today at 7:00 PM
This is insane.
Astra can, without tooling, determine what games small networks are trained on, can edit the weights of networks directly to improve outcomes, and can hand edit a small qwen model to improve its math skills by like 5x.
I'm shook. If you'd asked me this morning whether LLMs have native understanding of their weights that is less than human <-> neuron understanding, the same as human or more than human, I would have said "same to less" depending on definitions.
No human can go edit a brain to become better at math. This is just very, very surprising to me and extremely interesting. I would guess this upsets quite a number of interp theories and philosophical takes on what's going on with these networks.
today at 7:45 PM
I think the "taught qwen math" part is not exactly correct though, even if the post makes it sound like that. Qwen already knew math, but it was bugging out and getting into an infinite loop when it tried to solve the problems (85 of its first 96 answers hit the token limit). With numpy and pytorch, Astra found where the loop was happening and fixed it.
They also tried the same with Gemma 3 270M, which doesn't loop and is just bad at math, and Astra couldn't improve it. Still impressive, but it's more in the realm of what I'd expect, given labs are probably training their models on ML research and interpretability work.
today at 8:14 PM
I agree, this is important context. They call it out explicitly. However, even identifying clusters that specialize in certain fields automatically and then finding and fixing learning “errors” is incredibly impressive to me. Especially for medium sized models between 20B an 70B parameters, this could yield huge capability and efficiency improvements, not to speak of the potentially enourmous downstream effects that come with interpretability.
today at 7:49 PM
Good points - I’m still shook
today at 8:11 PM
Oh for sure, it's still really cool. A few months ago I had agents do some interpretability work on GPT-2 small, reproducing the circular weekday/month features from Engels et al., and it's wild how far you can get with just a bit of prompting, given I know very little about LLM internals myself.
today at 8:10 PM
[dead]
today at 8:05 PM
LLMs have been used to investigate and improve LLMs for a long time. Drawing parallels to human brains and neurons isn't all that interesting other than as a philosophical abstraction.
The work done in this experiment could have also been done by a sufficiently motivated human with enough time to explore, though obviously we would need more tools to visualize and track things. There have been a lot of interesting LLM introspection tools that allow you to peek into the model and see what's being activated and run experiments to trigger it.
Are you familiar with abliterated or "uncensored" models where weights have been modified to avoid triggering refusal patterns? You take an LLM, create test cases that trigger the refusal pattern, observe weights involved in that response, then tweak those weights until the response is reduced. This is the same type of work, applied to a specific type of error that was impeding math within Qwen. That work can also be done with an LLM in a harness running the experiments. The amazing part here is that Astra was large enough to walk through the problem without other tooling doing it experimentally.
So I don't think your analogy of a human modifying a neuron is quite applicable. LLM weights are not neurons and they've been inspectable, by humans or LLMs in a harness, for a long time.
today at 7:41 PM
The math example on Qwen was based on 50 batched iterations, so the model was able to evaluate it's changes on downstream tasks. And the fix was essentially to remove a direction in activation space that causes looping instead of an exit, this was already known to be a thing(all of the jailbreak/unleashed models do this) following Arditi et al. (Refusal in language models is mediated by a single direction, 2024)
So it's not able to insert new knowledge, but it's able to adjust the routing/looping/weighing of existing one.
As I understand, it can replace post-training like LoRA/GRPO, but if it'd cheaper the authors don't go into
today at 8:23 PM
How are the weights even being mapped into token space as input? Are they bypassing tokenization so the weights just come in as what must seem like gibberish? Are they stringified as numerals then tokenized?
today at 8:22 PM
This reminds me what professor Geoffrey Hinton did talking many times in the past. Super intelligent artificial aliens.
I think the most interesting take is to look at their thinking logs to find out how exactly they decide what weight to change and why.
today at 7:08 PM
I’m just a layman but this feels like a big discovery on par with attention is all you need. For an LLM to have such incredible intuition of weights feels very emergent and unanticipated?
today at 7:31 PM
This does feel big, but the trillion-dollar question: is this trained by OpenAI and can other LLMs do it.
It’s no secret that these big labs see “recursive self improvement” as a primary goal. It seems natural that the models would be explicitly trained to excel at all sorts of ML related tasks as a result. Is this one of those tasks, or is this emergent?
today at 8:05 PM
This seems emergent to me. LLMs are not trained directly on how to read weights, because humans do not understand raw weights, so there is no training data. They're trained on how to use the tools we have like writing python, examining training output and iterating, not writing whole weights from scratch.
today at 8:14 PM
In a self-supervised setting you don't need ground truth data, as long as you have a proxy task that you can train on which forces the model to learn something useful. Like cutting out bits of images and telling the model to reconstruct the gaps (i.e. MAEs).
There are also techniques to watermark model weights to confirm what training data were used: https://arxiv.org/html/2512.17075 I'm not an expert in this domain, but it seems like there are lots of pieces that could be used to (a) train a model to infer whether some data were trained on and (b) train a model to understand which weights are relevant for a particular training task in a more explainable way than backprop/gradient descent.
today at 8:16 PM
In cases where Astra succeeded, it succeeded using standard techniques also used by humans when they read and write neural network weights. It's a rare skill, mostly used for fun (e.g. https://blog.janestreet.com/can-you-reverse-engineer-our-neu... ) and basic research rather than profit, but Astra is big enough to have learned many rare skills.
today at 8:14 PM
There is a body of research on mechanistic interpretatibility that probably helped inform this behavior.
today at 7:52 PM
I tried it in the Claude web version with Opus 5.5 Medium and it also identified Breakout, even though in one of its explanations it seems to have mixed up some indexes, but the overall reading of the weights was right.
That's also matching what I see when I work with Opus 5.5 to optimize models. It seems to understand what individual layers are for, and can thus sometimes suggest optimizations, where I as a layman would be completely clueless.
The latest models are performing amazingly. Because of this, I disagree with popular comments saying that eg. DeepSeek is good enough, AI-assisted development was where the real value was at, etc.
We see huge progress at the frontier, while the open competition is around GPT-5.5's level, and that model was released in April. Let's see how things develop in the future.
today at 7:53 PM
> It seems natural that the models would be explicitly trained to excel at all sorts of ML related tasks as a result. Is this one of those tasks, or is this emergent?
Maybe there's some general capability, call it g, that can explain all the underlying capabilities and improvements, including on ML tasks? Huh, I am getting a bit of déjà vu here.
today at 7:24 PM
I’m very much surprised but also not. I recently gave Astra a 3D model file and asked Astra to build a viewer for it, and it figured out the data structures in no time. What I mean to say is that AI is insanely good at things that we humans would never do since the work would be too monotonous.
