Nic Becker

[max @maxbittker]

one day = one second of runebench public server bot activity

[Kevin Jiang @jiang_kev]

We just raised $5.7M for @PolarBrowser, the AI browser that beats Anthropic and OpenAI on every major web agent benchmark.

- 4,500,000+ actions taken for users, automating sales, recruiting, and ops
- One company cancelled Clay and saves 25+ hrs/wk per person
- Team is from MIT, YC, Prod, Citadel, Jane Street, Perplexity, Modal, and Apple

100 hours of work. From 30 seconds of typing.

Download the world's most powerful AI browser: https://t.co/lUOm01Stm0

[Applied Compute @appliedcompute]

Controlling staleness in an async RL stack has not been well understood, so we derived a closed-form formula that predicts staleness in advance. Our predictions match measured staleness from production RL runs within a fraction of a step. Building these simulations has led to key insights on how we maintain high GPU utilization during our RL runs without sacrificing ML performance.

[Applied Compute @appliedcompute]

Preserving entropy is critical for continued training; in modern post-training recipes, entropy is often a fixed resource that gets exhausted over the course of a training run, making it difficult for the model to improve and learn on new tasks.

Adaptive entropy control methods expose entropy as a controllable hyperparameter instead of a side effect. We find that in a continued training setting, GRPO leads to entropy collapse which stalls training performance on subsequent training phases. REPO-R (Petrenko et al., 2026) can hold entropy near a specific target, preventing this collapse, and pushing performance higher in follow-on tasks when the target entropy value is properly tuned.

[Nic Becker @caenopy]

@mmabrouk_ @appliedcompute @nvidia That’s right, the downstream model owns the trajectory. Mid-trajectory model switching can be helpful, but you you will pay for it with cache invalidation across turns where the model switches.

[G Jones @gjonesbass]

every time i see a new "create a song from a single prompt" kind of product being rolled out i think about these comments from @David_Rudnick

[neilson @neilsonks]

oakwood studio is evolving

[Nic Becker @caenopy]

@tobyshooters neogeocities

[cristóbal @tobyshooters]

I work on a tool for working with the information in videos called https://t.co/Z1hWUTrkqL

We just released "Videoboard," a 2d canvas for storyboarding + clustering. It lets you edit video on an infinite canvas—a “really non-linear” editor.

Some background...

[Nic Becker @caenopy]

new paper up on arxiv!

[Surya Ganguli @SuryaGanguli]

1/ Our new work: "How many degrees of freedom do we need to train deep networks: a loss landscape perspective." https://t.co/O7UKkSOO63 We present a geometric theory that connects to lottery tickets and a new method: lottery subspaces. w/ @_BrettLarsen @caenopy @stanislavfort

[Luiz Pessoa @PessoaBrain]

What does a leading 𝗻𝗲𝘂𝗿𝗼𝘀𝗰𝗶𝗲𝗻𝘁𝗶𝘀𝘁 think about *causation*? Summary of some positions by Buzsaki in The Brain From Inside Out (ch 2).
Plenty of discussions about causation but here we have someone at the very “center of the action” in neuroscience.
🧶

[Nic Becker @caenopy]

@_fernando_rosas Great list, here's another one: Projections by @KarlDeisseroth in June

https://t.co/PTlVFQRBtH

[Karl Deisseroth @KarlDeisseroth]

Some rare personal news…
My first calling, as many of you know, was writing. I’ve always found (unreasonable levels of) joy in working with words.
Now I’ve somehow written a book: Projections.
@randomhouse https://t.co/pVzT3E6SFT 1/12

[Nic Becker @caenopy]

@fchollet i imagine this will also upset what is deemed mainstream vs. alternative history. future historians will be inundated by the multiplicities of explanations at their disposal.