I created a self-referential game engine that uses AI to evolve games! Humans & agents play, and the games create & edit themselves. (Finally let your @openclaw agent have some fun.) Below is the AI playing a game it created where you play a bouncer in the famous Berlin

he terminated my dad who was auditing lifesaving aid (zidovudine for mothers in sudan to prevent HIV transmission). I bet this fool doesn’t even know that that medicine exists

How good are the models at making and playing games? Technically, very proficient; good at balancing game economies; creative at transferring mechanics between themes. But without guidance, they tend to optimize into a particular direction. Here is an experiment: I started with

These games are living things: because they fit into my game engine, the AI can constantly add new game mechanics and content. Below is the loop that runs. Every game play feeds back into improving the game. The game design agent is cycled between (1) adding new mechanics, (2)

I love the idea of using self-evolving games for educational purposes. The game design agent first plans and does web searches to best encode a theme into game mechanics. Watch this game that the AI created to teach quantum computing! The AI figures out how to build a quantum

How does it work? Game states are 100% text, but visualized through defined layout primitives. That lets the engine snapshot game state easily. All game logic runs in isolated Lua sandboxes. All games have a "heartbeat" that executes periodic updates. Games talk to the player

This is well said, heartfelt and true. Thrive is a gift to founders & all founders themselves. Congratulations @JoshuaKushner & team

Shifting structures in a software world dominated by AI. Some first-order reflections (TL;DR at the end): Reducing software supply chains, the return of software monoliths – When rewriting code and understanding large foreign codebases becomes cheap, the incentive to rely on

I wrote a self-referential engine that can create AND play games via AI & evolve them. Watch @claudeai Sonnet win the Cold War in a simulation game running in the engine (top right shows the agent thinking about how to play and the game's responses) More on the engine soon!

In ~10 years, Oscar went from non-existent to neck-and-neck being the largest ACA health insurer in the country, growing revenues at 61% to $18.7-19B into 2026, and increasing market share in our footprint to 30%. Incumbent insurers are mostly useless. But better,

My mother and my sister are nurses, I remember the nurses who helped give birth to my kids more than anyone, and nurses run our healthcare system. I have no idea how anyone could look at the killing of the VA nurse Alex Pretti & say it is justifiable by anything. Shame.

If you are signal boosting Cremieux, you are part of the problem. https://t.co/aYoXIxGKi3

I get that people have mortgages, private school bills and vacations to go on and pay for. But how any self-respecting engineer can still work for a company of a guy who amplifies garbage like this is beyond me. At this point I'd be glad to wait 5-10 years longer for another,

Uber driver? This guy is a hardcore lawyer 🤣🤣

This is completely not okay, and we can't become numb to repeated instances of illegal and unconstitutional action by government agencies. The recent days have been horrific.

Thrilled to see Oscar land top ranking in “The Book” for 2026! 🎉 This ranking recognizes Oscar as one of New York’s “Largest Public Companies,” a significant milestone that reflects our continued growth and stability. 👇 Check out the full list from @CrainsNewYork

There is ancillary work that also spikes ahead of an IPO that is indirectly related, the biggest one is probably compensation. Goes up to 9% of logged working hours in the months before the IPO. It's just a cool non-classical Claude Code use case & super fun for me 4/4

So, for example, here is the total amount of hours that a founder-CEO spends directly on taking a company public: 470 hours, between 20% and 40% of logged working hours for 5 months. (OSCR went public on March 3, 2021) 3/4

The agent can of course run any kind of query and turn it into a web page with observations. But it can also spawn lots of sub-agents to evaluate blocks of time spent and categorize it (like "was this a creative activity"), then persist the categorization. 2/4

My fun use case for Claude Code: since 2012, I've been logging what I do on a minute by minute basis in spreadsheets. (Here is a random day in 2022.) It's a 56,000 row table. I now have a Claude Code agent with various Skills definitions that can process & visualize it all. 1/4