🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

techcrunch.com/2024/08/25/ste… And I agree, when discussing some of the problems/opportunities in the field of NLP it becomes evident that ML researchers have missing pieces. At PNYX we have been working with philosophers since March, and the synergy is great, we are cooking big stuff

🤖Simposio Cientifico de Inteligencia Artificial y Aplicaciones 🤖 Desde la escuela de Ingenieria en IA y Matematica estamos organizando la segunda edicion del SCIAA Universidad de San Andrés: se busca promover métodos y desarrollos de IA y ML en diversas areas. Mas info udesa.edu.ar/sciaa

Open models are vital for research and development (specially for small teams/budgets). Skynet is nowhere near…

POKT Network incentives kicked in on Aerodrome, 750 % APR? 🤯

I hope that the first paragraph is sarcasm… Otherwise this is looking bad for AI progress

📢 New Listing! $POKT is now on Uphold. #POKT powers the POKT Network, which aims to address the growing demand for decentralized RPC services. Pocket Network has processed over 700B relays via 13K+ nodes since launching in July 2020. 🔗 Learn more: uphold.com/prices/crypto/…

dear Open at Microsoft see how 4 or 8 are not among the common factors of 10 and 40, which are 1, 2, 5, 10. Maybe it was not a good idea to use this for Phi-3-medium: `"num_attention_heads": 40, "num_key_value_heads": 10` Great model tho, but I need bigger GPUs...

We are cooking 🧑‍🍳

Por si pensabas que despues de FORMA y la firma de la constitución de Cardano Foundation no habia mas nada en Buenos Aires, miren lo que apareció 😱

And this is what real AI/ML looks like, congrats!!

seeing the "epiphany" moments of DeepSeek R1 generation is really funny and also interesting: `wait, what about the various test cases? Let's test it mentally` I don't think there is any "epiphany" there, but it kinda shows how they managed to guide CoT training

Nice paper, very well written and clear of any crypto-AI hype. Looking forward whats next, so far is a very clever way to reduce the overhead in verifiable ML!

This has been happening since gpt3. They don’t care because people think that ”vibe” is the way to select a model. Model’s capabilities should not be assessed by the same people that want to sell them, we need 3rd party testing, continual testing .

And you thought lm-arena was fair, the vibe must die… Great thread, I have a reading for today’s coffee!

Now, think of a permissionless version of OpenRouter, where anyone can place an anonymous model, with community driven generative benchmarks, mixing and hiding tests into users traffic... That's were we are going with the POKT Network , just wait a month...