The Signal: Meet Figure 02, the great AI talent shuffle, and Project Strawberry

Hey friends 👋 Happy Sunday.

Here’s your weekly dose of AI and introspection.

AI Highlights

1. Meet Figure 02.

This week Figure announced Figure 02. It is the world's most advanced AI humanoid robot capable of conversations with humans (speech-to-speech reasoning) using custom AI models trained in partnership with OpenAI.

Alex's take: This isn't science fiction anymore. It's happening now in factories and soon in our homes. For a company that was started only 18 months ago, they are building hardware faster than most people are building software.

2. The great AI talent shuffle.

There is an exodus of top talent at OpenAI. OpenAI Co-Founder John Schulman is leaving for rival Anthropic, and OpenAI's President, Greg Brockman, announced a sabbatical until the end of the year. This follows recent exits of other key figures, including AI safety leader Aleksander Madry, chief scientist Ilya Sutskever, and founding member Andrej Karpathy.

Alex’s take: It's like watching the All-Star team suddenly switch sides mid-game. Anthropic just got a major power-up in the AI alignment department, while OpenAI is left wondering who turned out the lights.

3. Project Strawberry.

Sam Altman recently dropped a cryptic tweet featuring a strawberry which further hints at OpenAI’s ongoing projects. Coincidence? This could be the first tease about 'Project Strawberry' or the mythical Q* AI breakthrough we've all been whispering about.

Alex’s take: OpenAI has been the market leader in AI breakthroughs. Needless to say, they’ve had a rough first half of the year. If they’re holding something in their back pocket, now would be a good time to release what’s next. We might see another quantum leap in AI capabilities.

1 Article I Enjoyed

Disease prediction with AI.

I recently came across a fascinating study on using AI for disease prediction.

It uses a ‘SEV-EB algorithm’ which helps doctors prioritise healthcare information such as blood pressure and cholesterol levels to make more accurate and confident diagnoses.

It boasts 95% accuracy in predicting disorders like heart conditions, diabetes and cancer.

While it's definitely not quite a crystal ball, it's a compelling step towards better patient outcomes and more personalised treatment plans.

I’m very excited about the future of healthcare, so I’ll be following this one closely.

1 Idea I Learned

Tipping LLMs.

I’ve found offering LLMs tips, such as adding, “I'm going to tip $10 for a perfect solution!” to the end of your prompt, to be quite helpful in increasing response performance. This performance increase could be up to 11%.

Here’s some more proof. It turns out that maybe $10 is the optimal tipping sweet spot.

I’ve also found that adding personal context, like “I’m struggling with this and would really appreciate your help; it would mean a lot to me,” has also increased the perceived quality of answers.

Prompt engineering is turning more and more into behavioural science.

So next time you use ChatGPT or Claude, try some honest tipping or infusing some more human emotion.

You might just get a better response.

1 Quote to Share

Tim Cook on problem-solving:

“When I don't have experience with something, I make a list of the people that I think that are the smartest people that I can contact to get advice.”

More than half the battle is won when you surround yourself with the smartest people.

It’s the who, not the how.

Source: CNBC’s interview with Apple CEO Tim Cook

1 Question to Ponder

Do you have a question about AI that you’ve always wanted to ask?

Have a ponder and reply to this email with your question.

I’ll pick one to answer next week.

💡 If you enjoyed this issue, share it with a friend.

See you next week,

Alex Banks

P.S. Jeff Bezos’ response to crisis