César de Pablo

MIT professor accidentally leaked his NotebookLM grading system during a Zoom call. Dude forgot to turn off screen share and we watched him grade 47 essays in 12 minutes. Here's what he was doing that blew my mind. He uploaded all student papers plus his original rubric into NotebookLM. Then asked it to "evaluate each paper against these specific criteria and flag any that deviate from expected patterns." But the crazy part was his follow-up prompt. "Now cross-reference writing styles with previous submissions and highlight potential academic integrity concerns." The AI caught three cases of weird style shifts he would've missed on his own. Final step killed me. He asked NotebookLM to "generate personalized feedback that connects each student's weak points to specific course materials they should review." What took him 6 hours before now happens in 15 minutes. And students get better feedback than his handwritten comments ever provided. The man turned grading from torture into actual teaching.

So what’s next? Well, now that AI can develop its own “limbic system”, how can it integrate into an agent’s behavior (or “mind”)? Here’s one way we could try:

We needed “interpretability mappings” to put names to those patterns. Here’s the LOVE 2:5x6 mapping we used, as “ideal references”. Can you match the learned patterns to the 30 ideal references? 👇

Can #AI learn and produce its own emotions, like natural ones? 🤖❤️ Meet LOVE (Latest Observed Values Encoding), a generic self-learning emotional framework for machines. Paper in Nature - Scientific Reports (open access): nature.com/articles/s4159… See how it works! 🧵⬇️

Entrenar #LLMs inclusivos, detectar sesgos en medio de comunicación, combatir el discurso de odio, construir narrativas sostenibles… aprende todo esto y más con nuestro Hackathon #Somos600M 🔥 ¡Únete ya a las casi 500 personas registradas! 🤯 hackathonsomosnlp2024.eventbrite.com/?aff=t

There are 600 million Spanish-speaking people in the world, but we don't have an instructions dataset nor an open #LLM leaderboard... let's solve this! 🚀 Join our Hackathon #Somos600M! This year we also welcome hablo-un-poco-de-español people 💛 ➡️ eventbrite.ca/e/registro-hac…

📢 Acabo de publicar GRATUITAMENTE mi libro "Seguridad ofensiva en machine learning. +100 prompts injections en ChatGPT. Atacando y defendiendo organizaciones con prompt engineering". Más de 160 páginas y 200 referencias. No se me ocurre mejor forma de iniciar al año. Se agradece difusión... #infosec #MachineLearning #ciberseguridad - github.com/mindcrypt/libr… Si te interesa el mundo de la seguridad ofensiva y defensiva y la inteligencia artificial te recomiendo su lectura. Si lo queréis en otros formatos lo tenéis disponible en papel o formato e-book en Amazon - amazon.es/Seguridad-ofen…

Feliz y rojiblanca Navidad ❤️🤍

“Los datos bien utilizados nos cuentan historias sobre qué necesitan nuestros clientes” bbva.com/es/innovacion/…

@BBVA_AIFactory @DotCSV Grande @DotCSV un placer, espero que coincidamos en otras ocasiones.

If you found this valuable, feel free to like/share with your friends 🙏

:-)

For nerds :-)

LLM Powered Autonomous Agents This is an incredible overview of LLM-powered autonomous agent systems. Includes case studies and proof-of-concept examples. lilianweng.github.io/posts/2023-06-…

:-)

:-)

I can’t believe they made this available for free. Huge thanks 🙏 @full_stack_dl fullstackdeeplearning.com/llm-bootcamp/s…

:-)

:-)

This is deliciously dystopian: here's what happens when you force a weird deepfake video/audio generator to keep on running past its natural conclusion