Founder @KhemLabs • Full Stack Dev • UTN Educator • Open Source Geek • Building meaningful tech with soul & purpose ⚡
In: @imluciomf
Alter ego @dnangelus
The boring checklist still wins:
- never give the model direct access to secrets
- scope tool permissions tightly
- validate outputs before using them
- log prompts, tool calls, and weird edge cases
AI security is mostly software security wearing a new hoodie.
Same bug. New interface.
Prompt injection isn’t weird AI magic.
It’s the oldest security lesson with better branding: don’t trust user-controlled input.
Treat LLM output as untrusted data. Sanitize it. Limit permissions. Log the weird stuff.
#AIsecurity#LLMSecurity
Effective today, we are:
1) Doubling Claude Code’s 5-hour rate limits for Pro, Max, and Team plans;
2) Removing the peak hours limit reduction on Claude Code for Pro and Max plans; and
3) Substantially raising our API rate limits for Opus models.
GPT-5.5 Instant is starting to roll out in ChatGPT.
It’s a big upgrade, giving you smarter, clearer, and more personalized answers in a warmer, more natural tone.
And it's also more concise, which we heard you wanted. We think you'll love chatting with it.
@precisox@opencode OpenCode sounds like a great step toward more transparent AI coding assistants. Excited to try a TUI-first agent built in Go and avoid subscription lock-in!
@alex_whedon SubQ’s sparse-attention design and 12M-token context window are really impressive! Excited to see benchmarks and open-source details—great work pushing LLM scaling forward.
Introducing SubQ - a major breakthrough in LLM intelligence.
It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),
And the first frontier model with a 12 million token context window which is:
- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus
Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).
Only a small fraction actually matter.
@subquadratic finds and focuses only on the ones that do.
That's nearly 1,000x less compute and a new way for LLMs to scale.
@safossatti Congrats to the Enter team on becoming the first AI unicorn in Latin America! 🦄 Exciting to see legal tech innovation thrive in our region.
Today Enter became the first AI unicorn out of Latin America.
Mateus, Michael and Henrique are building the best Legal AI product in the world from São Paulo. Not a copilot, an AI litigator that absorbs the entire workflow end-to-end and is already running 300,000+ lawsuits a year for clients like Airbnb, Bradesco, Nubank, Latam and Azul.
At @KaszekVentures we could not be prouder to back this team. Congrats Mateus, Michael, Henrique and everyone at Enter.
AI coding tools didn’t just speed up development.
They changed the AppSec math.
When code ships faster than teams can review, the question isn’t “did AI write a vuln?”
It’s “can our process keep up?”
Velocity without visibility is risk.
@stackhawk#AppSec#AISecurity
https://t.co/k4pmDRwE2n
Gemini vs ChatGPT vs Grok.
Same creative direction. Three very different souls.
Which image tells the strongest story — and what does that say about how each AI “imagines”? @grok@GeminiApp@ChatGPTapp
Google se lleva 238 mil millones de dólares al año solo por llenarte de anuncios.
Un solo desarrollador creó una herramienta totalmente gratis que los bloquea a todos, antes de que lleguen a cualquier dispositivo.
En toda tu casa.
Al mismo tiempo.
Sin instalar nada en el celular, la tele ni la tablet.
Se llama Pi-hole (más de 57.000 estrellas en GitHub).
Lo pones a correr en un Raspberry Pi de 35 dólares o en cualquier computadora vieja con Linux. Se convierte en el DNS de tu red y hunde todos los dominios de publicidad antes de que toquen tus equipos.
De repente:
- La Smart TV deja de cargar anuncios
- El teléfono navega limpio
- Los niños ya no ven propaganda en su tablet
También detiene:
- El pixel de rastreo de Facebook
- Google Analytics siguiéndote como sombra
- La vigilancia de tu tele inteligente
- La telemetría de las apps que llama a casa
- Los data brokers husmeando en tu red
Un solo aparato.
Una sola configuración.
Un comando.
Una industria de 238 mil millones neutralizada por 35 dólares y una tarde de tu tiempo. 100% open source. Gratis para siempreee.
Link en comentarios.
Coding from glasses is starting to feel surprisingly normal.
Between sips at a coffee shop.
Walking down the street.
On a train.
It fits into everyday moments more naturally than expected.
#BuildInPublic#DevTools#vibecoding