How do you pass time while your AI responds?

Happy Monday and welcome to your weekly ceremonial dose of AI updates.

The AI news cycle is relentless, and in this industry, not a day goes by without headlines, breakthroughs and shifting paradigms. Everyone I talk to agrees that it’s next to impossible to keep up, yet all agree that tracking the state of art helps make sense of where are (rapidly) heading. If only AI could help us keep track of itself.

This week’s edition is short but deep. I have little to add, but can wholeheartedly recommend the links below.

Until next week,
MZ

P.S. If you’re in Amsterdam, don’t miss our meetup next week at the Embassy of the Free Mind.

---

---

Quick Links:

• Vibecoding is mainstreaming (NYTimes).

• Everyone is doing Deep Research (You.com).

• Looks like GPT 4.5 performs well on things that are difficult to quantify. Only GPT gets close to replicating the line art.

• This is so trippy: next-gen voice models. Still feels uncanny, but less so than before. (Sesame)

---

How I use LLMs (2h)

Long and very hands-on video by Andrej Karpathy about his use of LLMs - loads of use cases and practical applications.

---

Deep Research and RL (30 min)

Lots of little insights about where OpenAI is taking their deep research tools.

Probably the number one lesson of machine learning is: you get what you optimize for.

---

Deep Q&A with the Sakana.ai team (1h40min)

Impossible to summarize, worth a listen. The future of science is artificial.

---

Testing AI Interfaces (36 min)

YC video testing different approaches for presenting AI in various services, including voice, chat and canvases. I’m always interested by the UX around AI.

---

You are receiving this newsletter because you signed up on envisioning.io or Substack.