News for hackers. Stuff that matters.

nblintao

Dedicated rollout and trainer nodes, with RDMA transfers between them.
Good point! First of all, none of the benchmark tasks were used in training, though the training data was designed to improve capabilities on tasks like TB4.1. If you consider TB4.1 the main target, the blog actually shows before/after results on several other benchmarks, with no regressions and even some improvements.
Doesn't look like just a thin wrapper to me. The interesting part seems to be the surrounding harness/workflow layer rather than only the model call itself.

I was trying to keep track of the better post-leak code-analysis links on exactly this question, so I collected them here: https://github.com/nblintao/awesome-claude-code-postleak-ins...

The part I keep going back and forth on is openness. This whole model assumes experiments, notes, and results are public. That works great for "open research for the love of it". But even in academia people worry about getting scooped, and industry is way more guarded. I'm not sure how large the audience for truly open ML experiments is. Curious if others have thoughts on this.
Re:Autoresearch on an old research idea on Sunday March 29, 2026 @11:15PM (#47570032)
The "price of an entire new employee" framing is spot on. I kept running into the same thing: individual experiments are cheap, but they add up fast, and nobody wants to approve that budget for speculative ideas.

I've been thinking of this as a gap between VC/Kickstarter and just doing it yourself. Most early ML experiments are too small for formal funding but too expensive to casually self-fund. So I built ML Patron where anyone can chip in a few bucks to sponsor an experiment they're curious about. I honestly don't have a good answer yet for how this turns into returns for sponsors in a traditional business sense. For now it's just open research patronage, like "I'd pay to know the answer to this". Platform runs it on cloud GPUs with public MLflow tracking.

Still very early: https://news.ycombinator.com/item?id=47563959.

Show HN: ML Patron – Run reproducible ML experiments with integrated funding

https://mlpatron.com

Hi HN, I’m Tao. I built ML Patron for a problem I kept running into with ML ideas: getting from “this is worth trying” to “the experiment actually ran and the result is inspectable.”To me, there are usually three missing pieces that kill an idea: funding, execution, and continuity. Good ideas often stall because nobody wants to pay for that first GPU run, there isn’t a simple execution layer...

( Read More... | 1 comment | on HN )