Live data from Hacker News

Autoresearch: Agents researching on single-GPU nanochat training automatically

github.com

61–66 of 66 posts

Re: Autoresearch: Agents researching on single-GPU nanochat training automatically

#61
post #8

but the experiments it did that "improved" validation BPB in the GH screenshot were all basically hyperparameter changes right? So is this better or worse, either per experiment or per unit time, than hyperparameter tuning techniques that don't involve an LLM? It's not clear from this if the LLM is more or less making random changes which sometimes work , and or the LLM thinking actually finds "good" changes because…

this is very far from hyperparameter tuning in at least three important ways: - it can modify code arbitrarily, the notion of a "hyperparameter" dissolves - there is no need to run "sweeps" - this is the standard parallel process that wastes compute. because LLM agents are sequential, they can do more efficient versions such as binary search to narrow in on the right setting very quickly (usually many parameters will…

The disposition problem you describe maps to something I keep running into. I've been running fully autonomous software development agents in my own harness and there's real tension between "check everything" and "agent churns forever".

It'a a liveness constraint: more checks means less of the agent output can pass. Even if the probabilistic mass of the output centers around "correct", you can still over-check and the pipeline shuts down.

The thing I noticed: the errors have a pattern and you can categorize them. If you break up the artifact delivery into stages, you can add gates in between to catch specific classes of errors. You keep throughput while improving quality. In the end, instead of LLMs with "personas", I structured my pipeline around "artifact you create".

I wrote up the data and reasoning framework here: https://michael.roth.rocks/research/trust-topology/

Re: Autoresearch: Agents researching on single-GPU nanochat training automatically

#62

As ai improves, most tasks will become something like this. Environments setup where the model learns through trial and error Any human endeavor that can be objectively verified in some environment like this can be completely automated

it's called reinforcement learning

Re: Autoresearch: Agents researching on single-GPU nanochat training automatically

#63
post #35

Earlier quoted context omitted.

don't forget the size of the search space...

this is why big tech is spending 500B on GPUs

that they don't even have the datacenters to plug them in, not the power generation needed to run them if they did

Re: Autoresearch: Agents researching on single-GPU nanochat training automatically

#64
post #52

Once this can run on stock hardware, set the goal to be replicating to other machines. You get a nice, massively parallel, intelligent guided evolution algorithm for malware. It could even "learn" how to evade detection, how to combine approaches of existing viruses, how to research attack methods, how to identify and exploit vulnerabilities in open source libraries, how to phish, how to blackmail, etc. Maybe even le…

Number of machines under control is a measureable target. Quite suited for this concept, at least in theory.

Re: Autoresearch: Agents researching on single-GPU nanochat training automatically

#65

As ai improves, most tasks will become something like this. Environments setup where the model learns through trial and error Any human endeavor that can be objectively verified in some environment like this can be completely automated

Many "subjective" tasks can also be done in an "objective" manner - as long as there is a large enough dataset to estimate what humans would evaluate the outputs - and the evaluators being reasonably consistent. Many human preferences are relatively homogeneous, or sometimes clustered into groups. And there are whole fields of study/practice of such phenomena, such as sensory science - with applications in food, audio, images etc.
Post reply on HN