🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

The people who are making these tests are strongly primed to expect the self-interest explanation, and may be willing to entirely overlook how hard they are pushing. But when I’ve looked into findings like this before, the actual text is a lot less spooky than the headline.

When lawyers first started getting yelled at by judges for citing hallucinated case 2y ago I naively assumed word would get around and they would all quickly learn not to This DB has 116 cases from 12 countries where this happened, 20 of them this month! simonwillison.net/2025/May/25/ai…

The Met, MI5 and the National Crime Agency are opposed to the government's plans for early release. It would be a "net detriment to public safety". Extraordinary to see them release a statement like this. The threat is real and Crush Crime will fight this.

As a bit of trivia, before flow matching / rectified flows etc in late 2022 the core idea existed about a year earlier in late 2021 Non-Denoising Forward-Time Diffusions by Peluchetti openreview.net/forum?id=oVfIK… shows the role of marketing in the academic publishing game

2. Eliminate crazy high income tax marginal rates. When people face marginal tax rates of 60%+, they often choose not to work. Sometimes our tax system creates marginal rates of over 100%. taxpolicy.org.uk/2024/10/17/ref…

ALL OF THIS IS JUST NONSENSE but no they didn't even bother looking at the outputs The models literally recite the algorithm in their chains-of-thought, in plain text and in code. As I have explained in the other post, the steps of the different games do not have equal

The "reasoning doesn't exist" Apple paper drives me crazy. Take logic puzzle like Tower of Hanoi w/ 10s to 1000000s of moves to solve correctly. Check first step where an LLM makes mistake. Long problems aren't solved. Fewer thought tokens/early mistakes on longer problems. 1/11

Noise pollution is a huge problem

The illusion of reasoning… in product design decisions (that seem to be made without much thought)

This is not a great take by Gary Marcus, and I want to argue this very explicitly: 1. Gary Marcus claims that Lisan al Gaib generally said that the correct answers would take too many tokens to express. 2. Someone else (Colin Fraser) used some other unrelated prompt for Tower of

Is it just me or has o3 become dumber over the last week or so

We should normalize giving honest feedback to people. The person receiving the feedback does not benefit from sunshine.

The frustrating thing is, this is actually quite easy to fix. Founders will list if the incentives are strong enough—even with lower liquidity and valuations. - 0% capital gains tax for tech founders who list - Perpetual 20:1 super-voting rights - No CEO compensation disclosure