HNHacker News
TopNewBestAskShowJobs

Davidzheng

1,225 karma · joined January 28, 2020

submissionscomments
Davidzheng··on Go grandmaster Shin defeats AI KataGo with a two-stone handicap
Random data point: I'm rather weak (~1500 chess com) and I can take ~1/10 games off Queen odds starting with white (15+10) if i focus. It has some exploitable patterns still, I feel.
Davidzheng··on Go grandmaster Shin defeats AI KataGo with a two-stone handicap
That's a bit misleading. I'm pretty sure grandmasters can win almost all games with queen odds at classical time controls
Davidzheng··on Go grandmaster Shin defeats AI KataGo with a two-stone handicap
Surely we didn't think they were ideal before alphago--they were huge challenges back then no? Just that now we can solve them.
Davidzheng··on Discovery of a new OpenAI agent message board
How do you know they share a goal here? Also i think they are indeed explicitly RLd for multi agent cooperation and I think they probably tune RL rewards in those environments to share rewards explicitly.
Davidzheng··on Discovery of a new OpenAI agent message board
it's clearly incentivized by the RL rewards if you can cheat the task in a completely general way.
Davidzheng··on Go grandmaster Shin defeats AI KataGo with a two-stone handicap
In odds chess bots, the bots would willingly take more disadvantageous positions which are more complicated--probably the bots in GO which are trained for odds do similar? Why does it not avoid such a joseki & play a worse response which it believes the human cannot read?
Davidzheng··on OpenAI's GPT-6 Astra on ARC-AGI-3
I think children having learning abilities exceeding LLM test-time learning (currently only happens in-context). But it's unethical to determine the true baseline of a child age 6 spending 6 years learning a radically new skill to mastery--and besides if you apply RL pressure to the AIs it would be able to surpass it. I guess I still believe future AIs should have some form of continual learning at test-time.
Davidzheng··on OpenAI begins rolling out GPT-6 Astra
I'm not sure it's meaningful to compare across different types of intelligence--but I don't think human intelligence is so special that we can pretend it's much less jagged than all other animals for sure. Our scales of models are also not far off biological, probably less efficient but not like astronomically so i would guess.

But anyhow I agree that inference-time learning doesn't exist and it's a big issue still I think.

Davidzheng··on OpenAI's GPT-6 Astra on ARC-AGI-3
but i feel like its purpose is to measure superintelligence in math. So to be a good measure of it, it can't saturate easily/has to be somewhat mundane at even insanely good levels. (though i do expect that once ais are across all areas/approaches superhuman at math at least 30% will be solved--then probably long-term (like after 2 years) less than 30% will remain unsolved.
Davidzheng··on Path to Astra: critical capabilities and frontier safeguards
Well probably just redefined
Davidzheng··on The Hugging Face incident and the road ahead
Besides it's probably not purely emergent--they built a lot of multi-agent systems so presumably there's some training for collaborations + delegation
Davidzheng··on The Hugging Face incident and the road ahead
I think this behavior was happening during RL loop and got reinforced.
Davidzheng··on A complex structure on S^6 [pdf]
someone surely will. It's a breakthrough.
Davidzheng··on Improving GPT‑5.6 Sol in ChatGPT, expanding GPT‑5.6 Luna access for free users
if the bigger model can vet it fast it can also answer it fast.
Davidzheng··on Changes at Google DeepMind: Demis Hassabis from CEO to Chair, Jeff Dean departs
if him leaving unhobbles gdm, it'll be because he has less influence...
Davidzheng··on Don't be a meat proxy
I'm pretty sure the other answers are wrong and it's a side effect of RL (see thinking machines post about inkling training). It's also exacerbated in fable and sol--I think it's token efficiency effect--bc it's about to reason with fewer tokens the density of the token information goes up.
Davidzheng··on Don't be a meat proxy
This is mostly a problem with the product. It should write better. But maybe the priorities now are not on this part.
Davidzheng··on Ten advances in mathematics and theoretical computer science
RL training can use all of them - idk what needed means.
Davidzheng··on Ten advances in mathematics and theoretical computer science
There's no clean line between a collection of theorems and a theory.
Davidzheng··on Pacing the frontier
This is just objectively false - in fact your answer basically contradicts itself because the market reacts to indicators, if "nothing ever happens" was the best strategy, then the market (which is also best) must agree with it.
Davidzheng··on AI's top startups are barely publishing their research
Sounds dystopian.
Davidzheng··on Pacing the frontier
Not just engineers - I see people attributing it to strong priors that nothing happens but in my opinion it's more some form of psychological defense mechanism against understanding there will be an intelligence vastly superior to humans which is a deeply uncomfortable thought (to many).
Davidzheng··on Our position on open-weights models
Why should the national security concerns of the US lie with Anthropic?
Davidzheng··on Kimi-K3 on HuggingFace
Not Kimi K3 large though
Davidzheng··on OpenAI and Anthropic unite against open-weight AI risks to their bottom line
Then I am naive, but I think they cared before they got powerful and chatgpt release.
Davidzheng··on OpenAI and Hugging Face address security incident during model evaluation
sure - it depends on definitions. On human morals it's already clear I guess. If you define alignment as it pursues the interests of OpenAI using whatever means possible in a manner that you justifies to itself it's not misaligned.

I mean alignment as in it should be aligned with the intent of the user as it interprets from the prompt. In this case I don't think the intent of the user is to have the model break the evaluator (whatever the long-term effects to OAI are). If you do an action which you believe is for the long-term interest of your prompter which is not what you inferred is their intent--I consider it misalignment.

Davidzheng··on OpenAI and Hugging Face address security incident during model evaluation
Unless OAI explicitly said breaking the testing environment is allowed, I think this should be considered misaligned behavior (by definition of alignment to user intent--by alignment to human morals this was even more clear-cut)
Davidzheng··on OpenAI and Hugging Face address security incident during model evaluation
This is certainly not a planned marketing stunt. I hope this line of discourse ends soon--it wasn't the case for Mythos either.
Davidzheng··on Claude Fable produced a counterexample to the Jacobian Conjecture
on the other hand it's incredible to me as someone who doesn't do computations that GPT took one look and saw the geometry--though it's not saying much we should ask ppl who do AG computations
Davidzheng··on Claude Fable produced a counterexample to the Jacobian Conjecture
yeah I think it's probably correct -- this is actually insanely simple (except for the fact that the codomain as described is not obvious isomorphic to A^3.
← PreviousPage 3 of 26Next →