Savlar4 sources healthy · latest episode 15 Oct 2026
Speaker

Alex Zhang

Guest on Latent Space: The AI Engineer Podcast. 306 claims across 1 episode, Oct 2026.

306claims, each timestamped
0 of 12signals on the brief
0verified changes of mind
29open predictions

Positions over time

Every pair of statements that was compared. Only a verified change counts as a change of mind; rewording, added nuance and different questions do not.

No two statements by this speaker have been compared yet.

Predictions

As stated. A window appears only where the speaker gave one.

ExpectsBySaidStatus
Alex Zhang and others thought that collecting enough data could enable GPU-kernel development to be scaled and automated, analogously to how Codeforces’ large problem collection supports programming. no date given2 Oct 2026pending
A knowledgeable person may identify an insight that lets a model solve a problem while spending fewer tokens—potentially avoiding as many as 100 billion or a trillion tokens. no date given2 Oct 2026pending
The speaker expects new types of models beyond standard frontier models to emerge. no date given2 Oct 2026pending
Large AI companies will keep pursuing scaling, which may create opportunities for new players if they discover something especially interesting. no date given2 Oct 2026pending
Any open-source Quinn model will perform worse than the system being discussed, because its creators' training method works very well. no date given2 Oct 2026pending
The speaker hopes newer models will be able to play all of the games in the suite very soon. no date given2 Oct 2026pending
Newer models will probably become able to play every game in the benchmark soon. no date given2 Oct 2026pending
The speaker plans to publish many additional ideas about RLMs. no date given2 Oct 2026pending
The speaker hopes that training on fewer environments will enable models to generalize more than existing harness-based, relatively naive training approaches. no date given2 Oct 2026pending
The speaker hopes that training on these tasks will generalize to a wider class of problems. no date given2 Oct 2026pending
Many kinds of harnesses beyond RLMs remain to be discovered and may provide similar generalization benefits. no date given2 Oct 2026pending
The speaker raises the possibility that a model will be trained to behave like an RLM implicitly within its forward pass. no date given2 Oct 2026pending
Researchers will uncover many approximations of RLM-like behavior, despite the challenge of non-differentiable code. no date given2 Oct 2026pending
The speaker expects creative, strongly opinionated harness designs to produce serious gains by making greater use of language models, which current harness designs underuse. no date given2 Oct 2026pending
The RLM approach may scale substantially better as available data and environments increase. no date given2 Oct 2026pending
The speaker suspects that Anthropic and OpenAI train only on their own harnesses and probably exclude competitors' harnesses from training. no date given2 Oct 2026pending
The speaker thinks that as models improve, differences between harnesses will matter less, and that Astra is capable enough to work in OpenCode without behaving badly. no date given2 Oct 2026pending
Models that correctly perform RLM-like workflows will be substantially more efficient. no date given2 Oct 2026pending
The speaker expects compositional harnesses to make each language-model call in distribution, even when the overall task is out of distribution. no date given2 Oct 2026pending
Whether training around a smart harness will yield better post-training scaling laws. no date given2 Oct 2026pending
Harnesses may emerge that work better according to the discussed compositional principles. no date given2 Oct 2026pending
Some harnesses that differ from the broadly similar current designs are emerging. no date given2 Oct 2026pending
Alex Zhang thinks Harvey's legal RLM work may have been done in collaboration with Base10. no date given2 Oct 2026pending
Users may eventually be able to query a well-scaling swarm, scaffold, or harness as if it were a language model, while seeing only a front-end interface. no date given2 Oct 2026pending
The speaker may work in the future on advancing tool execution, building on promising approaches developed by academics, particularly programming-languages researchers. no date given2 Oct 2026pending
The speaker believes it is possible to design a harness that makes long-running, simple tasks reliable. no date given2 Oct 2026pending
Many easy tasks in people’s workflows could be automated with a reusable general-purpose harness that users can direct, without designing a specialized agent for each task. no date given2 Oct 2026pending
The speaker believes models can be made to approximate human performance substantially more closely. no date given2 Oct 2026pending
A different solution or a new model might emerge within six months and work much better for a problem Alex Zhang is considering investing time in. 2 Oct 2026 – 2 Apr 20272 Oct 2026pending

Appearances

Claims per episode