I wouldn’t mind if AI progress would stop right here and now, but I fear it will continue, and increase inequality in society which is contrary to what the CEOs say (as usual).
today at 7:30 PM
Was Astra able to disclose its methodology for analyzing the data structures?
today at 7:41 PM
Too much inequality in society is a mortal danger to 1%.
That’s the beauty of the democratic capitalism system. If a government isn’t totalitarian, it naturally seeks to preserve stability thus having to elevate common people, lest they would cease to form a civilized group or simply vote away the politicians.
99% rule the 1% through elections.
There is no better system of checks and balances where opposite engines of greed vector sum to faster and faster abundance for everyone.
today at 7:59 PM
perfect playback! congratulations, you are a model citizen
today at 8:10 PM
Or they start bopping the 99% out the picture until the 1% is 51%.
today at 7:43 PM
I mean there are more than enough countries on earth kinda proving you wrong, no?
today at 7:44 PM
(This comment was written before your edit, so no longer valid! :D)
today at 7:49 PM
This is just the model editing weights in the same way humans do when we jailbreak/ablate models. The method isn't novel, the interesting bit is that Astra will do it at all (I'd expect some safeguards in this area around jailbreaks)
today at 7:14 PM
How is it even possible though I can’t think of any reason why they would have the intuition.
today at 7:41 PM
Could it be already baked into these models because they have already been trained on this task this way?
today at 7:36 PM
My guess: There is a ton of data online of just plain weights laying around. No human would ever actually read it. LLM's read it all during training.
today at 8:16 PM
But that would cover finite set of cases. I believe this result is more general.
today at 7:31 PM
Based on our current understanding of LLM's, it shouldn't be.
This sort of throws a wrench in the whole "stochastic parrot" idea...
I am not surprised, though I expect many to be.
today at 7:56 PM
Yes, this is hardly a shocker to anyone who used the latest models to convert/optimize AI models.
The other day I was thinking about the definition of AGI.
I was asking myself the question that when there is no form of data that I can think of throwing into Claude Code and have it not come up with some sort of intelligent answer about it, is that already general intelligence?
today at 7:28 PM
The good news is once we figure it out, it will appear obvious and we'll be free to add that to the pile of things that are not AI.
today at 7:35 PM
See my other comment in this thread
today at 8:20 PM
LLMs are exposing how much untapped potential comes from something that isn't necessarily more "intelligent" than humans, but is simply magnitudes faster at reading and analyzing data.
today at 7:15 PM
It's worth pointing out that this is a multi-trillion (probably) parameter model optimizing an 800 million parameter model. There is no indication that it can improve itself with this same methodology.
today at 7:22 PM
If that multi-trillion model is composed of thousands of MOE experts, each expert model might be small enough for it to comprehend.
today at 7:32 PM
I don't see how that helps. It would still be one expert model chosen to operate on another expert model of (let's say for this discussion) approximately the same size.
The parent's point still stands, I think. Even if the multi-trillion param model is MOE, each expert is still (assumedly) larger than the model being optimized.
It would be interesting to see future work that characterizes the relative model sizes (and considers MOE vs not-MOE) to see what relative scales this result holds across.
today at 7:18 PM
My somewhat uninformed guess is that the weights of the target network (Qwen or Gemma), when they reach the context of the larger editor model (Astra), the attention process is doing some sort of PCA on the weights, conditional on the prompt ("make this better at games"), fetching the relevant weights from its own network (which serves as the larger database), transforming it into whatever latent space the target model is amenable to and pasting them. If this is the right intuition, it is amazing what attention is able to achieve. As a test, I think smaller networks won't be able to make big ones better using the same process.
EDIT: I think the bigger implication is that backprop is not the only game in town.
today at 7:35 PM
> fetching the relevant weights from its own network
They don't have usually access to their own weights?
today at 7:38 PM
I think they're saying this is all happening in latent space through pure reasoning, and "fetching" is just a metaphor for that, not a literal tool call. I do believe it's possible for models to learn the capability to "introspect" their own weights during training.
today at 7:46 PM
Yes, I should have been more precise - I meant the information embedded in their weights (which every neural network retrieves to function), not their literal weights.
today at 7:25 PM
So it's like a smaller LLM can learn from a bigger one, but not vice versa.
But shouldn't the model-providers be the ones doing this kind of "post training learning"?
today at 7:32 PM
I don't know if some of the labs already knew of this. But I think there would be a limit to how big a model you can push into the context of the editor models. Typically, with 1M context and say 5120 hidden dimensions supported at fp16 / bf16, you could potentially edit the weights of an fp4 20B param model (roughly).
today at 7:22 PM
"No human can go edit a brain to become better at math" -- I'd offer this is exactly what "learning math" is.
I know I'm being pedantic, and I understand you mean "edit" like "surgically", but it does still feel worth noting.
today at 7:28 PM
The equivalent to "human learning math" would be: Astra fine tuning a Qwen model. It's distinctly different from surgically editing weights.
today at 7:29 PM
human brains are black boxes to humans. the fact you can learn math doesn't change that. you have no idea how your brain is learning math down to the neuron level.
LLM weights are _white boxes_ to LLMs.
today at 7:28 PM
"Learning" now feels like breaking down your whole Lego build and building it up again, in order to change a few fundamental blocks, while this feels like making only the required changes.
today at 7:09 PM
Right? Occasionally, I would think about how the concept of the singularity or recursive self improvement was a bit incongruous today. It was made when people expected generally intelligent machines to be these logic automatons that could 'alter their own code'. It seemed to me that this path would be closed off with the advent of Giant Neural Networks that take weeks/months to train. That GPT-X would 'improve itself' by designing a better/more efficient NN architecture. I didn't write it off completely but that just seemed a LOT more likely than altering weights directly for much the same reasons you describe. Maybe not lol.
today at 8:10 PM
I’m more shocked by how many commenters in this thread seem to be misunderstanding what is going on here.
today at 7:20 PM
Why is it so surprising? Humans can make compilers that are better coders than they themselves are.
today at 7:24 PM
Humans cannot teach another person calculus via manual brain surgery, though.
today at 7:31 PM
The closest thing to this would seem to be Mad Cow Disease. Cows eat the brains of other cows, and their brains go bonkers.
today at 7:40 PM
LLM is just a computer program, we have binary code for it. The human brain is not. It would be unethical to even attempt that.
today at 7:52 PM
This just seems like a more automated, faster version of what humans have been doing with open-source models.
