Totally agree, really liked the post I read about until I felt the hairs on the back of my neck go up...

While I have you, any particularly good resources you know of for having LLMs do code reviews?

Which produces better code:
1. Claude Sonnet 4.5
2. GPT 5.2
3. Zombie Charles Manson

@simonwillison.net I nominate Clawd (clawd.bot) for the perpetrator of the "Challenger Level Disaster", at least for some poor individual's machine.

Dear @tkingfisher.com I don't give a rat's patootie if you ever go on video, but *THANK YOU THANK YOU THANK YOU* so freaking much for being here and making bsky so awesome.

Oh, and your books are pretty good too :-)

I was in my first year at college. Came back to my dorm room to discover the pattern on the floor! I'd thought it solid gray.

It's cool that it recognizes it as a dog! And in a green muddy field, with a wet coat! And one gave a reasonable breed guess. But it just can't spatially analyze what it's looking at, or count. AGI is a ways away yet...

- One outline around the dog's head and neck.
- One outline around the upper torso and front legs.
- One outline around the lower torso and hind legs.
- One outline around each of the four legs individually.

This results in a total of six red outlines.

There are 6 red outlines in this image

I then asked it just to count the red outlines. I got:

In the image, there are several red outlines highlighting different parts of the dog. By counting each distinct red outline, I observe the following:

/3

I asked Mistral-3-large to count the outlines, and say what body part is encircled:

> In the image, there are four distinct red outlines. These outlines encircle the following body parts of the dog:

1. The front left leg.
2. The front right leg.
3. The back left leg.
4. The back right leg.

Um /2

Hey @simonwillison.net ! You probably saw this already on hn, but... what Vision models "know" affects what they see. As an example, counting legs on a dog (note: AI generated image ;-)

Note the five legs! BUT, if you ask LLMs which "understand" vision, they'll tell you it has four legs. /1

Really liked that article, the bit about container versus contents is very helpful. Looking forward to the follow-on articles!

Crepe myrtle maybe? The ones outside our apartment are not gnarled, alas.

I'm in!