@sidharthkmenon@GergelyOrosz@CompleteSkeptic I started using it for guardrails (openai decisions though). Working like a charm. I also use to understand which knowledge bundle to load into the agent
Há anos, eu e todas as pessoas ao meu redor, amigos e comunidade, sempre fomos contra qualquer tipo de violência discriminatória. Minha opinião política nunca foi mistério pra ninguém, e diariamente eu e meus familiares somos atacados e perseguidos por isso.
Mas eu também convivo diariamente com pessoas que tem sua existência questionada e ameaçada por ideologias completamente odiosas. O dia de ontem foi, honestamente, um choque e um presságio péssimo de algo que a gente não pode deixar acontecer de novo.
Uma comunidade grande inevitavelmente reúne pessoas que discordam ou simplesmente não conseguem compreender a proposta de uma obra ou trabalho, mas essas pessoas não representam o meu trabalho nem de todos ao meu redor.
Então, se ainda existe alguma dúvida, a resposta é simples: no segundo turno das eleições de 2026, dia 25, a única escolha é Lula 13. Nós não podemos permitir mais um Bolsonaro no poder.
I hate the feeling of no longer being fully in the loop.
I tell agents to debug an issue. They explain what goes wrong. Propose a solution.
That's sure a lot of jargon. Do I fully understand what they're saying? Kind of. But certainly not like if my mind was solely focused on that bug fix, instead of juggling 5 other threads.
I tell them to do it. Make the fix, create a PR, the usual.
Is their solution really the right approach? Will they not include a dumb invariant that might later bite me in the ass?
Glancing at code, it *seems* fine. But I'm not sure. I'd have to go way deeper into the issue to be able to tell. But that would defeat the purpose of having agents do the thing instead of me. That'd be *inefficient*.
So I merge it in, and move on to the next thread.
Forever falling behind so I wouldn't fall behind.
New Anthropic research: Natural Language Autoencoders.
Models like Claude talk in words but think in numbers. The numbers—called activations—encode Claude’s thoughts, but not in a language we can read.
Here, we train Claude to translate its activations into human-readable text.
Anthropic just dropped Free AI courses on:
- Prompt Engineering
- Building agents
- Best practices for Agentic Coding
- Collaborate with AI systems
.... and so much more!
9 best guides you don’t want to miss:
Why have you been stuck at the Senior Eng. level for so long?
After sitting on 50+ Staff+ promotion committees and performance calibrations as a Director from 2019–2024, here are the patterns I've identified & clustered that show up again and again.
🔽
🚨 BREAKING: HuggingFace just dropped 9 free courses to level up your AI skills.
From LLMs to agents, vision to gaming, audio to 3D and it's all open-source, all hands-on, and all free.
Here’s what’s inside ↓
We wrote up what we've learned about using Claude Code internally at Anthropic.
Here are the most effective patterns we've found (many apply to coding with LLMs generally):
Lately it does seem like the requirements for engineering managers have swung from "you just do people stuff" wayyyy over to "you should not only be technical, but also write code and review PRs every day" --which is nuts.
A good counter: https://t.co/Se42RwGgUJ by @spaceblenny
If you currently or at any point in the future will build APIs for public consumption, please read this entire series by @paul_asjes, it's awesome. https://t.co/FNppMTDYwQ