“Learn to build a local agentic RAG application for report generation using open-source LLMs! 🚀 Our friends at @AIMakerspace are hosting a live event next week (November 27) to teach you: 🔧 How to set up an “on-prem” LLM app stack 📊 LlamaIndex Workflows 🤖 Llama-Deploy 🏢 and
“I made an open source version of @AnthropicAI’s computer use. It automates a computer GUI with just the mouse, keyboard and screen. Built with: – @Alibaba_Qwen 2 VL – OS-Atlas by @zywu_hku – @E2B_dev cloud sandboxes
“🤯 Mind-blown! Just built a complete flashcard web app in less than 30 seconds using @Qwen’s new Coder demo! Like Claude’s artifacts but open source. One prompt = full web app with cards flipping. Try it:
“Let’s gooo! Tülu 3 – 70B & 8B by @allen_ai is OUT!! Competive with Claude 3.5 haiku, beats all major open models like Llama 3.1 70B, Qwen 2.5 (except MATH) and Nemotron! 🔥 Best part: All their recipe – code, datasets and model checkpoints are public and out in open! – Anyone
We can all be AI engineers – and we can do it with open source models
“🔗 Agent Protocol: interoperability for language agents LangGraph is a multi-agent framework. It can communicate with agents defined in other frameworks and languages. Today we are releasing: 📱 A standard interface for agent communication (Agent Protocol):
elvis on X: “Bi-Mamba: Towards Accurate 1-Bit State Space Models Presents Bi-Mamba, a scalable 1-bit Mamba architecture designed for more efficient LLMs with multiple sizes across 780M, 1.3B, and 2.7B. Bi-Mamba achieves performance comparable to its full-precision counterparts (e.g., FP16 https://x.com/omarsar0/status/1858878654736199850
“Bi-Mamba: Towards Accurate 1-Bit State Space Models Presents Bi-Mamba, a scalable 1-bit Mamba architecture designed for more efficient LLMs with multiple sizes across 780M, 1.3B, and 2.7B. Bi-Mamba achieves performance comparable to its full-precision counterparts (e.g., FP16
OpenScholar: The open-source A.I. that’s outperforming GPT-4o in scientific research | VentureBeat
“more fun open-source research news – new paper drops (nGPT) – claims 4-20x training speedup over GPT – shocking – very cool – very valuable – community tries to reproduce – doesn’t hold up – turns out baseline was busted – another cool new research idea oneshotted by github anon
DeepSeek
“🔥The competition for the best reasoning LLM intensifies! A few days ago, we had the Forge Reasoning API, now we have DeepSeek-R1-Lite-Preview which produces o1-preview-level performance on math benchmarks. Here are my observations after some initial tests on Deepseek’s new
“🌟 Inference Scaling Laws of DeepSeek-R1-Lite-Preview Longer Reasoning, Better Performance. DeepSeek-R1-Lite-Preview shows steady score improvements on AIME as thought length increases.
“Less than 48 hours ago, DeepSeek AI from China just dropped their AI reasoning model. And it’s on par with OpenAI o1-preview. Major shift. 10 examples (and how to try):
“Rumor is that DeepSeek R1-Lite is a 16B MOE with 2.4B active params if true, their MATH scores went from 17.1 -> 91.6
DeepSeek-R1-Lite-Preview AI reasoning model beats OpenAI o1 | VentureBeat
“🚀 DeepSeek-R1-Lite-Preview is now live: unleashing supercharged reasoning power! 🔍 o1-preview-level performance on AIME & MATH benchmarks. 💡 Transparent thought process in real-time. 🛠️ Open-source models & API coming soon! 🌐 Try it now at
“With the preview of @deepseek_ai R1 and results equal to @OpenAI o1-preview, you might want to take another look at “Stream of Search”. ‼️ When you test R1 “thoughts” are streamed, meaning that there is no MCTS used during inference. From looking at some of the thoughts they” / X
Grok
“Our frontend team for Grok in X just doubled. It’s now 2 people. Watch out.” / X
Hugging Face
“Our Omnivision-968M is ranked 3rd on @huggingface ‘s trending model list! It is already bringing powerful multimodal capabilities to the latest @AMD powered hardwares. Small models for the win!
