Many AI announcements these days. One of the coolest applications is to make the learning experience on YouTube more interactive, being able to ask questions about the video. If it works as expected it might be a great companion for YouTube learners.

Años 80. Una gran empresa española genera un proyecto donde perfiles diversos emplean design thinking.

¿Software? ¿IA? No.

La empresa es Renfe y van a reinventar el transporte suburbano.

Hoy, #GamuhiloRenombrado sobre la creación de la marca Cercanías.

Con Foro de Marcas Renombradas Españolas ❤️

Colab 📒 Notebook to fine-tune 💅🏽 Google AI #PaliGemma vision-language model 👓🔠🧠
on a free T4 VM!
colab.research.google.com/github/google-…

Nice blog - 'How We Saved 10s of Thousands of Dollars Deploying Low Cost Open Source AI Technologies At Scale with Kubernetes'

opensauced.pizza/blog/how-we-sa…

New open Vision Language Model by Google: PaliGemma 💙🤍

📝 Comes in 3B, pretrained, mix and fine-tuned models in 224, 448 and 896 resolution
🧩 Combination of Gemma 2B LLM and SigLIP image encoder
🤗 Supported in Hugging Face transformers
Model capabilities are below ⬇️

The first open Stable Diffusion 3-like architecture model is JUST out 💣 - but it is not SD3! 🤔

It is HunyuanDiT by Tencent, a 1.5B parameter DiT (diffusion transformer) text-to-image model 🖼️✨

In the paper they claim to be SOTA open source! I'm working on a Hugging Face demo

Amazing news re: our LLM fine-tuning course: All students get $1,000 in free compute credits from Modal and Replicate ($500 each) 💰

Course signups close end-of-day today.

You get more compute credits than the course costs 🤯

maven.com/parlance-labs/…

New short course: Building Multimodal Search and RAG', by Weaviate • vector database's Sebastian {W}italec ✊🏽✊🏾✊🏿.

Contrastive learning is used to train models to map vectors into an embedding space by pulling similar concepts closer together and pushing dissimilar concepts away from each other. This

Just added an /ai page to my website, inspired by Derek Sivers. It's a public statement on how I use AI for the notes and writings I publish. pelayoarbues.com/mocs/ai

Marcet siempre es inspiración, pero este artículo vale más que muchos libros de management y liderazgo.

LIDERAZGOS SENCILLOS

Liderar es influir más que mandar. Admiro a los líderes que hacen que parezca fácil. Sin sobreactuar. Alcanzando la sencillez.
Hoy, en La Vanguardia

The most interesting stuff in LLMs right now (to me) is:

+ figuring out how to do it small
+ figuring out how to do it on CPU
+ figuring out how to do it well for specific tasks

Thrilled to introduce RefuelLLM-2, our latest family of LLMs built for data labeling and enrichment tasks. RefuelLLM-2 (83.82%) outperforms GPT-4-Turbo (80.88%), Claude-3-Opus (79.19%), Llama3-70B (78.2%) and Gemini-1.5-Pro (74.59%) on a benchmark of ~30 data labeling tasks:

One rather strange event of the past ten days: like ~600,000 Australians, much of my retirement savings is with a superannuation (think 401-k) firm called Unisuper

Unisuper uses Google Cloud to store their data. Well, ~10 days ago, Google Cloud *accidentally* *deleted* *their*

Google Cloud accidentally deleted a company's entire cloud environment (Unisuper, an investment company, which manages $80B). The company had backups in another region, but GCP deleted those too. Luckily, they had yet more backups on another provider.
theguardian.com/australia-news…

'Folks who communicate a no effectively are not the firmest speakers, nor do they make frequent use of the word itself. They are able to convincingly explain their team’s constraints and articulate why the proposed path is either unattainable or undesirable.' by Will Larson

Over the last few months, we've been sharing a series of blog posts with MantisNLP about preference alignment techniques. If you’ve missed out, it’s not too late! We’ve gathered all the blogs in this thread 🧵 ⬇️

CuMo: Scaling Multimodal LLM with Co-Upcycled Mixture-of-Experts

Outperforms SotA multimodal LLMs across various VQA and visual-instruction following benchmarks within each model size group

repo: github.com/SHI-Labs/CuMo
abs: arxiv.org/abs/2405.05949

¡Mañana nos vemos en el Taller práctico de AI Generativa! 🤖
Aprende sobre chatbots, tokens, atención, transformers, RAG, tools, ciberseguridad y mucho más

Crea tu propio PyBot en Python
🏆 ¡El mejor PyBot se llevará un premio!

🙏 Gracias a BBVA AI Factory y PyDataMadrid
Últimas

🔔the guy who invented the LSTM just dropped a new LLM architecture! (Sepp Hochreiter)

Major component is a new parallelizable LSTM.
⚠️one of the major weaknesses of prior LSTMs was the sequential nature (can't be done at once)

Everything we know about the XLSTM: 👇👇🧵