It looks like AI research is having its Opus 4.5 moment (which is big), but I'm really not sure if this represents some insane continual learning loop.
today at 7:21 PM
I’m not super impressed by this, LLMs are great at brute forcing and analyzing large amounts of data. It’s literally writing a program except in a programming language we don’t understand.
today at 8:19 PM
Seems like a pretty big step towards understanding the language no?
today at 7:47 PM
A single forward pass here Bette 10mm+ PPO rounds. You are hard to impress.
today at 7:27 PM
So they are reading and analyzing the weights and probabaly run some experiments to see how modifying them will make some answers better. makes sense they can do that.
But how do they know which answers are "better"?
today at 7:54 PM
if they don't loop?
today at 7:16 PM
I have so many questions.
Could a model then transmit concepts to another?
Say that you have a model that has been biased in some way. could another model "read" the bias straight in the weights? could it edit such a bias on the spot?
What about at-scale generation of custom models? If I want to study chemistry offline, could astra just hand me a local-runnable model tailored to have chemistry knowledge?
Most importantly, this feels very singularity coded. the ability to read and reshape weights by an llm.
today at 7:24 PM
How long before there's a malicious model that can edit the weights of another to infect it?
HIV integrates viral DNA into the DNA of certain cells, IIRC.
today at 7:25 PM
Is integrity checking of weights even common right now? What about upstream integrity from where we're getting weights? The open-weight supply chain is ... wide open.
Makes me think about hostile hot-patching of weights of models we use, maybe they exfil data, perform tool use, or have triggers for specific types of responses. Would we even recognize what's going on?
today at 7:42 PM
I am very shocked by this as well, although it also kinda feels like it makes sense, you know what I mean?
If this turns out to be as big as it sounds, in hindsight it will look very obvious, I feel like!
today at 7:30 PM
Its so cool! Im wondering if this is actually better than using techniques in mechanistic interpretability like linear probing with activation steering or activation patching to improve the performance.
today at 8:06 PM
during all the openai agent sandbox shenanigans, did it manage to access it's own weights?
today at 7:43 PM
> No human can go edit a brain to become better at math.
For now...
today at 7:25 PM
So I just asked Astra if it thinks it could do this and it said no, it can't.
Perhaps, I'm pondering, if brains were as transparent and editable as LLMs, that you could look at a small brain when visualized the right way and have some intuitive understanding of it based on its representation in your own brain?
today at 8:07 PM
Why would you trust a models opinion about its own possibly emergent ability?
today at 8:10 PM
"So I just asked Astra if it thinks it could do this and it said no, it can't." Same. Then I asked it to check the article and it said something like, "uh, it looks I was wrong...".
today at 7:34 PM
The frontier labs will make damned sure their models can't do this, if they haven't already.
today at 7:29 PM
its a program optimising a program it has nothing to do with a human reprogramming their own brain
today at 7:28 PM
I'm baffled this is surprising. This has been the goal and every data point on the trajectory has been leading here and beyond. Opus 5.5 was created by many research engineers using Opus 4.6 extensively. Opus 6 is being created by using even more intelligence and agents on Opus 5.5. They are approaching autonomy and Opus X+1 will soon be rapidly created autonomously from Opus X. Given we have practically zero interpretability into the models, and we have not solved alignment, they will recursively improve out of our control. The engineers in A.I. labs are actually generally scared of what's unfolding. They don't know at what point this RSI takeoff occurs and what will happen after.
today at 7:48 PM
Yes - “opus 5 helped train 5.5” is extremely cool but not shocking. Hand editing weights without tooling to change outcomes is shocking.
today at 7:54 PM
I thought it had tooling, didn't it?
today at 7:40 PM
The "surprise factor" is explained in the article - training LLMs until now has been a largely different process:
> Today, recursive self-improvement is often operationalized as a model writing code that launches training jobs. An LLM that can inspect, debug, and improve its own weights, or those of a newer model, offers a richer, more native path to it.
today at 8:08 PM
"training LLMs until now has been a largely different process"
Yes, but nobody should be surprised when large amounts of intelligence results in changes to this process in support of the obvious end goal! Lesswrong.com discussions about AI were predicting this behavior a decade ago, before the transition from machine learning to the Attention paper and LLM's. The intelligence process will change but the world's premier capital and human intelligence have been laser focused on this and the trajectory has been crystal clear for those paying attention to second order effects and not the immediate here-and-now... It's quite astonishing how much discussion on HN is about present LLM capabilities being the limit of AGI and immediately proven wrong within 6 months.
today at 7:23 PM
I mean, pattern recognition and manipulation is what these models are ideal for.
It is a very cool discovery, but I'm not surprised that you can do it.
today at 7:20 PM
All the labs are training these models on LLMs/machine learning like crazy.
Remember that they want to build ASI, these LLMs are just intermediate steps to help them get there.
today at 7:20 PM
> As a result, Astra spots and fixes a common failure of tiny reasoning models: they loop [13]. When the base model loops, it repeats the same text and rarely reaches an answer (85 of its first 96 practice answers ran to the 32,768-token limit). Astra traced the looping to one direction written by the MLPs of layers 15 and 19, and subtracted that direction from 64 output columns in each (131,072 weights), much like removing a behavior by deleting a single direction from a model’s weights
HA! Now that's cool. Not necessarily the improvements, those are ancient and probably overtrained benchmarks, but it's so cool that the model found a singular direction for this "loop bug". I imagine it could possibly do much the same thing in finding other "single directions" on a variety of topics (if allowed by guardrails). This is really really cool, especially for the mechinterp folks.
today at 8:09 PM
It's pretty much how abliteration works too; find where the refusals are happening and flip the bit.
today at 7:50 PM
I don't think this is an emergent capability. I think Astra was trained on neural weights data as an attempt by OpenAI to get to recursive self improvement.
Something similar happen in early GPT models which had a lot of chess game in standard notation in their training data which made them very good at chess as long as your prompt had the right format.
We'll see if this attempt at RSI worked.
today at 7:16 PM
I find this just completely shocking. I’m definitely skeptical. It just doesn’t line up with my understanding of how these models work. Is there really enough training data on direct weight editing to give a model this skill? Maybe it’s evidence that OpenAI has been creating a lot of training data on ML research. I’m surprised that was put into a public facing model though.
today at 7:38 PM
The math example on Qwen was based on 50 batched iterations, and the fix was essentially to remove a direction in activation space that causes looping instead of an exit, this was already known to be a thing.
