is beyond misleading. There is no pixel data, just sparsely coded feature representations capturing things like edges, textures, motion, color contrast and the like, already, at the retina.

While hundreds of trillions of photons might hit our photoreceptors, > 99% of that is filtered and or compressed before even reaching retinal ganglion cells. Only a tiny fraction, about 10 million bits/sec, of the original photon signal rate is transferred through the optic nerve (per eye). This pattern of filtering and attentive prioritization of information in signals continues as we go from sensory fields to thalamus to higher cortical areas.

So while we might encounter factoids like: on the order of a billion bits per second of data hit photoreceptors or [10Mb/s transferred](https://www.britannica.com/science/information-theory/Physio...) along optic nerves, it's important to keep in mind that a lot of the intuition gained from digital information processing does not transfer in any meaningful sense to the brain.

I think the point remains that few have been able to catch up to OpenAI. For a while it was just Anthropic. Then Google after failing a bunch of times. So, if we relax this to LLMs not by OpenAI, Anthropic or Google, then Deepseek is really the only one that's managed to reach their quality tier (even though many others have thrown their hat into the ring). We can also get approximate glimpses into which models people use by looking at OpenRouter, sorted by Top Weekly.

In the top 10, are models by OpenAI (gpt4omini), Google (gemini flashes and pros), Anthropic (Sonnets) and Deepseeks'. Even though the company list grows shorter if we instead look at top model usage grouped by order of magnitude, it retains the same companies.

Personally, the models meeting my quality bar are: gpt 4.1, o4-mini, o3, gpt2.5pro, gemini2.5flash (not 2.0), claude sonnet, deepseek and deepseek r1 (both versions). Claude Sonnet 3.5 was the first time I found LLMs to be useful for programming work. This is not to say there are no good models by others (such as Alibaba, Meta, Mistral, Cohere, THUDM, LG, perhaps Microsoft), particularly in compute constrained scenarios, just that only Deepseek reaches the Quality tier of the big 3.

There is skill to it but that's certainly not the only relevant variable involved. Other important factors are:

Language: Syntax errors rise, and a common form is the syntax of a more common language bleeding through.

Domain: Less so than what humans deem complex, quality is more strongly controlled by how much code and documentation there is for a domain. Interesting is that if in a less common subdomain, it will often revert to a more common approach (for example working on shaders for a game that takes place in a cylinder geometry requires a lot more hand-holding than on a plane). It's usually not that they can't do it, but that they require much more involved prompting to get the context appropriately set up and then managing drifting to default, more common patterns. Related is decisions with long term consequences. LLMs are pretty weak at this. In humans this one comes with experience, so it's rare and an instance of low coverage.

Dates: Related is reverting to obsolete API patterns.

Complexity: While not as dominant as domain coverage, complexity does play a role. With likelihood of error rising with complexity.

This means if you're at the intersection of multiple of these (such as a low coverage problem in a functional language), agent mode will likely be too much of a waste for you. But interactive mode can still be highly productive.

LeBron is one of the rare individuals at that intersection of high athleticism and mental capability. It's why at the age of 40, well past his athletic prime, he's still a top NBA player. He has Magnus-level chunking ability enabling prodigious memory for games, he has fast processing and court vision, being able to leverage symmetries to automatically adjust for current player orientations to predict opponent plays. It's what allows him to make passes that seem impossible--he sees windows open up based on predicted player movements, not just current positions. Like that famous Wayne Gretzky quote.

It's a super rare archetype of athleticism/size+mental that only the likes of LeBron, Jokic and Magic Johnson have occupied (not meant to be an exhaustive list).

The essence of the article is that self-correction exists as a nascent ability in base models already (more robustly in some like Qwen than others). This is highly reminiscent of Chain of Thought, which was found to be a capability already present in base models too. The result of RL is to reinforce already present authentic self-correction patterns and down weight superficial self-correction.

Thoughts:

- An analogy you shouldn't zoom too close into is going from CoT to reasoning traces is like going from purely ballistic trajectories to including navigation and thrusters. RL is for learning how to use the thrusters for adjustments based on its internal encodings of rare samples† where some author fully spelled out their thought process.

- This might also explain why SFT on reasoner traces seems to be surprisingly effective. If it were purely an RL mediated phenomenon, SFT for reasoning would not work nearly as well.

