musebox35

I agree that it is a bit weird, kind of coding in natural language but also not exactly like that. Maybe as the agents stabilize and there are proper new tools for this new process we can relax a bit and it will become the new normal. For now it feels like surfing over an ever changing seascape, exciting but tiring at the same time.

> Some things you need to understand-others, not so much.

This is the core of the matter though, knowing what you need to understand and what you can ignore is the actual programmer's skill. It requires you to have a clear mental picture of both what you are trying to build and what the underlying machine will do when you are finished.

You need to understand the abstractions, but also where they leak, when they won't match reality, and how. This is why knowing computer architecture and assembly helps you to optimize your code even if you are coding in a high level language.

The problem with coding agents is that they are tuned to work on all contexts so they always fill an underspecified request by optimizing the average case and often without stating all the assumptions that they make. So you still need to understand what you specified and what got filled in automagically by the agent. My experience is that they (even the paid frontier models) are poor judges of the most important assumptions they make, which will might be corrected by a prompt or a tool output in which case it is fine. Otherwise it will be ignored and steer the model into a weird loop. It then tries to fix things but can not do so since its mental model is totally broken now.

Do not get me wrong, I am so happy to let the agent handle tool building (especially those that involve a web UI) and fill in the CLI command line argument parser. But every time I trust the agent by relying on it to drive the mental model of what we are doing, I got seriously bitten. Well, maybe that should not be surprise me, but I can understand the confusion of less experienced programmers and non-coders. It must really be frustrating to be able to build so much, but also not to be able to fix seemingly small issues.

It was Nvidia’s competitors’ job to ensure this never happened but hardware companies rarely value the software stack as much as they should have. AMD screwed up several times to build a similar tool and Intel did not manage to create a proper programming model for their vector instruction sets despite having some promising internal efforts. Cuda allowed AlexNet to usher the deep learning era, written by university researchers. The flash attention implementation allowed quadratic attention to have reasonable memory requirements again written by a university professor and used by everyone. If I were to complain about Nvidia, I would complain about them abandoning the personal computing track for datacenter profits, definitely not about pushing Cuda.

This is one question that is not just for kids but for all "gpu poor" adults as well. One concept that I found useful is "toy models": small compute models that replicate a certain, not very well explained phenomenon that the large models have. Of course, you can't have a toy model for everything, some behaviors are only visible at scale. But, there is a bunch of things like in-context learning that you can play with without burning a ton of compute.

The best analogy is strip mining (big labs) vs cave exploration (solitary/small teams). I think this is how science progresses at the boundaries by smart/curious/hardworking individuals because depth is a requisite for finding the right questions and then the answer. It is not for everyone and it does not always work. But you learn a ton even if it doesn't pan out to be a big breakthrough.

A more apt analogy is the diminishing understanding of assembly language and machine instruction sets. I love programming and something got lost in both transitions, first to high level languages and now to coding agents. I am quite sure some contexts will require programming by hand for a long while as there are still people who are writing amd64/avx512/sass by hand. I hope in time we will find new ways to enjoy solving problems with this new programming modality. By now, I am pretty sure it is here to stay and evolve from its current crude form.

That also surprised me, the diffusion gemma is actually a bunch of architectural changes to make discrete diffusion practical and an sft + distill of the Gemma 4. One cool trick they do is to take advantage of the entropy of the estimated probability distribution to adaptively adjust the required number of diffusion steps for inference. Image diffusion only yields the score function, a gradient instead of the probabilities, so the same trick does not apply.

This is one of the dilemmas that I am trying to wrap my head around. I love optimizing software pipelines, which often boils down to figuring out the operational constraints that the compiler and the generic libraries can’t assume. Then I exploit these to squeeze out performance. But in a world I can start from scratch and code a domain specific solution from line zero in a matter of hours/days, I do not need general libraries as much as I used to. On one hand the code won’t be as well tested as a good general library. On the other hand, it also won’t have a plethora of library bugs that are there because the code is generic and opaque. One counter argument is that things are never static and you can’t have specific code for too long. A counter to that is that you can then change the code to be specific to the new reality at very low cost. This is the mental loop I ride constantly. Disclaimer: My circumstances are definitely not general, I am not writing code that is truly large scale.

I would add that it is not just capacity, but also negotiation ability. With scale comes the ability to negotiate better prices than everyone else. Even if you can find capacity for your smallish user base, your inference cost can not match these companies unless you have a technical advantage for your inference cases. Squeezing the hardware requires request batching and caching which are far easier at scale and sustained user activity.

