Until you're 10+ in a startup , the only roles that exist are head of get shit done & chief common fucking sense officer.

se redes sociais tivessem lista de bloqueio padrão, o felipe neto certamente estaria na minha, enfim, feito

Not every researcher is a good one — or a good reviewer. Still, we need academics in academia and business minds in the market. It’s about trying, being part of the process, and learning to live with it.

oh no

Paul Graham Moderately intelligent humans have _a lot_ of implicit context, that's why they can figure out what is needed from a couple of words or even a non-verbal gesture. What we call "prompt engineering" is primarily about providing AI with context. We'll need more of that, not less.

Officially a PhD Candidate.

PROVOCA | ANDERSON SOARES | 20/05/2025 youtu.be/9BZI6LcuD3U?si… via YouTube

CUDA Teaching Center youtube.com/playlist?list=… via YouTube

Can’t wait to ICLR 2026 at home.

If software engineering leverage increases dramatically thanks to AI-assistance, but still requires some engineers to drive (looks likely currently), I think we'll see work shifting to indies getting paid per outcome faster than companies adjust to high salaries.

I thought more people used Bluesky.

Se você olhar bem, a base de desenvolvimento de software hoje é composta por coisas “antigas”: Linux, RDBMS, SQL, processos e filas.

Se vc não pagar 1000 reais no meu curso vc ta deixando dinheiro na mesa

Forget bluesky, Twitter/X still the best. Just need to avoid toxic content.

Não gosto de leetcode, decoreba não mede experiência.

KV caching in LLMs, clearly explained (with visuals):

Some advice to anyone starting a PhD in ML, or things that I heard from more experienced researchers and I tried to follow: 1) focus on a real problem. Something tangible, that can benefit people. Talk to industry folks if you're looking for open problems. Talk to the end (1/8)

ICLR 2026 will take place in 📍Rio de Janeiro, Brazil 📅 April 23–27, 2026 Save the date - see you in Rio! #ICLR2026

🤔 What is KV cache offloading and why does it matter for LLM inference? #LLMs use the KV cache to accelerate inference speed by avoiding recompute for KV during the prefill phase, however cache grows linearly with requests and has to be evicted to due to GPU memory limit. KV

decades of human evolution just for this