I think there is a good chance natural languages become more formal because of AI. AI makes formalism more accessible than ever. The next pidgin could be a natural / formal hybrid.

Expert led evals for LLMs are hot right now, Wiener had figured it out in the 60s that teaching AI via graded exercises was the way to transfer values and tacit knowledge to AI

Over the 10 days, I wrote a new Racket library for "expect testing", a style that Yaron (Ron) Minsky has advocated for in OCaml. github.com/samth/recspecs/ As an experiment, I built it entirely with Codex (the OpenAI async AI programming tool). I have some thoughts.

Notion of personalised proactive agents for much longer, this is Licklider and Taylor talking about Selfridge’s ideas from 1968

Start of the bring your own LLM era, next step is bring your own context

As the cost of UI and modifying workflows goes down, contextual context becomes the interesting place. When do you offer that discount to the customer? Which product do you try to upsell, it depends 😁

Cost of UI heading south 😊

It’s easy to imagine a near future where ChatGPT Voice Mode is more like “show and tell mode” and it live generates visuals that correspond to whatever you’re talking about. That would be a new medium. For example, when I ask about fixed points in RG, it would spontaneously show

Whats a minimal viable eval setup? Error Analysis + Notebooks are all you need (for a while) 1 of 3

If context engineering is the new hotness for engineering, context management is the new UI hotness/challenge for designers.

Most paths on AI for white collar workers lead to Knowledge Navigator like systems backed by domain specific models. My bets are on one of OpenAI, Cohere, and Replit ⠕ making the first release by end of 25. youtu.be/umJsITGzXd0?si…

Why is this feature not available in Google meet?

Where does your ‘work’ happen? Does the LLM come to your app or do the apps come to the LLM? With the OpenAI Agent, the negotiating hand of the classical software apps just got weaker. Next move from the Anthropic market place.

Not VSCode, its the Office folks who should worry about Cursor

The MDS saga on compressed time scales, consolidation ahead. techinasia.com/news/accel-lea…

People debated the value of the unit drive: motors attached to each machine in a factory as it was more expensive than group drives or the status quo of belts and shafts. We now debate the value of LLMs. History of industrial revolutions certainly rhymes.