- Deepseek struggled to get RL to work on smaller models, if this is replicated, it might be the case that larger models encode self-correction patterns more robustly while having them as more probable.

- Imitating traces is easier than pure RL for bringing such patterns to the fore, for smaller models. However, we still want models to learn how to dynamically adjust their thrusters, SFT does not provide ample opportunity for this. Further training with RL or alternatively, replacing SFT with methods like [Critique Fine-Tuning](https://arxiv.org/abs/2501.17703) are needed.

- The article incidentally reinforces that having a low temperature means consistency not correctness. Except for high confidence scenarios, the highest greedily computed probability answer is generally less likely to be among the best ones the model can give.

†Question: First thought is blogs by people who discuss what didn't work. But, I wonder how much of reasoning model patterns and ability is shaped by Detective Conan transcripts?

I'm also ashamed to say I've also never seen any of his movies and TV series but this still hits hard because of his influence on some my most cherished fictional properties. These are Alan Wake/Control, Silent Hill 1&2, Returnal and Disco Elysium.

Actually, his influence on how surrealist fiction is presented throughout all media cannot be understated. I was surprised to read even the original Zelda has him as an influence. Majora's Mask does feel particularly Lynchian.

It would not surprise me if the Souls games and at least the later Berserks (late 90s/early 2000s forward) were either directly or 1-step indirectly influenced by Lynch.

This is an interesting proposition. Have you tested this with the best open LLMs?

This seems wrong. Reasoning scales all the way up to the discovery of quaternions and general relativity, often requiring divergent thinking. Reasoning has a core aspect of maintaining uncertainty for better exploration and being able to tell when it's time to revisit the drawing board and start over from scratch. Being overconfident to the point of over-constraining possibility space will harm exploration, only working effectively for "reasoning" problems where the answer is already known or nearly fully known. A process which results in limited diversity will not cover the full range of problems to which reasoning can be applied. In other words, your statement is roughly equivalent to saying o3 cannot reason in domains involving innovative or untested approaches.

The information on the creative step which you provided to o1, was also the key step and contained almost all the difficulty. The hope is that 2025 models could eventually come up with solutions like this given enough time, but this is also a toy problem. The question is how much clever answers will cost for real world complex problems. At present it looks like, very much.

Excellent point. The hope is reasoning LLMs will make a difference for such problems. But it's also a great example of why the those who think being able to have the LLM iterate more will be crucial to reasoning are off base. There are many computations that a transformers (or humans for that matter) are not well equipped to represent internally, tool use during the reasoning process is unavoidable for all but the artificial or knowledge heavy problems.

Small examples, throwaway but involved calculations, prototypes, notes of what didn't work and what's promising are what's crucial for novel reasoning. It goes beyond just search or iterative refinement; there is no royal road to reasoning.

I've also used it to make a function tail recursive, which is useful in languages with TCO.

Note that this is an expressibility (upper) bound on transformers granted intermediate decoding steps. It says nothing about their learnability, and modern LLMs are not near that level of expressive capacity.

The authors also introduce projected pre-norm and layer-norm hash to facilitate their proofs, another sense in which it is an upper-bound on the current approach to AI, since these concepts are not standard. Nonetheless, the paper shows how allowing a number of intermediate decoding steps polynomial in input size is already enough to run most programs of interest (which are in P).

There are additional issues. This work relies on the concept of saturated attention, however as context length grows in real world transformers, self-attention deviates from this model as it becomes noisier, with unimportant indices getting undue focus (IIUC, due to precision issues and how softmax assigns non-zero probability to every token). Finally, it's worth noting that the more under-specified your problem is, and the more complex the problem representation is, then the quickly more intractable the induced probabilistic inference problem. Unless you're explicitly (and wastefully) programming a simulated turing machine through the LLM, this will be far from real-time interactive. Users should expect a prolog like experience of spending most of their time working out how to help search.

Trivia: Softmax also introduces another problem: the way softmax is applied forces attention to always assign importance to some tokens, often leading to dumping of focus on typically semantically unimportant tokens like whitespace. This can lead to an overemphasis on unimportant tokens, possibly inducing spurious correlations on whitespace, this propagating through the network with possibly unexpected negative downstream effects around whitespace.