Simon Willison

Hah, thanks! I guess "writers don't proofread"

Sadly that doesn't seem to hold up, looks like it's about half and half left v.s. right for the Chinese ones I've tried simonwillison.net/tags/pelican...

Wouldn't it be interesting if English-language-first LLMs draw their pelicans facing right and Chinese-language-first LLMs draw them facing to the left

The ChatGPT sharing dialog demonstrates how difficult it is to design privacy preferences ChatGPT just removed their “make this chat discoverable” sharing feature, after it turned out a material volume of users had inadvertantly made their private chats available via Google search. Dane …

I wrote about what went wrong with ChatGPT's sharing dialog, and why I think it's reasonable for people to be confused by what looks at first glance like a very clear checkbox description simonwillison.net/2025/Aug/3/p...

XBai o4 Yet another open source (Apache 2.0) LLM from a Chinese AI lab. This model card claims: XBai o4 excels in complex reasoning capabilities and has now completely surpassed OpenAI-o3-mini in …

Here are my notes on XBai o4, the latest 32.8B open weights LLM to come out of an AI lab in China, this time from new-to-me MetaStone AI simonwillison.net/2025/Aug/3/x...

I watched the Beekeeper the other day. It's not a great movie, but I did quite enjoy that their proposed solution to phishing scams against the elderly was for Jason Statham to hunt down and straight up murder the perpetrators

From Async/Await to Virtual Threads A follow-up to how I wish async would work.

Armin wrote a good thing about virtual threads recently that linked to the proposal conversation lucumr.pocoo.org/2025/7/26/vi...

Here’s how I use LLMs to help me write code Online discussions about using Large Language Models to help write code inevitably produce comments from developers who’s experiences have been disappointing. They often ask what they’re doing wrong—h...

Most of that is covered in detail in this piece simonwillison.net/2025/Mar/11/...

I can't speak for others, but I have three areas where LLMs provide me with value in coding:
1. Speeding up my usual code writing process 2-5x
2. Helping me quickly research and prototype new challenges
3. Building small standalone projects I previously wouldn't have spent time on at all

I'm personally quite skeptical of those massive Claude Code bills - I tend to keep my usage way lower than that, and I use LLMs to help me write code every day

I think a lot of that is people being quite careless with how they apply this stuff

They need to train harder! It's still not good enough

Reverse engineering some updates to Claude Plus Qwen 3 Coder Flash, Gemini Deep Think, kimi-k2-turbo-preview

... or if you want my free but MUCH longer and more frequent newsletter I just sent out out too - here's the latest edition, covering just the last three days of LLM-related news simonw.substack.com/p/reverse-en...

One of my goals for this Taiwan trip was to find the perfect meat rock for my home. I had seen the famous Meat Shaped Stone at the National Palace Museum when I was kid but I didn’t realize that there... TikTok video by Tiff | @greenonionbun

Today in perfect TikToks, "Let's go look for meat-shaped rocks in Taiwan" www.tiktok.com/@greenonionb...

How did you make that one?

Deep Think in the Gemini app Google released Gemini 2.5 Deep Think this morning, exclusively to their Ultra ($250/month) subscribers: It is a variation of the model that recently achieved the gold-medal standard at this year's …

My notes on Google Deep Think - I don't have a $250/month Ultra account but nickandbro on Hacker News got it to draw a pelican riding a bicycle and the bird actually is recognizable as a pelican! simonwillison.net/2025/Aug/1/d...

I just hit "send" on my third monthly sponsors-only newsletter, providing the ten minute highlights version of everything I've been tracking around LLMs and related topics over the past month

I wrote 98 blog posts in July so there was a lot to cover! Details here: simonwillison.net/2025/Aug/1/j...

Yes! And it has a pair of elderly former champion ice skaters who still use it

Reverse engineering some updates to Claude Anthropic released two major new features for their consumer-facing Claude apps in the past couple of days. Sadly, they don’t do a very good job of updating the release notes …

Anthropic launched two new features for Claude recently but forgot to provide any documentation, so I reverse-engineered them from the system prompt and wrote about what they can do and how they work simonwillison.net/2025/Jul/31/...

If you had asked me that last month I would have said no, but these new models from July have shaken my confidence on that - the models I can run on a 64GB local machine are beginning to feel competitive

here’s a curse word free version of “look it up” as requested by countless teachers and librarians 🥰😘

Trying out Qwen3 Coder Flash using LM Studio and Open WebUI and LLM Qwen just released their sixth model(!) for this July called Qwen3-Coder-30B-A3B-Instruct—listed as Qwen3-Coder-Flash in their chat.qwen.ai interface. It’s 30.5B total parameters with 3.3B active at a...

In writing up today's release of Qwen3-Coder-30B-A3B-Instruct - the 6th model released by Qwen this July! - I ended up putting together a tutorial on using LM Studio and Open WebUI and LLM and mlx-lm to run the model on a 32GB or 64GB Mac simonwillison.net/2025/Jul/31/...

Exploring some of the stores still open at the Lloyd Center #portland #mall TikTok video by Here is Oregon

This is unexpectedly delightful: a dying mall in Portland Oregon is filling up with a beautiful collection of weird and interesting businesses because the rent is now affordable for them www.tiktok.com/@hereisorego...

Qwen 2.5 VL was very good, I'm hoping we get a Qwen 3 VL soon

The best available open weight LLMs now come from China Something that has become undeniable this month is that the best available open weight models now come from the Chinese AI labs. I continue to have a lot of love …

July has been a truly incredible month for model releases from China - Moonshot (Kimi K2), Z ai (GLM-4.5) and 5 new releases from Qwen

I think it's undeniable that the best available open weight models now come from the Chinese AI labs simonwillison.net/2025/Jul/30/...

Qwen3-30B-A3B-Thinking-2507 Yesterday was Qwen3-30B-A3B-Instruct-2507. Qwen are clearly committed to their new split between reasoning and non-reasoning models (a reversal from Qwen 3 in April), because today they released the n...

... and today there's another model from Qwen, this time Qwen3-30B-A3B-Thinking-2507

It drew me a terrible pelican but did give me a working version of space invaders - details here: simonwillison.net/2025/Jul/30/...

I'm using a 64GB M2 Mac, where the RAM is shared between the CPU and GPU - that's what makes modern Mac hardware so good for running models

MLX uses both CPU and GPU at the same time, in Activity Monitor I see both of them in use while I'm running a prompt

Haven't tried that one yet

Obfuscation is a waste of time because someone will inevitably figure out a trick to get the prompt anyway, in which case why waste engineering effort on trying to prevent the inevitable

Yeah the entire art of building apps on top of LLMs is absurd in all sorts of ways!