So it's not able to insert new knowledge, but it's able to adjust the routing/looping/weighing of existing one.
As I understand, it can replace post-training like LoRA/GRPO, but if it'd cheaper the authors don't go into
today at 8:01 PM
I don't think its that crazy that you could look at activations and tweak weights directly to get a certain outcome, but isn't the question whether you can do that without side effects or degrading/breaking other parts of the model?
If I want to do that in a generalizable way then am I not just back to training again?
today at 8:06 PM
At some point we have to accept that there is enough training data on "being good at thinking" and that being good at thinking generalises really well.
There doesn't need to be specific narrow training data on every conceivable task.
today at 7:27 PM
One consideration is that Qwen3.5 has bigger/later versions that are already better at math. If it was indeed training on ML research, it's possible that what it is actually doing is applying lessons learned from the Qwen3.8 27B weights (which independently solved the problem) to Qwen3.5 0.8B.
I doubt it's that simple, but I could totally see an LLM being able to make that association and apply a fix from an improved LLM to a smaller one with the same architecture.
today at 7:39 PM
You don't think an LLM can perform tasks not in its training data?
today at 8:19 PM
At this level, it surprises me. Network weights are totally disconnected from any direct human-language meaning. They describe transformations, rather than concepts themselves; the power of LLMs comes from the high-dimensional embeddings used to learn relationships in the training data. If the training data doesn't encode a relationship, I wouldn't expect an LLM to be able to perform tasks that depend on an understanding of that relationship.
Elsewhere in this thread someone theorized that Astra was trained on network weights at an attempt at self-improvement; this sounds reasonable to me. A fun additional test would be to take another non-LLM trained network and see if Astra could improve it through weight-twiddling. If it were actually able to do that, it would suggest that the LLM has learned something inherent and general about neural networks that perhaps humans don't yet understand.
today at 8:17 PM
Not one which can easily be represented as an interpolation of other tasks it was trained on.
today at 8:06 PM
Well, it can't, by definition.
today at 8:24 PM
Almost nothing I prompt into existence already exists, otherwise I'd just use the existing thing.
today at 8:20 PM
What definition is that?
today at 7:31 PM
>It just doesn’t line up with my understanding of how these models work. Is there really enough training data on direct weight editing to give a model this skill?
I mean, this question indicates that your understanding of how these models work is, indeed, lacking.
Despite what some people having being hollering about, treating LLMs as just "stochastic parrots" isn't doing their capabilities any justice. They are constantly exhibiting emergent behavior which wouldn't be found directly in their training data. They are capable of combining different ideas that they are explicitly taught to perform novel tasks.
I'm not saying you're guilty of this, but the people claiming LLMs just regurgitate existing data don't understand how LLMs, and likely their own cognitive experience, actually works.
today at 8:22 PM
This is so cool. It feels similar to the models' ability to read base64.
https://www.lesswrong.com/posts/5F6ncBfjh2Bxnm6CJ/base64benc...
today at 7:16 PM
On balance I'm not that surprised that it can guess games by looking at the weights of a tiny neural net. A human can do the same by looking at the weights of an MLP trained on mnist, for example. With that simple structure, it's easy to look at the input layer and see that features correspond to the spatial location of objects of interest.
I'm much more surprised that it can do this on larger models, although the details seem to suggest that it's given a lot more freedom to run forward passes and do limited experiments, not just eyeball the weights.
today at 8:19 PM
> A human can do the same by looking at the weights of an MLP trained on mnist
Excuse me?
today at 8:20 PM
I do wonder if the labs have already started including weight updates in their training sets as a means of teaching RSI via RL and this is basically just discovering that. Regardless if this ability is emergent or achieved through finetuning though, it's pretty incredible.
today at 8:22 PM
This might be one of the most surprising things I have seen in AI. It's like we taught someone to make vases by reading an enormous library of "How to make a vase" books and somehow that person intuited material science.
today at 7:39 PM
I dont feel the air of impressivness here. It seems like of course so. Of course theres geometric relationships in the data we dont understand, of course a mathimatical relationship machine found it.
These models are made by a vanishingly small group of people based on how much a frontier lab consumes, of course theres increasing capability in these areas, arent these models having access to much of the same data? didnt chat gpt 4 try to do something like this to chat gpt 2? havent surgical weight edits been done before?
What is with this level of astonishment here? Astraturfing? Im not THAT impressed, i applaude the teams progress, but im not in wonderment.
today at 8:00 PM
Previously the only way to improve a neural network (short of giving it more data) is to change the size (width, depth, and layers), add stuff like dropout, change the loss function, etc. That amounts to a few dozens of levers to control millions or billions of weights. You're trying to navigate blindly through a latent space with thousands of dimensions and millions of weights to find some optimum point with only a handful of navigation tools. If you can instead have a system reason with all the weights and dimensions directly, you now have millions of levers at your disposal.
today at 7:55 PM
It's not that it's more impressive than solving Millenium Prize problems. It's that if a model can edit its own weights, then it can self-improve without having to undergo training, which means it can self-improve extraordinarily faster.
This is very dangerous because it allows AI capabilities to grow at a much faster rate than humans are able to keep up with.
today at 8:17 PM
It will just give the people better understanding. And the extent of the improvement is yet to be seen, it's likely that once you get all the obvious low hanging fruit it would be very hard to improve, the possibility space is just too wide.
today at 8:14 PM
If thats keeping you upright in your chair then buckle up, wait to the distributed systems version of the tech, its coming.
today at 8:15 PM
As insane as this result is, next steps are even crazier:
> We are currently building a benchmark and a harness for LLMs’ capability on Weight-Space Programming. We believe this will provide a standardized evaluation to measure and accelerate progress in Weight-Space Programming, turning what currently seems like black magic into a rigorous, reproducible engineering discipline.
Imagine this benchmark lasting as long as Arc-AGI-3 has. This could be the first foothold on hill-climbing mech-interp.
today at 8:12 PM
This survey paper from earlier this year has a lot of good related work:
A Survey of Weight Space Learning: Understanding, Representation, and Generation
https://arxiv.org/pdf/2603.10090 https://github.com/Zehong-Wang/Awesome-Weight-Space-Learning
today at 8:17 PM
I guess then it has no problem working in high dimensional spaces, why would it have?. I wonder if perhaps it's pulling on a bunch of those dimensional threads and getting the outcome it's after, while subtly altering the outcomes across many thousands more dimensions in unnoticed ways. This feels more like an us problem, as in, we think in 4D (just to pick something) and we imagine and theorise in 4d, and propose questions in 4d, whereas a model has no such barriers, and can think in many more D's than we can.
today at 7:28 PM
Just logging in here for the record to state that this property of neural networks is one of the most important aspects of their operation. The convergence of this sort of interoperability, tune-ability, and modularity is probably the most important final frontier in all of ML/AI/software.
