AI: Weekly Summary

...week ending December 15, 2023

Welcome to the Saturday weekly AI summary here on ‘AI: Reset to Zero’. 

Here are some highlights from this week:

1. Microsoft & Mistral release smaller ‘SLM’ AIs: The momentum to ‘Small Language Model’ AIs continue, with both Microsoft and European’s newest Unicorn AI startup Mistral announcing open source ‘Small AI’ models. For Microsoft, it was an upgrade to its Phi models, the diminutive 2.7 billion parameter Phi 2, capable of running on mobile phones. And Mistral, the second best known AI startup after Hugging Face, announced new funding and their 7 billion parameter Mixtral, with ‘high-quality Sparse Mixture of Experts’, competing with OpenAI’s GPT 3.5 LLM AI. These developments accelerate the move to ‘Small AI’ I’ve been discussing for a while now. Deeper dives on ‘Smaller AI’ trend here and here.

   

2. Google rolls out Gemini Pro for enterprise: A week after last week’s big Gemini pre-release release, Google unwrapped Gemini Pro, it’s ‘middle’ next gen Foundation LLM AI, which sits between Gemini Ultra and Gemini Nano, for developers and enterprises to start building with Gemini Pro via APIs. Free for now, with ‘competitve pricing’ to come. Goes up of course against Microsoft/OpenAI GPT-4 for enterprise and others. Separately, Google also released ‘Duet AI for Developers’, their AI code assistant against Microsoft CoPilot for Github, Replit and others. Also released ‘Imagen-2’, their new text-to-image model. Busy AI  product update week from Google. My take on Gemini release again here.

   

3. European Union prints EU AI Act: EU got EU AI Act on the 2023 board, even though key provisions don’t go effective until 2025. A PR victory vs US and other jurisdictions on seemingly proactive AI regulations. Even though the Act was percolated pre-OpenAI ChatGPT a year ago. A victory in regulatory optics for now. Deeper take here.

4. OpenAI partnership and product updates: OpenAI also had a busy week, releasing ChatGPT Plus back again after pausing a bit earlier. Also announced their ‘big media’ partnership with Axel Springer, the European media conglomerate. Separately, there were percolating new rumblings on GPT-4.5 potentially out shortly. 

   

5. Stability AI tries ‘omni’ subscription model for open source AI: For those wondering about evolving business models for open source LLM AI models, Stability announced a 3-tier ‘membership’ set of subscription plans for its AI models. This holds firm their commercial rights over the core technologies, while generating subscription revenues for personal, mid-tier enterprise, and custom enterprise pricing. Expect to see this trend from other vendors as well. 

Other AI Readings for weekend:

1. New AI paper from Google Deepmind, ‘FunSearch’, focusing on discoeries in mathematical sciences leveraging LLMs. Focus on human supervised AI augmented with AI that is ‘more intelligent’ than just humans.

2. Another notable AI paper ‘Weak to Strong Generalization’, from OpenAI’s ‘Super Alignment’ team, was announced a few months ago by Chief Scientist Ilya Sutskever. 

3. I wrote a piece comparing industry expectations and eventual expectations around 5G with AI today. An ‘wait and see’ exercise in seeing if history rhymes.

   

Thanks for joining this Saturday with your beverage of choice. 

For your Sunday tomorrow, this week’s ‘The Bigger Picture’  is coming up. Stay tuned. 

(NOTE: The discussions here are for information purposes only, and not meant as investment advice at any time. Thanks for joining us here)