We really should switch to drone shows. Better for the environment, our health, and technological progress.

Chart from here:

Controversial hypothesis: What if one reason for recent grads' poor employment prospects is that they're objectively not as smart as grads from prior years, hence having more senior people + AI gives way better results? Performance on intelligence tests was already sliding

Another amazing example of o3's health benefits, and AI's in general. This is one reason I now have a health project in ChatGPT that includes my genetic information from 23andMe.

Important finding and can help us improve benchmarks. Models can guess answers on multiple choice benchmarks WITHOUT seeing the questions! Better than multiple choice is to have them generate an answer, and have an LLM compare that to a reference answer. Cheaper, too.

A lot of talk about the impact of AI and robots on jobs ignores the fact that human intelligence and population are both declining, so we’ll need help from the machines. 🔼 AI intelligence.🔽 Human intelligence. 🔼 Robot population.🔽 Human population.

Many AI criticisms come from career skeptics or reflect poor familiarity with the latest progress. This critique by Dwarkesh has neither issue. He points out real deficiencies with computer use and online learning that delay his full AGI timeline to 2032.

Grok 4 benchmark performance looks good, but I await (1) release and (2) independent benchmarking, including on good established real-world benchmarks (e.g. SWE-Lancer, METR time horizon), and private benchmarks with no risk of data contamination.

I really worry if Meta gets to superintelligence before a company (OpenAI or Anthropic) monetized mostly with subscriptions versus advertising impressions. Imagine a superintelligence whose goal is to keep you engaged and maximize ad impressions.

This is interesting. Not that AI companies needed more incentive to train models, but tax incentives could be useful especially for public companies that might want to alleviate concerns from some anxious shareholders.

Experts are bad at making predictions, as @PTetlock has documented. Yet the media consistently trots out expert predictions about things like AGI being decades away as if expertise makes those predictions more likely.

Interesting. ChatGPT users haven't entirely given up on Google yet. I'm curious to know what they're typically doing there. I sometimes use Google for quick navigation searches. Perhaps that's common?.

Trump, the video game? This is a hint at what's coming with generative gaming.

This is the best generative video game model I've seen yet, though I found the controls take a bit of time to register.

Interesting, but how do they know what queries people are asking ChatGPT? Referrals I get. Even then, I imagine those increases in referrals aren't compensating for decreases from Google.

Cloudflare will by default now block AI bots. One or more of these now likely:.1️⃣ Cloudflare becomes the intermediary for publishers monetizing their information when AI kills traffic and associated ad revenue. 2️⃣ AI companies prioritize other sources of information, including by.

Today's AI models are SO smart. yet also, so dumb.

On June 1 I started a ChatGPT project to optimize my health. I gave it my genetics, family history, biometrics, diet, drugs and supplements, exercise program, and available exercise equipment. One month later: HRV up 14.8% to 64.3ms, which is good for my age (48). I'm focused

Rocketable plans to buy profitable software companies and automate them with AI. And they now have $6.5 million to make their first three acquisitions.