AI Optimism, Silicon Valley Tax, & Self-Play RL for AI Agents

Americans are getting way more optimistic about AI than the media narrative would make you think. The interesting part is who is most bullish: younger people, men, and racial minorities are leading the optimism, and the biggest fault lines are age, gender, and race. Sample size was massive too, roughly 30,900 respondents, fielded Dec 1 to Dec 10. 

A lot of this vibe shift is probably because the early “AGI any minute now” timelines that some corners of tech pushed have not played out the way people expected. When the world does not instantly flip into sci-fi, people stop bracing for impact and start thinking in terms of tools, leverage, and opportunity.

Verification becomes the real bottleneck in the Age of AI

Martin Kleppman, the author of Designing Data-Intensive Applications, states:

As models get stronger, the winning move is less about generating more content and more about proving what is true. The advantage shifts to anyone who can build tight feedback loops: tests, checks, audits, evals, provenance, citations, and systems that force AI outputs to earn trust. In a world where everyone can produce answers, the scarce asset is “this is correct” and “this holds up under scrutiny.”

Self-play RL for AI agents is a huge step toward autonomous improvement

The Self-play SWE-RL line of work is basically pushing agents to create their own training signal by iteratively injecting bugs and fixing them inside real codebases, rather than relying on humans to curate tasks. That matters because it points toward compounding capability: agents generating experience, learning, and leveling up with less human babysitting. The paper reports measurable gains on SWE-bench style benchmarks, which is exactly the kind of “agents get better by doing” loop people have been waiting for.

Ro Khanna proposes an idiotic policy: wealth tax aka a 5% unreunrealized gains tax

https://x.com/PalmerLuckey/status/2005200055846723876?s=20
This proposal is one of those policies that sounds clean in a tweet and gets chaotic the moment you map it onto reality: liquidity issues, valuation games, incentives to move assets, weird second-order effects for founders, investors, and anyone sitting on long-duration equity. Palmer Luckey, who is absolutely correct in this case, is going after Khanna hard on this

ChatGPT and AI is still the fastest consumer tech shift we have ever seen

The curve speaks for itself!