A week in Generative AI: Claude-3, Sergey Brin & Perplexity
News for the week ending 10th March 2024
The big news this week was Anthropic’s launch of their Claude-3 family of models. There wasn’t any news of note from Google this week (a first for 2024!), but there is a great video of a Q&A with Sergey Brin, that I highly recommend watching.
Anthropic release Claude-3
Anthropic has released the Claude-3 model of families, which now seems to be the trend - don’t just release one model per generation, release a family of models with different sizes, capabilities and use cases.
Claude-3 Opus (the most advanced model) has been shown by the accompanying research and test results to outperform GPT-4 and Gemini 1.0 Ultra across many domains. This is great to see, but according to the Chatbot Arena it currently has an ELO score of 1233 vs 1251 for GPT-4 Turbo so it’s still rated as a ‘GPT-4 class’ model.
However, this research claims that Claude-3 is the first AI to pass an IQ of 100 for the first time, which is very impressive!
Time will tell, but I still don’t think we’ve seen a truly ‘next gen’ model released yet. With all of OpenAI’s rivals having recently released new models, it’s probably over to them to move the game on again when they release GPT-5 sometime later this year.
Sergey Brin talks AGI, Gemini 1.5 Pro and the future of search
I haven’t seen Sergey Brin talk much in public about AI, but I had heard that he came out of retirement so he could be more actively involved at Google over the last couple of years, purely focused on this area. This is a rare opportunity to see Sergey talk about AI, what Google is up to and answering some really great questions from the audience. Worth a watch.
OpenAI announces new board members, reinstates CEO Sam Altman
So, the independent review is out and there’s no real news or any additional drama following up on Sam Altman’s ousting and reinstatement as OpenAI CEO last year. Sad to see a last minute bit of mud slinging aimed at Mira Murati in the NYT from former board members. I followed the events at OpenAI closely as they were unfolding last November and from my perspective Mira acted with absolute honesty and integrity with her loyalties clearly with OpenAI and its mission first.
AI startup Hugging Face is looking into open-source robotics
It’s great to see Hugging Face start to expand their open source philosophy to robotics, and Remi Cadene is a great hire from Tesla to lead this. Really looking forward to seeing what they do!
Perplexity Poised To Become Latest AI Startup To Hit Unicorn Status
Perplexity is trying to take on Google Search by building a search experience on top of GenAI technology from scratch. Great to see them getting more funding, we’ve been waiting 25 years for someone to truly challenge Google’s dominance in search, maybe Perplexity will have more luck than its predecessors!
ChatGPT can now read its answers out loud
OpenAI has announced a new feature where you can get ChatGPT to read its answers out loud in 37 different languages. Just another small step to true multi-modal capabilities.
AI Ethics News
Long Reads
Demis Hassabis - Scaling, Superhuman AIs, AlphaZero atop LLMs, Rogue Nations Threat
Yann Lecun - Meta AI, Open Source, Limits of LLMs, AGI & the Future of AI
CNN - Move over, artificial intelligence. Scientists announce a new ‘organoid intelligence’ field
Wired - The Mind-Blowing Experience of a Chatbot That Answers Instantly
“The future is already here, it’s just not evenly distributed.“
William Gibson