Ricardo Herrmann
4.9K posts
Ricardo Herrmann
@rgherrmann
Computer Scientist. Jack of all trades, master of some. Also @[email protected]
Brazil Katılım Ocak 2009
887 Takip Edilen414 Takipçiler
After much consideration, my users have convinced me to add documentation to PufferLib for use by LLMs. It will politely inform them to RTFM
English
Ricardo Herrmann retweetledi
'The purpose of abstraction is not to be vague, but to create a new semantic level in which one can be absolutely precise' - Edsger Dijkstra
English
Ricardo Herrmann retweetledi
Ricardo Herrmann retweetledi
I want a universal AI agent that just takes all of the code that was ever written and collapses it all down to the absolute minimum code that needs to exist. 95% of the code that exists today, probably doesn’t need to. And it is only getting worse.
English
Ricardo Herrmann retweetledi
Negative Externality
I work on a large open-source project (FreeCAD). This is the growth in daily Pull-Request submission over the last 18 months. Submissions have basically doubled.
All those submissions have to be reviewed by an already overstretched maintainer team.
A significant part of this growth is driven by AI. It helps good developers go faster and lets bad developers generate atrocious code that they don't understand.
...And the trend line is not flattening out.
AI assisted code-review would help but the AI companies do not make the use of their tools free for open-source maintainers. The result is that maintainers have to make a choice:
1) Pay out of pocket
2) Suffer the wrath of the community because code review gets slower.
AI companies train their models on open-source code. While I'm generally an optimist about these tools, this pisses me off.
English
Ricardo Herrmann retweetledi
"Our goal should be to give an LLM coding agent zero degrees of freedom." As usual a thought provoking article from John Regehr. "Correctness oracles abound. We have test suites, fuzzers and property-based testers, runtime sanitizers, static analyzers, linters, strong type systems, and formal verifiers. Any time such a tool can be made available to the LLM, we’ll reap the benefits in terms of not dealing with bugs the hard way, later on." john.regehr.org/writing/zero_d…
English
Haskell's revenge is coming 😁
Paul Snively@JustDeezGuy
This AI hype bubble (#4 in my lifetime) will burst, but when it does, it might just leave in the crater an awareness of the Curry-Howard Correspondence those of us in software have been unsuccessful in advancing for 40 years.
Sao Paulo, Brazil 🇧🇷 English
Ricardo Herrmann retweetledi
I would like to invite @elonmusk, @sama, @DarioAmodei, @sundarpichai, @demishassabis, @tim_cook, @satyanadella, @JeffBezos, @finkd, and all the other tech CEOs to join the "AI race to cure cancer."
The winner gets humanity's forever appreciation.
Luiza Jarovsky, PhD@LuizaJarovsky
Everybody wants AI to help cure cancer. Why isn't every AI company obsessively focused on that?
English
Ricardo Herrmann retweetledi
Ricardo Herrmann retweetledi
> You have zero fucking idea what's going on because you delegated all your agency to your agents. You let them run free, and they are merchants of complexity.
mariozechner.at/posts/2026-03-…
English
I can’t stop thinking about how
1) RL is the dominant paradigm in AI and everyone is designing around it but
2) the old school RL people are almost all uninterested and are sitting the LLM stage out on the sidelines
Not what I would have predicted 10 years ago
English
Ricardo Herrmann retweetledi
Been experimenting with procedural locomotion on #GaussianSplat environments from @theworldlabs .
Built a little multi-legged robot in Unity that raycasts against the splat data to figure out where to place its feet, no meshes or colliders involved... Pretty fun to climb surfaces, walk on walls, and set a weird number of legs on the fly :)
Still rough around the edges but pretty fun to watch it figure things out.
#GaussianSplatting #Unity3D #WorldLabs #ProceduralAnimation
English
Ricardo Herrmann retweetledi
Consider that the dumbest people you know are repeatedly being told "You're absolutely right!" by LLMs.
English
Ricardo Herrmann retweetledi
As a statistician, I keep asking myself how all these AI people are dealing with the massive potential for catastrophic errors in critical analyses, and the answer keeps being they either didn't think about it at all, or they don't care.
English
Ricardo Herrmann retweetledi
JEPA are finally easy to train end-to-end without any tricks!
Excited to introduce LeWorldModel: a stable, end-to-end JEPA that learns world models directly from pixels, no heuristics.
15M params, 1 GPU, and full planning <1 second.
📑: le-wm.github.io
English
Ricardo Herrmann retweetledi
Ricardo Herrmann retweetledi
Statistical Rethinking 2026 is done: 20 new lectures emphasizing logical & critical statistical workflow, from basics of probability to causal inference to reliable computation to sensitivity. It's all free, made just for you. Lecture list & links: #calendar--topical-outline" target="_blank" rel="nofollow noopener">github.com/rmcelreath/sta…
English
Ricardo Herrmann retweetledi
This is more evidence that current frontier models remain completely reliant on content-level memorization, as opposed to higher-level generalizable knowledge (such as metalearning knowledge, problem-solving strategies...)
Lossfunk@lossfunk
🚨 Shocking: Frontier LLMs score 85-95% on standard coding benchmarks. We gave them equivalent problems in languages they couldn't have memorized. They collapsed to 0-11%. Presenting EsoLang-Bench. Accepted to the Logical Reasoning and ICBINB workshops at ICLR 2026 🧵
English
Ricardo Herrmann retweetledi
Universal embeddings FTW 😊
One of the flagship projects at FAIR was to "embed the world" (i.e. represent every entity on Facebook).
The name was soon changed to "Filament", deployed internally, and eventually open-sourced as "PyTorch-BigGraph"
The techniques were more primitive than today!
github.com/facebookresear…
English