Rafael Pinto
@rcpinto.bsky.social
4.6K followers 240 following 950 posts
CS / AI / ML PhD / Professor. 🇧🇷 Working on meta-learning, continual learning, control, evo, RL, game AI. Hobbyist game dev (@megamanbeyond.com) Local news and useless facts about my life in PT-BR. EN otherwise.
Posts Media Videos Starter Packs
rcpinto.bsky.social
Buttobi CPU, I see you're a person of culture.
rcpinto.bsky.social
Counterfactual reasoning✔
rcpinto.bsky.social
The best reason to root for Helix and their Figure robots so far:
rcpinto.bsky.social
Michael Levin would like to have a word.
rcpinto.bsky.social
After using Gemini 2.5 Pro, Claude Sonnet 4.5 and GPT-5 for highly novel theoretical neural networks research, I must say that, at least for my case, GPT-5 is the clear winner. The "Ph.D. level" claims are correct (and of course, like any human Ph.D., it also makes stupid mistakes now and then).
rcpinto.bsky.social
Pensei que fosse o Toninho do Diabo.
rcpinto.bsky.social
I wasn't familiar with that concept at all, but it seems very relevant to the subject indeed, thanks!
rcpinto.bsky.social
rcpinto.bsky.social
IOW, there are at least 2 kinds of "knowing" / "understanding" when talking about LLMs, and people usually confuse and/or conflate both.
rcpinto.bsky.social
Do you realize the abysmal difference between being a user (and "looking into it") and doing decades of research and, in fact, creating the very foundation for those things?
rcpinto.bsky.social
The hatred coming from the butthurt anti-AI mob is now comparable to that of far-right scumbags.

It's one thing to dislike the technology, but the way they're responding to some people here is sick to say the least.
rcpinto.bsky.social
If you confidently assert that LLMs (or anything else besides yourself, for that matter) are or aren't conscious, you're wrong.
rcpinto.bsky.social
I dream of a social network that's like a generalization of that. For any topic, you have to prove knowledge before posting.
rcpinto.bsky.social
rcpinto.bsky.social
Even if we had perfect knowledge of neuroscience, that wouldn't imply perfect knowledge of psychology.

We have perfect "LLM neuroscience", but not "LLM psychology". We know how parameters are updated and outputs are computed, but not what internal mechanisms it is learning and how it uses them.
rcpinto.bsky.social
The results show exactly the opposite.
rcpinto.bsky.social
Here we go, another thread full of geniuses and experts that know everything about how AI is just autocomplete. I'm glad to grow my block list, tho.