The biggest advantage of tpus is the high bandwidth fiber optic interconnect between them that allows distributed computing on pods with thousands of tpus and the co-design of cooling systems that go with their racks. I do not think that we will see personal tpus any time soon.

Sadly no. Despite the name, the experts are not routed per concept or topic but per token. So for the same sentence you might activate multiple experts for different tokens. What it solves is the distributed training and inference problem. As long as each expert fits a single gpu, coordinating the model evaluation is much easier and it is faster. It does not buy as much for running on a single device though still less costly than a dense version.

dev variants are usually cfg distilled which means that directly finetuning isn’t as effective. In the past, for the flux2 klein models,they released base versions that are not distilled. So it will probably be a while before you can fully take advantage of the open weights.

Thanks for posting this. This is the tinkerer mentality. It is not for everyone, but certain things can only be learned in that way. It is the best antidote to AI paranoia. There is much that does not transfer between frontier models and local ones. There is that. But you can not tinker as much as you can with the former.

Training isn’t a single homogeneous step. It starts with pretraining which requires bulk PB of data but you have less quality concerns here. You cover the whole data distribution. Later stages require less and less but increasingly higher quality and complex datasets. The late stage ones are highly curated and might even be sourced from world subject experts. This is where frontier labs with big pockets have the advantage.

Thanks it looks promising :)

I also have been using emacs for almost anything for the past 20 years. I had to switch to VSCode for coding over a remote ssh connection to cloud VMs. The client/server split of vscode felt superior over the ssh connection and the emacs alternatives was not up to the same level of performance two years ago. Do you know any progress on that front? I would love to go back to emacs as my daily driver but I am sensitive to lags when I type / execute commands. Have you worked with ai assistants over a remote connection?

True enough, but that is true for all the products I buy. I do not expect to control every product I own. For some I prefer to have more control, for others I just need something that works out of the box. There is always an initial bias for trust when you buy something otherwise you would not spend your hard earned money on it.

“Fool me once, shame on you. Fool me twice, shame on me. Fool me three times, shame on both of us.” -- S. King

It is much more reasonable to do it in a visible / flagged way. At least you have visibility over the quality of service you get as a customer.

Silent treatment is a breach of trust, what you buy changes depending on the context based on the goals of the producer. It is like your computer silently blocking ads from competitors at the hardware level, which is crazy. I think they erred on the wrong side of things due to IPO pressure.

At least there is competition from multiple companies. Still it is best to have personal benchmarks for the domain you are working on to have a real evaluation of the value you get for the money/time you spent on these products. Without trust, that might be the only way forward to keep the companies honest.

This happens eventually in all sectors, a good magazine/website that does independent product evaluation is priceless. Sadly, the new ad-driven internet decimated those that worked great in the 90/00s. Still there are independent blogs that does some evaluation and that is better than nothing.

I work on open source text-to-image finetuning of open source models like zimage/flux2 klein 4b and inference time latency optimization. The moment I read the silent treatment, I went ahead and cancelled my subscription too since I would never know whether the models they launch will silently corrupt my output. This is totally unacceptable. There is a big difference between silent / flagged if you are doing ml research but not at frontier capability.

This goes on to show that - All that interpretability / safety research they are doing can also be weaponized against customers (steering vectors, intent classification, ...) in the name of safety from malicious actors. - If they deem profitable, they might nerf to original model and its training data for ml research at a bulk scale and then they won't even have to announce it so long as the overall benchmark score stays high enough.

As the IPOs get closer, they can do whatever they want to assure the investors that they have a moat that can not be crossed over by their own products. Considering this affects all ML researchers/students at universities, smaller scale research labs, this is just "cutting the branch you are sitting on".

SFT + RL connection to model/hypothesis search is insightful. Brute force / scalable search is where Sutton's Bitter Lesson also points to. Once your search domain is small compared to your search budget, that makes a lot of sense.

If I get your meaning right, SFT creates the right inductive bias so that the RL search + reward guidance does the trick.

For novel discovery, the question might then be whether the inductive bias builds a strong enough prison so no new discovery is possible by RL or if the search can escape the boundaries set by SFT given enough randomization and the right reward function.

I know that RL is usually not performed at inference time, but in-context learning mechanisms might be developed by RL to discover at test time. Edit: I would love to hear if that actually happens or not, like new induction heads (https://transformer-circuits.pub/2022/in-context-learning-an...) forming during RL. I really have no idea.