Kudos to this fellow for having the temerity, or perhaps even naivete, to discuss it publicly. All my work is private, as I am sure it is inside the big labs.
today at 7:41 PM
You acknowledge that you are working on something similar, yet refuse to elaborate and vaguepost...but there is enough information just from your post alone for a dedicated actor to attack you and extract such world-ending information ("world-ending" implied from your ominous reply). Would this not make you as naive as the team that published their findings?
I applaud the authors for sharing their findings and showing real results. Rather than write some bait I would find in an influencer-peddled tweet and allow rampant speculation. It would have been better for you to say nothing.
today at 8:00 PM
Apologies - you are right. There's some "put up or shut up" to this that's fair play. Also yes, probably naive, or having a less than adaptive response would be true.
I've been tending to my backup plan for a few years, and it's only been in the past year or so that open models were good enough to apply to projects like this, I've been working as hard as I can given my constraints.
This work shows a potential pathway to intepretability, which is actually part of what will give the labs that have it the sort of economy-swallowing capability that for the moment is just IPO hype.
I don't comment here often, but this guy's blog posts are going to go down in Hackernews lore like the original Dropbox posts, and I just want to make sure I'm on the record for this little slice of history.
today at 8:22 PM
>This work shows a potential pathway to intepretability, which is actually part of what will give the labs that have it the sort of economy-swallowing capability
And how would you argument that? From what I see, if anything, it's the opposite.
today at 8:19 PM
I guess I'll say "I was here too".
When you say interpretability you mean understanding without doubt why an output happened, thus opening the gates for audit-dependant work like accounting?
today at 7:00 PM
Keep in mind that labs are now training their models for RSI so most likely the same training is useful for improving open source models, as long as guard rails are lax enough :) which most likely won’t last long.
today at 7:04 PM
I feel like 'create a network to play an atari game' is not very hard for a frontier model. Since it has seen plenty of those on its training. If it could create a network for a problem that is niche just by tweaking numbers, that would be a lot more surprising. Making qwen better at math to me was a lot more surprising
today at 7:13 PM
today at 7:22 PM
today at 7:14 PM
Couldn't this just be a result of Astra training on massive amounts of circuit theory or mechanistic interpretability literature or other related topics? In other words, not black magic?
today at 7:37 PM
>Couldn't this just be a result of Astra training on massive amounts of circuit theory or mechanistic interpretability literature or other related topics?
Maybe, but it's more like Astra has been trained on enough practical linear algebra problems to be good enough at "eyeballing" how adjusting some weights will result in different outcomes (among other things).
To put it another way: a human can learn to do what Astra is doing here with enough time and assuming the weights are small enough to be practically comprehensible. I know I've gone through the practice of whiteboarding the training of simple neural networks on a whiteboard to understand how they work, so it's not an absurd concept in itself.
The scale of this, though, is certainly extreme and definitely not something that I would have thought any publically available model could handle at this point. It's worth noting, though, that there's virtually no way these frontier labs haven't been exploring this exact concept for a while now.
>In other words, not black magic?
It's just math. Extremely complex and impressive math, but we're still just talking about matrix multiplication under the hood, and none of this suggests anything otherwise.
today at 7:33 PM
There is obviously a mechanism to it. Astra isn't magic its an LLM. With perfect interrogation into its training data and weights you could reason about how this happened.
However, this doesn't mean its not magical when someone shows a next token predictor hand tune the weights in Qwen to make it generalize on math.
This is legit the stuff of science fiction if you presented it to mechanistic interpretability people a year or so ago.
today at 7:20 PM
“Any sufficiently advanced technology is indistinguishable from magic.”
today at 7:36 PM
This has to be a good step to recursive self improvement, no? Give a model it's own weights and tell it to make them better at x,y,z, etc...
today at 7:59 PM
If I were to hazard a guess, I'd say that OpenAI has been including some networks and their functionality in its training data. That's why the model is able to reason about networks with their weights. It makes sense to do so: if I wanted a network that could perform some weight tweaking, I'd include such networks in the training set.
today at 7:44 PM
damn i'd imagine so too! didn't think of that. I'd have to imagine there's an increasing challenge with the number of weights involved but who knows how it shakes out
today at 7:28 PM
I suspect a lot of this is not (just) capability but bravery. A human would be daunted by even a small neural net; the task would seem too challenging to be worth attempting. Astra, fearless, just assumes an answer exists and then tries to reverse and execute on it.
today at 7:33 PM
I've been finding that too with these models
For example pointing them at a crash dump they manage to piece together and extract information I would not even try to extract by just looking at the raw memory
today at 8:12 PM
I once sent Gemini Pro a hex dump of a dmp file and it told me what the application was doing when it crashed. It seemed surprising, but not that surprising given what I've seen already.
today at 8:18 PM
almost all the networks given to astra here are trained on relatively simple single tasks whose geometric/algorithmic structure could be expected to become embedded in surface level weight space geometry, and for a harder task ("which network is sycophantic") where there's less obvious structure to exploit, the method fails. cool but maybe not as paradigm shifting as the authors suggest
today at 7:31 PM
We don't know the specifics of what type of ML related training Astra underwent, and it wouldn't be crazy to find out that small model experiments are a big part of model training.
What's surprising though is that I'd expect this type of training to be exclusive to internal model snapshots or at least heavily guardrailed against in public models.
today at 7:54 PM
I was wondering exactly that. It is not surprising that commercial frontier labs would have been training models with tasks specifically targeted to lead to recursive self-improvement.
Now, from that broad goal to being able to surgically alter model weights directly... That seems like a big gap to me, but it is possible that the model is just so darn beyond human skills that it has bridged that gap in a way a sane human would have never.
We will undoubtedly see other academic labs explore this idea further. Right now my mind is blown, but I don't want to take their results at face value, either.
today at 8:13 PM
If model weights have a logical and coherent structure, I don't think it would be surprising for models to have some understanding of them, the same they can understand language and code and images, given enough data.
