Español
Alvaro Correa
2.7K posts
Alvaro Correa
@corrius
Software engineer, building AI agents for the last two years
Madrid Присоединился Ağustos 2010
292 Подписки198 Подписчики
@Movistar_Team @SprintCycling Ostras,¡enhorabuena!.Me lo he perdido, habían pasado 7' de la victoria de Pogaçar y he dejado de verlo. Cruzar la meta ahora es la ambición de Movistar Team.
Español
Enric Mas ha sido el primer corredor del equipo en cruzar la línea de meta en Sondrio, en una exigente etapa inaugural del 🇨🇭 #TourdeSuisse.
Tras una jornada disputada íntegramente en suelo italiano, la carrera entrará mañana en Suiza, con final en Locarno.
📷 @SprintCycling
#RodamosJuntos | @movistar_es
Español
Hoy tengo que volar, me recomiendan algún libro en Kindle? Me he terminado el libro con el que estaba y no tengo nada 🥲
Español
Acabo de descubrir que los scripts de PowerShell, si los ejecutas con Windows en castellano, esperan otros parámetros diferentes a los del inglés, como S en vez de Y, por ejemplo. WTF
Español
you guys aren't understanding this
this isn't memory
this should be a internal stackoverflow, where once an agent figures out a bug it posts a little "question/answer" into the folder
that way future agents can refer to that if they run into the same issue
vogel@ryanvogel
I still think we need some /learn command I just burned 1.3M fable tokens on some weird expo bug, then fixed it, but then spawned a new session and fable made the same mistake AGAIN. There needs to be like some internal repo stackoverflow reference guide
English
@javisantana La gente de madbuilders están moviendo mucho en Madrid sin ningún objetivo comercial
Español
Do you still remember when we organized tech events without a company trying to sell us something and just for the shake of learning new stuff and show what we were doing?
English
Alvaro Correa ретвитнул
This is important, if you're working on AI research they'll actively and silently boycott your work.
today is AI research, be prepared for it to be absolutely anything where they want to compete (eventually all economy)
jonready.com/blog/posts/cla…
English
Alvaro Correa ретвитнул
@karpathy This is not a day for celebrating, Andrej.
It's a very dark and very sad day, and the damage may be impossible to undo.
English
Anthropic ha modificado su nuevo modelo para que no sirva para mejorar otros modelos, ni destilar ni colaborar a mejorar y lo peor de todo es que no será transparente si no que dará respuestas erróneas aposta
Nathan Lambert@natolambert
Labs starting to pull up the ladders on the ability to diffuse AI was inevitable. Doing it without telling the user is misaligned.
Español
@badlogicgames Sec filing is confidential, you shouldn’t talk about it
English
not throwing stones in my specific glass house. i test in prod too.
but then i'm also not a gazillion dollar comoany with a sec filing.
English
with all their model, infra, eval and massive talent might, i do wonder what the test regime before a claude code release looks like, that has things like this fly under the radar.
not meant in a mean way. genuinely curious.
ClaudeDevs@ClaudeDevs
We've reset 5-hour and weekly rate limits for all users on Pro and Max plans. We fixed an issue that caused some Claude Code sessions to spawn excessive parallel subagents, burning through usage faster than expected.
English
@flanagrama De Indra, me habré postulado n veces (n > 20). Sólo conseguí una entrevista. Cero respuestas
Español
Lo de hacer 3 entrevistas y que te ghosteen pasa solo en España?
Español
@mitsuhiko Whatever you choose don’t choose NordPass, terrible UX
English
I remember when I enjoyed 1Password. Those days are long gone. And bitwarden is down a dark path. Is there something actually good in that space? If not: why not?
English
Yet another reason why you should build internal datasets and benchmarks that you can run for each model.
Benedict Kerres@benedictk__
Quite misleading 5.5 xhigh is at 70% at 6.6 dollars. Why not show that @datacurve ? Btw 5.5 high is at 62% and 4.47 dollars average cost.
English
@davidalarse @jacintofleta @fr0ziisc2 Os diría que contempléis Codex, porque revisa un modelo diferente al que ha generado el código. Me funciona bastante bien para encontrar cosas que no habíamos contemplado.
Español
Claude code con este prompt (el default de claude)
```
REPO: ${{ github.repository }}
PR NUMBER: ${{ steps.pr.outputs.number }}
Please review this pull request and provide feedback on:
- Code quality and best practices
- Potential bugs or issues
- Performance considerations
- Security concerns
- Test coverage
Be constructive and specific. Use `gh pr comment` for top-level
feedback and `mcp__github_inline_comment__create_inline_comment`
(with `confirmed: true`) for inline diff comments. Do not approve
or request changes — only post comments.
```
English
Como programo hoy en día en Baker:
1️⃣ Uso Conductor para paralelizar
2️⃣ Lanzo multiples agentes en plan mode (5 es habitual), cada uno con su misión (proyectos diferentes)
3️⃣ Leo el plan, ejecuto y se crean las PRs
4️⃣ No leo las PRs
5️⃣ En Github, se lanza un agente revisor
6️⃣ El feedback vuelve al agente inicial y lo aplica (sigo sin leerlo). A veces hago dos loops de feedback.
7️⃣ Una vez aplicado el feedback, leo las PRs para asegurarme de que no hay liadas y para producción.
Feliz de aprender de ti si lo haces diferente!
Lo cuento en el podcast de @mafortperera 👇
Español
@sanch_ia Yes, maybe I was wrong. IMO the fundamental are being able to learn and adapt fast to whatever the new best thing is plus the first principles you mentioned
English
@corrius So I don’t fully buy the idea that time-in-market translates cleanly anymore.
That doesn’t mean fundamentals disappear. It means they get compressed. You still need product thinking, systems thinking, and user empathy — but they no longer map linearly to advantage.
English