I understand the skepticism. I am worried about the implications of AI as well. The deeper issue at stake is that the depth of scientific knowledge has been increasing for a very long time. Now you get to have a PhD in esoteric subproblems and that slows down research especially if the discoveries require depth in multiple subdomains. Socially and economically training people in every combinatorial combination of subfields at the required depth may not be possible. I am especially interested in two problems to be resolved and do not care if an AI scientist performed the discovery. It will be humbling, but totally worth it:

- Fusion (a clean sustainable form): Without this I think we are heading in a very wrong direction, whether it is conflict or climate change does not matter. Everyone is aware of this and instinctively afraid of the implied loss of quality+quantity of life.

- Cure for Cancer: It is a world wonder even in Civ. I and for good reason. As a father of a teenager, every time I hear a story of someone losing a parent/child I cringe. We have to accept this as a reality of life until a proper/generic cure is found that eliminates the most common offenders.

I am skeptical that we will have AGI anytime soon and I think the social aspects will help balance the technical developments even it becomes a reality (Three laws, A Butlerian uprising, you name it).

Chess bots can beat grandmasters, but I have a friend who takes his son to tournaments. Humans are still playing chess, kids in the same tournament with grand masters. We have to have faith in the humanity, or all else will not matter.

And I will definitely keep playing Factorio even if AGI comes to pass ;-)

The most successful applications like coding are not the result of pure LLM/generative modeling. They come from closing the loop with an agentic harness. The generate-test-selectively refine loop is the core modality of scientific work. An LLM + RL with Verifiable Rewards + feedback from compiler/terminal runs mimics this process to a great extend.

This is Fisher/Box feedback loop (https://www-sop.inria.fr/members/Ian.Jermyn/philosophy/writi...) implemented on a modern computational system. LLM is just a component. I wish Sutton had commented on this fuller picture of what we have now instead of commenting just on the LLM/Backprop side of things. I am honestly curious of whether such a loop can at least partially automate discovery.

There are more elements to discovery though. It is still not clear where the initial working model/hypothesis comes from or how the updates are selected (unless it is just parameter induction). I recently read about Hanson's Patterns of Discovery which aims in that direction. I have still not read it, but I am curious if it has any mechanistic clues.

I was about to post your last point / quote. Going multigpu is relatively not so though but once you go multi-node you have distributed storage/io/compute system which is highly non trivial. Add that the long training times now you have robustness/fault-tolerantness concerns with hardware failures and restarts. Today’s training systems are engineering marvels.

Not understanding the whole does not completely remove an ability to analyze. An interesting direction is individuality and having a notion of self. It is difficult to demarcate the individual for a model given how much the system prompt and the fine tunes / distills affect the behavior. So with computational intelligence in its current form either we can not talk of an individual or we can have a nearly infinite set of individuals corresponding to variations of the context window including the system prompt. So I do not think it can have the same kind of consciousness as biological embodied individuals. It might have something else or maybe embodied robots will one day have a similar consciousness in a similar sense to the one we think we have.

I think the complexity issue in science and engineering has also been growing for some time beyond what can be analyzed/designed by a person or a group with conventional software and math. Wolfram argues that some processes are so complex, only a computational method can solve them. If that is the case, AI might be the only path to help us in designing and discovering novel tech / science. It might be the bicycle for the mind that Jobs envisioned.

The cutoff date is early 2025 so make sure to enable web search when experimenting. I was expecting something more recent, took a while to notice this.

Thanks, that sounds like a good direction to try.

Most of your analysis I can easily relate to except “There is evidence that the Chinese models are falling further behind, not gaining.” Where is that evidence? Deepseekv4 claims to be trailing front runners by six months. I read people agreeing with this. I watched Eric Schmidt to recently make similar comments. Is he just scaremongering? Why do you claim they are falling behind?

I watched the last one S5:E17 What jobs are AI jobs and I think it gives the right framing to think about this. It is not prescriptive, it does not give a list which is smart. The job title might be the same but the actual role might have different context so the best is to have the right frame to explore your particular situation.

I am exploring ways to document the design for the agent to read and update. What makes it difficult is the lack of structure. Spec writing is not my core skill. Schemas and APIs are easier, there are declarative ways to document them. Runtime concepts and workflows have less structure and writing prose seems so unstructured for my taste. Formal languages are too rigid. But I could not find a better way.

I totally agree. I loved coding because of its closed feedback loop. Since last November, I also delegated it mostly to agents. Now I concentrate more on the design part, which is not the same. However, you move with the times and hope something else will become exciting. I do not know a more worthwhile and satisfying way than computing to spend my work hours.