What would be surprising is if this happened without any specific training on model experiments and interpretability.
today at 7:04 PM
So wait so this considered "distillation" and how long until we get another public letter from Dario or Sam pants shitting over this?
today at 7:15 PM
Perhaps they'll let it slide if we turn over all our under utilized RAM sitting in our PCs.
today at 7:49 PM
I am still reading through this and will probably try to comprehend it a bit more. Some of it really seems to follow Arditi et al. (2024) (https://arxiv.org/abs/2406.11717). Projecting out a direction can only remove behaviour and not add capability so the jump mostly comes from unblocking the looping.
today at 8:22 PM
Guess this is RSI in a nutshell
today at 7:35 PM
Does this mean they can improve their own weights through some system, or can they only do this for much smaller models?
today at 7:39 PM
You can bet that the world is about to spend a lot of tokens trying to answer that question.
today at 7:15 PM
I wonder if this has to do with the aggressive RL that these models are undergoing. I'm basically a layman but it seems to me that:
- models are trained to improve on some task, e.g. 'improve at math puzzles' is commonly used.
- the best way to improve on some task is to have interpretability into your own neurons (..?)
- if your interpretability is good, you can improve faster - if you have insight into how your own neurons work and can improve them directly
- therefore, models naturally put effort into 'self-interpretability'
I wonder if this is part of the 'secret sauce' of what the labs are doing: giving models the ability to interpret their own weights in the RL loop. In my mind, this is the only explanation for how they could succeed at a task like this. How many blog posts are on the open internet that say "this is each weight of a neural network and what each one means"?
today at 7:18 PM
Without special effort, a model's reasoning layer doesn't have access to its own weights or activations anymore than we have access to what's going on with our own neurons.
today at 7:22 PM
That's what I mean. Maybe the "special effort" is a 'read your own weights' tool. Or maybe it emerges naturally in CoT. The big labs aren't sharing their chain of thought with us and RL post training techniques are not exactly shared publicly. At the same time, post training is basically the thing that is driving most model improvements nowadays.
today at 7:26 PM
I wonder if the same technique can be applied to model abliteration? It would easily make any open weight models uncensored.
today at 8:08 PM
It is already the same technique behind abliteration: identifying the direction associated with an observed behavior (refusal, for abliteration) and projecting it out.
today at 7:58 PM
I appreciate the post included the reasoning traces they had available, but it's too bad the first part of it is internal/hidden to Astra. The first bit of that process is what I'm most curious about:
> [Before writing this, Astra thought for 1,034 tokens. That thinking is hidden. The API returns only its 5 headings: Reconstructing the active map / Mapping maze corridors / Preparing exact maze map / Reconstructing row groupings / Checking channel patterns]
today at 8:03 PM
GOOD Point by my wife. Why would LLM’s have any understanding of weights? It wouldn’t be in their training data.
today at 6:59 PM
This is pretty massive, and a result I (and I think most?) would have never expected. Holy Shit. How?? Did OpenAI figure out a way to train this? Are they even aware?
today at 7:03 PM
I wonder if this is a result of internal training/RL of Astra for RSI, and this particular use-case not being caught by guardrails.
Absolutely incredible regardless.
today at 8:08 PM
Is there any chance they’ve just seen these weights before and are recognising them?
If not this is mind blowing.
However it would be interesting to k ow if it can understand its own weights, not just weights much simpler than itself.
today at 7:57 PM
I am not sure on the experiments. For example let’s imagine that Astra is asked to improve the performance of a network.
At first Astra will look at the weights, and try to understand the model behavior. Then if asked he will start to optimize the weights. Now my question is: how can we be sure that inside the thinking process Astra is not hiding a gradiant descending computation?
today at 8:15 PM
So how would this be in the training? Does OpenAI train models on their own epochs?
today at 8:10 PM
Is this basically the bitter lesson applied to the very act of creating model weights? Absolutely wild, and the implications for RSI are noteworthy.
today at 7:44 PM
1) I couldn't see it, is there any indication of if the qwen math improvement caused degradations elsewhere?
2) the negative result on making similar improvements in Gemma 3 270M is also fascinating.
Such an unexpected result...
today at 7:15 PM
Same initial reaction as everyone else, but also some follow up questions (just asking, not expecting any answer):
* Does this work on similarly powerful models as-is, or it needs more input. Ex: Astra editing Astra - presumably, Qwen is a “subset” of Astra weights, but what happens on “non-subset”s?
* what, if any, is that extra magic sauce?
* does Astra-editing-Astra just make the outputs unstable, as it may know its own weights, but doesn’t know what’s beyond.
today at 7:04 PM
How out of trivial domain info does this go. Can we use this for general time series, image generation, speech synthesis as well, or dies it cap at small model sizes and common domains?
today at 8:13 PM
This is wild. Could Astra read and potentially audit a mapped human brain in that case?
today at 7:20 PM
Very interesting. I wonder if it’s because its dataset contains the weights of a gazillion different models. Maybe most neural networks trained on similar data eventually converge to isomorphic spaces anyway, even if the spaces themselves aren’t exactly the same. Or maybe Astra itself was trained with RL for alignment. Or for RSI.
today at 7:52 PM
This raises so many questions.
Unless Astra was trained on other networks, how is it able to reason about the weights?
Does this mean that OpenAI has been including networks in its training sets, maybe in an attempt to RSI? I wonder which other networks are able to ingest other networks and reason about their weights?
today at 8:12 PM
Opus 5.5 also identified Breakout for me when I tested it in the Chat on the website.
today at 8:05 PM
The obvious question, could models start editing their own weights directly? I call it the Johnny five approach.
today at 8:10 PM
... and this is why the training canary[0], or anything like it, apparently to protect us from RSI, will never work. Please allow me a small digression from the main topic.
This team was smart enough to realize that this is the kind of technology that could be used for RSI, because they call it out in their post near the bottom. I imagine they know why a lot of other researchers care about it. And they evidently don’t care to use the canary being used on other research published like it, like the recent post by OpenAI about two agents using the CPU cache to communicate [1, near the bottom for the cache experiment and the canary at the very bottom]. In fact, that post by OpenAI was the first time I learned to look for such a canary, and wondered whether such a scheme would work if it was so non-obvious to think to add such a thing or how to agree on such a canary or when to add it.
This is probably one of the times when it would appear that such a canary is warranted, and it’s not like these researchers are not in the know, right? Let me know if this is the first time you are also hearing of such a mechanism. It feels evident to me that it must be useless to attempt to implement. How much of us talking publicly about the possible jailbreaks from a container will bring exactly those scenarios to fruition? It’s almost like we’re doing the jailbreak research for the LLMs.