“HF posts are becoming the best place to post AI news & updates. You can now check who’s getting most visibility on their posts:
“Biggest open text dataset release of the year! 🚀 SmolTalk is a 1M sample big synthetic dataset that was used to train SmolLM v2. It is available under Apache 2.0 and combines newly generated datasets with publicly available ones! 🧬 TL;DR; 🧩 New datasets: Smol-Magpie-Ultra
“An ordinary day @huggingface: – We released SmolTalk, a 1M-sample synthetic dataset used to train SmolLM v2 – We released Observers, a Lightweight SDK for AI Observability – Qwen2.5-72B is now the new default model for HuggingChat – We dropped a new “recent activity” feature on
“Open Source AI With SambaNova & Hugging Face Tuesday, December 10 5:00 PM – 8:30 PM PST Join Sambanova and Hugging Face for an Open Source AI night to remember with Silicon Valley’s best AI minds rsvp:
Judge Arena – a Hugging Face Space by AtlaAI
Meta/Llama
“AI2 @allen_ai just released Tülu 3, a new family of LLMs 🦙 Based on Llama 3.1, so Llama 3.1 license 🦖 Comes in 8B and 70B 🔖 Aligned using SFT, DPO, and a new technique they created: RL with Verifiable Rewards (RLVR) Load and use easily @huggingface transformers 🤗
“Cerebras is capable of offering Llama 3.1 405B at 969 output tokens/s and they have announced they will soon be offering a public inference endpoint 🏁 We have independently benchmarked a private endpoint shared by @CerebrasSystems and have measured 969 output tokens/s, >10X
[AINews] Llama 3.2: On-device 1B/3B, and Multimodal 11B/90B (with AI2 Molmo kicker) • Buttondown
Llama 3.1 405B now runs at 969 tokens/s on Cerebras Inference – Cerebras
Mistral
“Pixtral Large 124B
“@MistralAI is growing! We opened a new office in Palo Alto, CA! We are now hiring across research, engineering and business roles in US. Join us:
“open weights Pixtral Large with 124B params ✨ we are so back” / X
Mistral unleashes Pixtral Large, upgrades Le Chat with image gen | VentureBeat
“Yes we now support image generation on @MistralAI Le Chat, powered by @bfl_ml. And did I mention it’s free?” / X
pixtral-large is now in anychat
“⬆️pip install -U vLLM You can run Pixtral Large with vLLM today!
“At Mistral, we’ve grown aware that to create the best AI experience, one needs to co-design models and product interfaces. Pixtral was trained with high-impact front-end applications in mind and is a good example of that.” / X
Pixtral Large | Mistral AI | Frontier AI in your hands
Release 1.5.0 – Mistral Tokenizer v7 (new System Prompt + Fn calling) · mistralai/mistral-common
Mistral has entered the chat | Mistral AI | Frontier AI in your hands
“Today, we are announcing two new exciting updates: Pixtral Large: Frontier-class 124B multimodal model, powering the new Le Chat. Brand new Le Chat: With web search, canvas, image-gen, image understanding & more- all for free! 1/3
Mistral unleashes Pixtral Large, upgrades Le Chat with image gen | VentureBeat
“Yes we now support image generation on @MistralAI Le Chat, powered by @bfl_ml. And did I mention it’s free?” / X
NousResearch
“Expanding from a science company to a science and product company was no easy task, and that release is a very significant milestone in our journey. We’re looking forward to how you’ll use le Chat, now a slightly more mature animal” / X
Introducing the Forge Reasoning API Beta and Nous Chat: An Evolution in LLM Inference – NOUS RESEARCH
“Today we are launching the Forge Reasoning API Beta, an advancement in inference time scaling that can be applied to any model or a set of models, for a select group of people in our community.
Qwen
“🚀 @Qwen just dropped 2.5-Turbo! 1M token context (that’s entire “War and Peace”!) + 4.3x faster processing speed 🔥 Check out the demo:
Snowflake
Snowflake shares surge on rosy forecast, AI deal with Anthropic
We’re proud to partner with Hyatt to help them deliver a best-in-class experience to their guests and drive great business outcomes. With SnowflakeDB they’ve been able to:
➡️ Reduce time spent accessing and managing data
➡️ Unify data into one consolidated platform
➡️ Collaborate easily across disparate functions
➡️ Launch applications and innovate quickly
Leave a Reply