Hive’s Post

Want to see what Airtable is cooking up in AI? 🤖 (from Howie Lui, CEO of Airtable) We'll share in the newsletter tomorrow. 👉 Find out by signing up at builtonair.com/join

Great presentation by James Zhou (Stanford University ) on mixture of agents and text grad at hackathon organized by SambaNova Systems , Together AI , Numbers Station AI at AGI House. Multiple agents collaborating to achieve a task is becoming an increasingly important aspect to develop useful enterprise products! By combining multiple agents, they were able to beat #GPT4 on AlpacaEval leaderboard. These systems require switching between multiple agents and running them at extremely fast speed. This is where SambaNova’s SN40L chip really shines. 100s of models on a single node each at 1000s tokens per second. Try our apis at - 1. https://fast.snova.ai/ 2. https://lnkd.in/gM4G4gDA

Own your space and be an AI prenuers in that space." (Ubio Obu). Were you at the AI Prenuers event? If not, you missed out! But don't worry, I've got you covered. Check out our blog post for all the highlights and get inspired to be an AI-PRENUER! https://lnkd.in/dwVQN93E

What does 2024 look like for AI? Here are 6 highlights from Clem Delangue, CEO of Hugging Face:    1. 🍃 More emphasis on the monetary and environmental impact of AI. 2. 📉 One of the upcoming AI firms will go bankrupt or be bought at a low price. 3. 🤖 AI will be used to enhance jobs, not replace them (for now). Especially in specific domains, such as content generation. 4. 👐 Open source LLMs >= closed source LLMs. Giving organisations more options will become critical over the next few years. 5. 💡 More breakthroughs in videos, time-series (e.g. in Finance), biology and chemistry. 6. 👷♀️ Increase of AI developers. Agree or disagree, buckle up for another exciting year full of AI newsfeeds, memes, and innovation.

🗓️ Were you unable to attend our Intro to AI webinar last month? ⏯️ Not to worry- you can register to watch the full session on-demand anytime at https://lnkd.in/dn-CNDYQ 👇 Catch a sneak peek of what you're missing below:

Some very interesting predictions from Hugging Face CEO Clem Delangue 🤗 As our seniors and juniors approach the end of their High School career, many should be considering not only College, but what projects they can begin in order to “stand out”. Open source companies like Hugging Face can provide students with AI education and opportunities for practice on technology that they may not have access to currently. There are so many fields that will be effected by AI in the near future. Business, Science and Journalism are just a few.

My preferred prediction is the first one Clem Delangue 🤗 mentions. The equation of AI has 1⃣ quality (speed of inference, accuracu of content) on one side and 2⃣cost (economic and environmental) on the other ; and this equation can be optimized big time.

I agree to most of what Clem Delangue 🤗 says in this video. However, I think that Huggingface should invest more in curating their platform. I know it is a fine line between censorship and anarchy. Nevertheless, huggingface is becoming less and less helpful due to the sheer number of models and datasets. Only a few of the thousands of models are really helpful. Most are just experiments with no public value and for interested users it is becoming harder and harder to find the gems among all the stones. Maybe AI can come to the rescue here as well. We need good model cards with working code including requirements, so that running the model on premises becomes easy. We need information about the prompt templates to use and ideally tests on a large number of standard benchmarks. Open LLM leaderboard is a good start here, but it could use a lot more benchmarks (which needs an awful lot of processing time to run). We also need transparency regarding datasets used for finetuning and foundation models, the current model is based on. Polluted training data to climb the leaderboard seem not uncommon today. There are some good examples, where all these information are readily present in the model card, including tags for languages, ML tasks and so on. However, the majority of models has no model card at all. Let's work together to build high quality models including documentation instead of competing for being on top of the leaderboard with just 0.01% better scores.

The easiest way to deploy a private multi-modal API with the new Pixtral model by Mistral AI in <30 lines of code 🤯 Take it for a spin ➡ https://lnkd.in/eRKtTu5r

"But having access to ISAIC, to cover those 3 areas was a critical essential for our business." - Geoffrey Shmigelsky from One Cup AI. Learn about ISAIC's latest panel discussion at