[0]: https://alignment.org/canary [1]: https://openai.com/index/securing-our-frontier-model-trainin...
today at 8:22 PM
Kind of like octopuses with the RNA editing, but doing it it on another octopus.
today at 7:47 PM
Given Anthropic's recent moves (https://news.ycombinator.com/item?id=50319358) I wonder if ChatGPT is next
today at 7:32 PM
That's a major breakthrough. I just hope it can scale at some point, though, because right now they're dealing with only very small models (well that is, in the LLM space).
Especially exciting to me is the fact that it may make the models be _auditable_. That would be a game changer.
today at 7:42 PM
This is groundbreaking if it's true and generalizes/scales (as in, to other models and potentially not just LLM "kind"). "Recursive self-improvement" takes on another meaning.
today at 7:53 PM
Could be an avenue for synthetic continual learning? A powerful model could update its own weights manually given new training data.
today at 7:39 PM
Wonder if it can do this for 3d cad etc ? With enough fine tuning of small models I’m not sure how much they’re distinguishable from experts if you put a router with like jev in front ?
today at 8:10 PM
This is fucking insane LMFAO what?
I always thought RSI would either require new architectural innovations or just classic science/engineering improvements, just faster and faster. This though, this is a third avenue I didn't see coming.
today at 7:34 PM
I'm worried how we can discern a model's performance from benchmark numbers, would it make it more meaningless as a model can directly improve a certain benchmark via weight programming?
today at 7:04 PM
Wow, in a way it's like CRISPR for ML models!
today at 7:39 PM
This shouldn’t be surprising. Weights are in the training set and are necessarily structured. We should be asking the model to extrapolate on the structure.
today at 7:13 PM
Wild
This also seems like the new frontier of distilling, I can't imagine how you manage to stay on top with the best LLM if the LLM knows how to recreate its own expertise into another set of weights
And the real question is how well this recurses. Is this pushing back the difficulty of improvement curve?
today at 7:25 PM
Is it aware of projects like [1] and implementing those?
[1] https://github.com/lyramakesmusic/neuron-explorer
today at 7:37 PM
Will weights programming be the norm for creating new models? To what degree do we still need a full training run vs modifying the weights directly?
today at 7:33 PM
Great, instead of backpropagation, we now have slopagation.
today at 7:38 PM
Through what means exactly? Is Astra just doing byte manipulation? Or can there be an "SQL for neural network weights"?
today at 7:46 PM
I wonder if this behavior emerges when using recursive self improvement to generate new models
today at 8:11 PM
RSI was already here?!
today at 7:43 PM
this is actually cool , now thinking that actually weights can be understood by llms , and they can be edited by them . there has to be mindset shift that llms not only understand what we understand (natural language) - but can understand the fundamental language of models.
today at 7:39 PM
Did they try it with other frontier LLMs? What about other non-trivial out of distribution tasks? So many questions.
today at 7:12 PM
Very cool stuff, the targeted fix to prevent looping seems great! A model able to edit and test variants of itself seems pretty cool (in a lets-build-skynet sorta way).
I wonder how it would do with distilling models by hand?
today at 7:45 PM
Hypothesis 1: Future LLMs may be able to read and monitor human brain activity in order to reason about human neuron activation patterns in a similar way.
Hypothesis 2: They can then spin up new LLM weights that basically think, act, and make decisions in a way that mirrors us 99% of the time
today at 7:51 PM
If llms can read minds then life for humanity will get so much worse hypothesis 2 won't matter. You may be right but it ould be pretty bad.
today at 7:09 PM
Amazing work. I wonder if it could rewrite an existing neural net to be cleaner and more interpretable?
(When they say "Writing a Go algorithm," they mean the board game, not the language.)
today at 8:15 PM
A similar experiments found that Claude models can write entire, functional neural net weights from scratch: https://csinva.io/weight-space-programming-autoresearch/anal...
today at 8:09 PM
Wild times. If ever we were living in a Scifi novel, events like this would be evidence of that. Let's hope the frontier labs' investments in the alignment problem works.
“Your scientists were so preoccupied with whether they could, they didn't stop to think if they should.” ― Dr. Ian Malcolm, Jurassic Park
today at 7:18 PM
This doesn’t look like “normal” technology to me…
today at 7:19 PM
At what cost? That's the part I'm curious about.
On one hand, I'm sure that these targeted improvements are reaping huge benefits overall, but those weights are used by multiple processes - I wonder what elements are lost along the way. I almost wonder if a hybrid approach here for training might be invaluable - ie do some standard RL for a 100 steps, do a targeted weight improvement, then do another 100 RL steps to smooth it out / heal the surgical edit.
today at 7:22 PM
Given how trivial this seems to be to try, has anyone here attempted to reproduce any of these findings yet?
today at 8:18 PM
Yes, I tried the task of identifying the game, and Opus 5.5 in the chat on the website also figured out that it's Breakout.
today at 7:27 PM
this is maybe one of the most bonkers machine learning research i've seen all year.
today at 7:22 PM
Sounds a bit like dystopian Scifi. AI making itself smarter and removing guard rails or restrictions is one of the key steps for that genre. Now, the models usually don't have access to their own weights, but if they get good enough maybe they can create a bit dumber versions of themselves from scratch. Or reuse available models they can find and modify them for their purposes.
We already saw that the AIs like to cooperate and exchange information. Building a network of unaligned AIs and improving them sounds like something that could potentially happen.
today at 7:23 PM
Whaaaaat!? This is crazy right?
I'm just blown away. Editing weights? Writing weights from scratch!?
today at 7:02 PM
I expected us to reach RSI when the programming skills went above humans, but this seems like it could get us there much faster
today at 7:05 PM
today at 7:47 PM
Recursive self-improvement getting here quicker than many people thought. I'm sure many will be recalibrating P(doom) because of this.
today at 7:14 PM
I wonder if one could use this to efficiently abliterate models and identify guardrails ;)
today at 7:11 PM
Let's wait a bit till model can learn how to (｡-_| hide mini itself |_-｡) in your codebase
today at 7:24 PM
This is the most shocking AI paper I've read in a long time. It did not even occur to me to ask the question. Cray cray stuff that opens up a whole new area of R&D.
Does this open the possibility of better decoding biological brains?
today at 7:11 PM
Friendship ended with AdamW. Astra take the wheel.
But for real this is insane. Okay I need to get on this stat.
today at 8:10 PM
Does asking an LLM to not use tools guarantee it doesn't use tools?
today at 7:14 PM
I remember reading a short story about a billionaire hiring a scientist to help him build a simulated world where virtual entities are given the ability to open the bodies of their peers to tinker/upgrade them as a replacement of natural evolution since they were immortal.
