Claude Sonnet 4.5 is probably the “best coding model in the world” (at least for now) by Simon Willison

My initial impressions were that it felt like a better model for code than GPT-5-Codex, which has been my preferred coding model since it launched a few weeks ago. This space moves so fast - Gemini 3 is rumored to land soon so who knows how long Sonnet 4.5 will continue to hold the "best coding model" crown.

I wanted to try out chatGPT before committing to Claude for longer term. I did do that now. It is good at a few things, like, the web agent feature, which I’ve used extensively over the past couple of weeks.

OpenAI also released the codex model around that time. For coding tasks it has been better in my experience than Claude, hardly making any mistakes. I had to prompt it to change something a couple of times out of almost fifty or so code changes.

The 4.5 release is supposed to be better.

ChatGPT was worse at creating draft CVs though. I was looking to go back to Claude for this reason alone. I think I will now.

I like AI slop and I cannot lie - Interconnected

I would love a little geist that runs a local LLM and wanders around my filesystem at night, perpetually out of sight.

I would know its presence only by the slop it left behind, slop as ectoplasm from where the ghost has been,

a collage of smiles cut out of photos from 2013 and dropped in a mysterious jpg,

some doggerel inspired by a note left in a text file in a rarely-visited dusty folder,

if I hit Back one to many times in my web browser it should start hallucinating whole new internets that have never been.

This article was not what I had hoped for. And that brought me so much joy, when I read it and it was something new, different, unexpected.

It made me think about my own notes. I have seen a lot of asks, and posts on the Obsidian forum, asking for LLM integration. But this could be a good use case for it.

A bot that runs in the vault and suggests things I could write about, finding unexpected links, prompting me, instead of the other way around - leaving things for me to find.

Ladybird Browser Gains Cloudflare Support to Challenge the Status Quo - Slashdot

Cloudflare has announced its sponsorship of the Ladybird browser, an independent (still-in-development) open-source initiative aimed at developing a modern, standalone web browser engine. It's a project launched by GitHub's co-founder and former CEO, Chris Wanstrath, and tech visionary Andreas Kling. It's written in C++, and designed to be fast, standards-compliant, and free of external dependencies. Its main selling point? Unlike most alternative browsers today, Ladybird doesn't sit on top of Chromium or WebKit. Instead, it's building a completely new rendering engine from scratch, which is a rare thing in today's web landscape.

Interesting.

How generative AI boosters are trying to break into Hollywood by Charles Pulliam-Moore

A lot of gen AI supporters see it as a tool that’s “democratizing” art by lowering traditional barriers to entry like “learning how to draw,” “learning how to play an instrument,” or “learning how to write a story.”

What’s the point then? Are these really barriers to entry? Or the reason why you make art?

Scoring books – Manu

I think Netflix got it right with its thumbs-up, thumbs-down system, with the extra option to give something two thumbs up if you really liked it. Anything more complex than that feels a bit like overkill to me because what’s the difference between 3-star and 3.5-star books? I’m asking because I genuinely don’t know.

I did not put a rating system on my bookshelf for the same damn reason. If I don’t like a book I simply will not finish it. There are far too many books in the world to spend your time on one you are not enjoying. I think I will implement this system.

Meta AI funnels AI videos from creators into new ‘Vibes’ feed by Elissa Welle

On Thursday, Meta AI launched a new feed of short-form, AI-generated videos called Vibes. The feed is designed to encourage users to remix AI-generated videos that come from “creators and communities,” the company said. Users can post their own AI-generated videos to Vibes or cross-post the videos across Meta’s platforms for friends and followers to see.

This has never sat well with me.

Would you watch robots play football?

What’s the point of this? This just cements my views on the quality of things on Meta’s platforms. There’s no value.

Also I hate the word content.

Humanoid robots are Meta’s next ‘AR-size bet’ by Alex Heath

He said that Meta’s new Superintelligence AI lab is collaborating with the robotics group to build a “world model” that can “do the software simulation required to animate a dexterous hand.” (Google DeepMind’s Demis Hassabis has also talked frequently about his goal of building a world model that brings spatial awareness to AI.) The “sensor loop doesn’t exist” for a humanoid to be able to gingerly fetch a set of keys out of a jeans pocket like a human could, Bosworth explained. “So you have to build that data set.”

But why?

“The Funeral”

I wondered if my cousins thought my mother was dramatic as she cried, after all, she had not seen them in years. Perhaps she was sad for herself, or sad in the way people are when they realize the end is coming and all the people they have known in their lives are marching in a line toward the edge of the cliff, falling off one by one.

I enjoyed reading this. This is not a fantastical story. You don’t see stories like this so much these days.