qz_kb

This is not "AI", it's non-linear optimization...

DeepSeek R1 just uses crappy PPO ("GRPO" is just using the sharpe ratio as a heuristic approximation to a value function) on top of distilled existing models, with tons of pipelining optimizations manually engineered in. I don't see this making leading edge research any less expensive, you won't get a "smarter" model - just a model that has a higher probability of giving an answer it could already give. If you want to try and do something interesting with the architecture the pipelining optimizations now slow down your iteration capability heavily.

The RL techniques present will only work in domains where you can guarantee an answer is right (multiple choice questions, math, etc.). It doesn't really present any convincing leap forward in terms of advancing the capability of LLMs, just a strategy for compute efficient distillation of what we know already works. The fact this shitty PPO proxy works at all is a testament to the fact that DeepSeek is bootstrapping its capability heavily off of the output of existing larger models which are much more expensive to train. What DeepSeek R1 proves is you can distill a ChatGPT et al. into a smaller model and hack certain benchmarks with RL.

If you could just do RL to predict the best next word in general this would have been done already - but the signal to noise ratio on exploration would be so bad you'd never get anything besides infinite monkeys at a typewriter. It's not a novel/complicated idea to anyone familiar with RL to try and improve probability of things you like, and whoever decided to do RLHF on an LLM surely thought of (and did) regular RL first - and found it didn't work very well with whatever pretrained model and rewards they had. it was like two weeks ago people were going crazy about O3 doing arc-agi by running the exact same kind of traces R1 is doing in "GRPO" at test time rather than train time. Doing this also isn't novel and also only helps on shitty toy problems where you can get a number to tell you good vs bad.

There is no mechanism to compute rewards for general purpose language tasks - and in fact I think people will come to see the gains in math/coding benchmark problems come at a real cost to other capabilities of models which are harder to quantify and impossible to generically assign rewards to at internet scale.

To explore the frontier of capability you will still need a massive amount of compute, in fact even more to do RL than you would need to do standard next token prediction - even if the LLM might have fewer paramters. You also can't afford to do all the optimizations as you try many different complex architectures.

now add error handling.

I'm not suggesting you build a a whole build system with python (which is basically bazel and it seems to be good enough for google.)

A lot of originally little automation/dev scripts bloat into more complicated things as edge cases are bolted on and bash scripts become abominations in these cases almost immediately.

This is why you force yourself to use nearly zero dependencies. The standard library sys, os, subprocess, and argparse modules should be all you need to do all the fancy stuff you might try with bash, and have extremely high compatibility with any python3.x install.

Using python and constraining yourself to only use a basic subset of the standard library modules so you can run the script in pretty much any environment is almost always a better choice than trying to write even one loop, if statement, or argument parser in a bash script.

bash script is "okay" I guess if your "script" is just a series of commands with no control flow.

How the hell does the Seek by iNaturalist app work so well and also be small/performant enough to the job completely offline on a phone? You should really try it out for IDing animals and plants if you haven't, it's like a real life pokedex. Have they released any information (e.g. a whitepaper?) about how the model works or how it was trained? The ability to classify things incrementally and phylogenetically makes it helpful to narrow down your own search even when it doesn't know the exact species. I've been surprised by it even IDing the insects that made specific galls on random leaves or plants.

Regression to the mean

Just leaving a comment to say this is amazing.

This is usually done with shaders and a circle of buffers which maintain state.

No one ever considers the equally likely scenario of a technological plateau instead of singularity. Complexity/Entropy always forces things to level off. There's an plausible scenario that GPT# "replaces" all knowledge work, but cannot move anything forward. All humans become comfortable and the skills/knowledge/tools required to improve anything are lost to time as systems producing capable humans erode and we gain an overreliance on GPT# to solve every knowledge problem, but the knowledge problems that both we and GPT# care to solve plateau because were all synchronized to the same crystallized state of the world that the final GPT# model was trained on and "cares" about.

Maybe at some point maybe we only act as meat-robots which shovel coal into the machine, but a lack of redundancy in GPT# due to it's own human like blind spots means it shuts down. Humans can no longer get it running again because they can't query it properly to help fix the complicated problems. The ability to even do the tasks or design systems required to keep modern world robust to unknown future disasters or breakdowns does not and will not exist in any of the training data. If we get rid of all knowledge work, we can no longer bootstrap things back to a working state should everything go wrong.

Maybe the current instantiation of GPT#/SD etc. pollute the training data with plausible but subtly flawed software, text, images etc. halting improvement around here. Maybe the ability to evaluate if the model improved becomes more noise than signal because it gets too vague what improvement even means. RLHF will already have this problem, as 100 people will have a 100 slightly different biases about what constitutes the "best" next token.

No matter how hard it tries, I think we can say GPT will not solve NP-Hard problems magically, it will not somehow find global optima in non-linear optimizations, It will not break the laws of physics, It will not make inherently serial problems embarrassingly parallel. It will probably not be more energy efficient at attempting to solving these problems, maybe just faster at setting up systems to try solving them.

Another trap, as it becomes more human like in its reasoning and problem solving capabilities, it starts to gain the same blind spots as us too, and also gains stochastic behavior which may cause it to argue with other instances of itself. I'm not convinced an AGI innovates at an unfathomable rate or even supersedes humans in all contexts. I'm especially not convinced a world filled with AGIs that is indistinguishable from a very intelligent human or corporation or what have you through imitation does any better at anything than the 9 billion embodied AGI agents that currently populate the earth.

hybrid seeds don't breed true. You need to research what hybrid means, I have no idea why you bring up "organic markets" as if the market/growing conditions affect the genetics of a plant??

What kind of simple features are you writing that 50 lines seems like enough to do anything useful??