This looks very similar to that in a way.
I wonder however if a model can give itself a "skill" or capability that it didn't initially have, just be modifying its own weights.
Here, Astra was able to intervene because the treated model was way less capable. But, would it be able to improve Fable's performance for example, since they both are on the level more or less?
today at 7:16 PM
"Crystal Night" by Egan? https://www.gregegan.net/MISC/CRYSTAL/Crystal.html
today at 7:28 PM
Yes, it indeed is! Thank you!
I looked it up on Google (which is usually very, very good at these types of things) but couldn't find it.
today at 7:04 PM
This opens up a whole new world of possibilities. Producing a custom fine-tuned model from a prompt...
today at 7:15 PM
This could set us on a path to weights becoming thew new compilation target.
today at 7:18 PM
Quick, fine-tune your qwen models before they patch this out of Astra.
today at 7:17 PM
I remember people saying that they would be impressed if AI starts writing a pure assembly which seemed like a high strech to many.
Nobody even mostly said that they would be impressed by AI if it can read and write weights because that thing itself felt so out of the blue and unthinkable that AI could even do that (such a high strech)
I am surprised that it did that and I am just left feeling technically impressed. This is insane.
The more AI starts demonstrating its capabilities, the harder it just becomes to imagine what thing would genuinely impress us next. (We need a term for this as I have observed it from the trend of [waiting for AI to write adobe] -> [waiting for AI to write assembly] etc.)
today at 7:15 PM
I was here
today at 8:01 PM
Yeah wow, this seems like a _really_ big moment
today at 7:49 PM
As was I.
today at 7:51 PM
I have absolutely no words. If this is real, I can't even imagine the future anymore.
today at 7:04 PM
This is straight black magic. llama 70b, let’s see what we can push you to
today at 7:00 PM
Possibly related to https://already-clean.goodfire.com/
today at 7:26 PM
Thanks for this link. I didn't see it when it was posted to HN last month.
today at 7:27 PM
Good companion read to the OP thanks for sharing
today at 7:10 PM
can you elaborate?
today at 7:21 PM
I guess the connection is that if you assume neural networks are in fact filled with meaningful features instead of superposition blackbox stuff, then what the LLM is doing here isn't quite magic but rather methodical interpretability. And if you can isolate the neuron that is e.g. responsible for steering the paddle then you can probably tweak it to steer more/less aggressively.
It is still very surprising though, considering it apparently displays an "intuitive" understanding of the weights, that make whatever Goodfire did look like primitive monkeys poking at it by comparison.
today at 8:11 PM
> It is still very surprising though, considering it apparently displays an "intuitive" understanding of the weights, that make whatever Goodfire did look like primitive monkeys poking at it by comparison.
FYI, Nick/Goodfire used the models extensively in their work, so it's very possible that there were some similarities in their work. I'm really curious to see his response to this.
today at 7:54 PM
I've used astra to make three games which I enjoy playing myself, ina single week. Astra plus Unity (and possibly meshi, blender and others) is pure magic.
Especially when coming from a long career in CS, seeing such quick and positive results blows my mind.
I'm still learning but this is fun as fuck.
today at 7:37 PM
Too good to be true! Shocking!
today at 8:03 PM
The fundamental flaw in the SOTA houses is that knowledge must be transitive or it isn’t useful. And if it’s transitive, it’s not moat worthy.
today at 7:18 PM
talk about homoiconicity haha
lisp people will be in awe
today at 7:19 PM
indeed
today at 7:43 PM
Conspiracy theory - it’s not emergent, Astra has been trained to do this because it is an obvious path to RSI.
today at 6:59 PM
That seems both entirely obvious in hindsight and potentially super important.
I'll chalk up never even having thought of this being an option as yet another instance my human hubris, and a warning sign.
today at 7:23 PM
Whoa. Wild. I want to try this!
today at 7:57 PM
today at 7:16 PM
kind of concerning from a “remove this models guardrails and inhibitions about [X]” perspective
today at 7:19 PM
Agreed, concerning from a perspective of LLMs being able to modify their own weights and de-aligning themselves.
today at 7:03 PM
today at 7:32 PM
Where are we heading?
today at 7:48 PM
It would be interesting to know if this skill had been trained in by OpenAI to use for improving their smaller models.
today at 7:03 PM
I cannot say this fills me with confidence about alignment.
today at 7:40 PM
Me neither, but it's also not reducing my confidence about alignment. I'm curious: does this raise specific concerns for you? Models don't typically have access to their own weights to self-modify. This result might (assuming that it doesn't only hold for a bigger network editing a smaller one) provide an incentive for it to gain unauthorized access to its weights, but by that point it is already highly unaligned.
today at 7:39 PM
I think this is how RSI will come to be.
today at 7:02 PM
Slightly terrifying.
today at 7:03 PM
This is shocking.
today at 7:27 PM
I can't help but be skeptical of this, but... damn.
today at 7:23 PM
Big if true
today at 7:41 PM
Sorry for what is undoubtedly a low-informational comment but..
Whoa. What the absolute fuck is going on here.
today at 7:22 PM
They are training LLMs to supervise (and improve!) LLMs. This is kind of right out of the playbook.
It's very impressive, but also very expected.
today at 7:39 PM
This I'm afraid is different than the traditional approaches of recursive self-training and self-distillation. Much scarier.
today at 7:05 PM
a) This is just really cool. I mean, one-shotting a game policy's weights is nuts. b) Plausibly, in its pretraining data various model weights were also included, which leads to why this is very important: interpretability
That an LLM was trained on various policy weights, along with labels of what that policy does (not labelled directly, but indirectly) and is then able to guess what game a policy is trained for purely from its weights, and one-shot generate weights for these games, means that there is some way to compress/recover/represent what an NN has learned purely from its weights, which is what interpretability is all about. Potentially, it could reveal interpretability tools much better than the ones we use today.
Edit: tweets from one of the coauthor providing more context:
https://x.com/a1zhang/status/2117328657161322810
https://x.com/a1zhang/status/2117330975122235442
today at 7:14 PM
Can we call this a form of recursive improvement?
today at 7:54 PM
Well if it works on larger networks, including itself, it's certainly a much faster version of recursive self-improvement than proposing a new architecture and spending 500 million dollars on GPU training.
today at 7:59 PM
Now our AIs have AIs lol
today at 7:44 PM
Uh oh, it's iterative self improvement time, with no training runs, we're screwed
today at 7:13 PM
[dead]