Rafael Pinto

Ah não.............

Olha a pegadinha, hein... não vai ter anistia só pra quem planejou matar?

Open model performance rankings across most benchmarks these days.
Top half: China
Bottom half: Everyone else

So many such cases.

Certamente tinha uma base do Hamas escondida embaixo da tenda.

EU TÔ ME CAGANDO DE RIR!!!

HHUAUHAHUAHUUHAHUAH VTNC

Jeff Hawkins even wrote a book about that back in 2004.

From the accelerate community on Reddit: GPT5 Progress Is Right on Track - 3 Charts Explore this post and more from the accelerate community

www.reddit.com/r/accelerate...

Even IQ tests have subcategories. Peforming badly at one of them doesn't make you straight dumb. I'd say you can even get a Ph.D. (or a few)

It could be argued against that analogy and that's fine. Maybe "LLM physics" would be more adequate to refer to the part we know. What's important is that we know the lower level but are still struggling with the higher level that emerges from it.

IOW, there are at least 2 kinds of "knowing" / "understanding" when talking about LLMs, and people usually confuse and/or conflate both.

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.

Seems like you're a parrot.

Agora a Atlas faz enquete? Suprassumo da metodologia estatística.

lol

Ué

Tadinha da Áustria, tão pobrezinha.

NÃO POSSO MAIS COMETER CRIMES!!!!!!!!1!!1!111

The "AI is useless" crowd should feel ashamed.

The original definition of AGI in the 00's was basically "not-narrow-AI" or "multi-task-AI". The goalpost has moved so far away by now, and will keep moving...

Compartilhei com essa esperança mesmo hahaha

YouTube video by Bob the Hamster ohrrpgce 2025 07 timelapse

@daysdanilo.bsky.social olha quem recém postou vídeo: www.youtube.com/watch?v=ZrN5...

"I want AI to do my laudry so I can make art, not the other way around!!1!"
Ok, what now?

Tem sim, não seja modesto.

Nossa, ele não vai nem dormir hoje.

TACO

Hierarchical Reasoning Model Reasoning, the process of devising and executing complex goal-oriented action sequences, remains a critical challenge in AI. Current large language models (LLMs) primarily employ Chain-of-Thought (CoT...

HRM: Hierarchical Reasoning Model

ngl this sounds like bullshit but i don’t think it is

- 27M (million parameters)
- 1000 training examples
- beats o3-mini on ARC-AGI

arxiv.org/abs/2506.21734

I automatically dismiss any argument that starts with "AI will never".

“Both architectures are optimized with Adam.
Who/what is ‘Adam’? I think this is a very serious typo…”