Friday Links: Learning by copying, AI Smell and Voice grants from ElevenLabs

This is a shorter dispatch than usual because it’s Slush week and there were lots of in-person meetings. (The event has been amazing - I’d highly recommend it to both startups and investors + Helsinki has been amazingly welcoming in November!). On to the links:

• AI Agents can copy humans. Part clickbait title, part helpful insight. Google DeepMind research on AI systems observing the actions of others and copying them shows that learning can jump ahead in certain scenarios. This isn’t terribly surprising, but it’s an important factor in how systems will train in the future. Collections of AI/Robotic systems will also co-learn behaviors (there’s plenty of research on this.). An important caveat is that learning like this risks groups of AI agents falling into sub-optimal patterns that they might not use if they learned from first principles.
• Mapping out smell molecules for AI. A big limitation of any digital system is that it can only act effectively on what is already digital. Smell is still a barrier, so it’s cool to see work on capturing smell information as data.
• Amazon’s Q - The Chatbot for your company data. One of the most critical considerations for AI startups and investors is “Where is the data.” Amazon just underlined how important this is again with the launch of Q which is a bot system you can customize with your own company data. If done well, this could make life challenging for many of the new companies setting out to do the same thing. On the other hand, it may end up being too generic or general to be truly great. Companies like Notion and Interkom may be in the best spot: close to the data, but data with a specific purpose.
• Eleven Labs offers Startup Grants. The AI platform wars are underway in earnest. If you’re starting an AI Voice project, now might be a good time to start if the ElevenLabs platform is one of your choices. The more prominent players in each AI infrastructure space need to build a customer base and have investor money to do it. While you can say it’s a cynical play, it actually helps the ecosystem a ton to put the horsepower that is now available to use and see what works.
• Reshaping organizations for AI. Nice post from Ethan Mollick which talks about starting the process to adapt teams and organizations to AI. Processes will change and team needs will change. Some change will be painful but if it doesn’t happen more agile teams that started “AI native” might end up forcing the change anyway.

As an extra free link: Christphe Janz on the right opportunities in AI (keep in mind Des Traynor has a strong book to talk from Intercom’s point of view :-)).

Have a great weekend!