Greetings,

The first wave of AI were bots that could classify data. Generative AI is the second wave, it talks to us and creates things but it’s still not the Halo-like Cortana that does any task we ask. For that, we’re going to have to wait until the third wave – Interactive AI.

• The next wave of AI will be interactive bots that can carry out tasks for us, says the founder of Inflection – like Yann LeCun, he’s not afraid of LLMs destroying humans, and prefers to focus on practical issues like privacy and bias. - MIT Technology Review
  
  
• When you combine LLMs with self-driving AI, you get Lingo-1 – it’s a vision-language-action model (VLAM) that can explain in detail what it’s ‘thinking’ about while driving around and making decisions. - Wayve
  
  
• 60 potential genAI use cases across 6 major industries – the Generative AI Dossier from Deloitte is a great resource to get inspiration on how to use genAI in your business. - Deloitte
  
  
• Petals is a community-run, “BitTorrent style” distributed system where people combine GPUs to enable everyone to use and fine-tune large, SoTA open-source models like Llama 2 70B using just their local machines or Google Colab. - Petals
  
  
• Optimizing software from the ground up – Meta AI tuned the smallest Llama 2 model to optimize LLVM assembly code size, outperforming two SoTA baselines and generating compilable code 91% of the time. - arXiv
  
  
• Microsoft’s AI researchers accidentally expose 38 TB of private data when sharing open-source training data – a good case study of AI development risks with data oversharing that could easily lead to supply chain attacks.  - Wiz
  
  
• Base LLMs are limited to their training data, retrieval-augmented generation (RAG) is a way to overcome this and teach an LLM about a specific topic – see how it’s done in this guide for building RAG-based applications. - Anyscale

----

Found this helpful? Forward this email to a colleague.

Was this email forwarded to you? Sign up here to stay on top of AI news.