niobe

Completely Claude written FWIW. I recongise the style.

I felt it had been enshittified 1-2 weeks back, not in Feb. But it's very subjective.

Steeped in confirmation bias.. the whole article - and apparently author's methods - are written from the point of view of trying to prove that Satoshi is Adam Back. This cannot be trusted, no matter how many times Back is mentioned in a single article

Very cool

Magnus has really become the master at this. If you look at his best classical or rapid wins (as in most interesting or novel) of the last few years, they often follow a similar pattern:

- pick a playable but rare line in a solid opening - go out of book early - just scrape through the middle game with often questionable moves - later, sacrifice material for positional/ strategic advantages - rely on his greater 'pure' chess genius to close it out.

The idea is to avoid deeeply studied "safe" lines where it's possible the opponenet has deeper knowledge and then be slightly less off-balance than your opponent throughout. This is all in deep contrast to how the GMs played before Deep Blue where it was all deeply conservative well studied lines. I strongly suspect Magnus, who is known to have an incredible memory for games, keeps a mental database of these obscure lines and plays them only once!

A workaround for their own badly thought out permissions system? More likely another internal fix they decided to release.

journalctl anyone?

This is for co-ordinating instances of Claude or Codex, not something you do inside each instance.

Very nice. However, I do like to read every agent summary before letting them move on. I'm not sure I'd be able to apply this level of automation to many tasks.

I had to reread the first paragraph several times before I understood - the author was misuing a term.

> unexpected-behaviour.exe.dev

That is not a URL, that's a fully qualified domain name (FQDN), often referred to as just 'hostname'.

But that's one of the first things you fix in your CLAUDE.md: - "Only do what is asked." - "Understand when being asked for information versus being asked to execute a task."

But is anthropic trying to solve it? The current permissions solution is unbelievably poor for a product with this much traction.

Sensitive lad.

You can also turn off tips, but that's quite handy to learn when a new feature pops up

The entire kernel on every arch is 40 million lines, but the kernel running on your desktop is probably less than 2 million of those lines.

it's more the way they do them.. you've used them right?

The US is such a shit show. Personally I hope this doesn't affect Anthropic's growth and development because I quite enjoy using their products and see them evolve.

I find it very intuitive as is

it almost sounds like a personal attack on Peter Steinberger

Here’s an idea: What would happen if you changed a bunch of different bytes in an AI’s model file?

paywalled

HEADLINE: Electromagnetic radiation can be used to see!!!!

Right, that's what your eyes do. Radio is much longer wavelength than visible light (~5-10cm). So at best it offers extremely crappy resolution unless - you're doing something clever with second order information.

That's not impressive. That's an incredible amount concentrated in the hands of a few looking for a place to live. It has to end up somewhere. Some of it goes everywhere.

can we not come up with a way to separate the control and data planes in LLMs?

I think that's the kind of intuitive decision that comes from years of troubleshooting experience. It's not obvious that would be the place to start. It's impressive to me at least he got there.

Utterly brainless article. Why am I even commenting.

Real risk of them blocking or crippling it at some point. They've nerfed a lot of useful things over time, Wi-Fi scanning being one good example.

I assume this allows more granularity. Many apps avoid you blocking their marketing by not using the notification categories system. It's all or nothing. This app would presumably allow me to differentiate between the two if it can't be done with notification categories.

So fast mode uses more tokens, in direct opposition to Gemini where fast 'mode' means less. One more piece of useless knowledge to remember.

Was it leading with a bad analogy that gave it way?

Postgres has a large manual not because it's overly complex to do simple things, but because it is one of the best documented and most well-written tools around, period. Every time I've had occasion to browse the manual in the last 20 years it's impressed me.