Clay

Honestly CLAUDE.md barely does anything to manage pitfalls for me. I tell it not to maintain backward compatibility and multiple implementations, not to use fallbacks, I tell it to push back... and even with Opus it does the same things over and over again.

I think this is helping to manage sycophancy in Claude code:

If "three strikes" laws seem overly draconian to too many people, we could easily get ~the same outcomes with ten strikes

Anyone still focusing on wokeness as the threat to free speech is a fool. We literally have the president suing people and his goons directly targeting shows that criticize the dear leader.

Congress today held a hearing on whether to force states to sell pork from crated pigs. This is a wildly unpopular policy that Congress is likely about to enact via the Farm Bill. Today's hearing showed how Big Ag lobbyists get DC politicians to do their bidding. A 🧵...

Have people had any success modifying the claude.md system prompt to get Claude Code to be less of a sycophantic booster? I'm pretty sure I could propose both A and not A and get "that's a brilliant insight" in both cases.

Let me tell you, Grok is not happy about the enemies of Ataturk

In order to supervise an automation tool (or another person!) effectively, you need to be able to do the same job yourself. Doesn't matter if you rarely ever do the job yourself (like a manager who no longer codes), you need to be *able* to do it.

Texas officials have long worried about the dangers to children who packed into camps each summer along the flood-prone Guadalupe River. They knew that warning systems were inadequate. So why was nothing done? We have reviewed documents spanning a decade to find answers. 🧵 1/9

CLAAAUUUUDDDEEEEE!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

The most telling part of this case is when the Trump administration took the unprecedented step of releasing the full Kamala Harris interview. Once it was out, there was practically crickets—because it just wasn’t that bad. Networks edit interviews for clarity all the time.…

Kleiman used to observe that when a good has severe negative externalities, it is a policy success to have inefficient providers and bad market design. The exact reason we want state retail for bad things like liquor is why we want capitalism for good things like carrots & eggs.

I asked Oxford philosopher Toby Ord to explain 'The Scaling Paradox': AI 'scaling' is one of the least efficient things on the planet, with costs rising as x²⁰ (!). Also how OpenAI accidentally put out a graph suggesting o3 was no better than o1 — plus revealed their latest…

Maybe they’ve gone beyond the optimal amount of psychedelic use in tech?

Incredibly important cases. Training a stat model on a word order distribution across a corpus is clearly transformative fair use. Specific expression, not "style", is protected. *Using* those outputs can be a copyright violation, just as paying an artist to copy something is.

God help the doctors who are using 4o. Someone really needs to tell them.

Since medical doctors are likely using ChatGPT informally, could be that OpenAI's insane numbering system under which o3 is better than o4 has cost some lives.