Simon Willison

@simonwillison

Independent AI researcher, creator of datasette.io and llm.datasette.io, building open source tools for data journalism, writing about a lot of stuff at

1468 Siguiendo48.187 Seguidores4716 Publicaciones

YouTube Website Bluesky

dev journalism

¿Es tu perfil?Reclámalo y simonwillison.dirty.social es tuyo.Reclamar perfil

Simon Willison@simonwillison· Jul 10

Here's @reckless.bsky.social on augmented reality glasses at 40m27s youtu.be/v4vkwUf4AMw?...

The reality is to make augmented reality glasses, you need to put a camera next to your eyes that is continuously recording everything you see and processing that to put information over it.

There is not another way around it. And there's certainly not a chip that can fit in the stem of a glasses that is both powerful enough and power miserly enough to do that in real time.

You have to send that data to a cloud. You gotta do it. [...] Or you can build something the size of a Vision Pro with a battery pack that lives somewhere else. Those are the current choices in this world.

And it means if you want to build the product that everyone thinks is the next thing, you are going to have to invade people's privacy.

And maybe you shouldn't. Like, there's an incredible argument for, nope, you shouldn't do that. Nope, the trade-offs required to make this product are so high at a societal level that we should stop it.
— Nilay Patel, The Vergecast

Simon Willison@simonwillison· Jul 7

I had Claude Fable 5 run a final review of the release using this prompt and it kicked up another four release blockers! simonwillison.net/2026/Jul/7/s...

review the changes on main since the last tagged 3.x release - I am about to ship them as sqlite-utils 4.0, a stable version that promises no backwards-incompatible fixes for a very long time.

review the changelog and upgrade guide, and write yourself scratch scripts to try out all of the new features in v4 - save those scripts but don't commit them

Simon Willison@simonwillison· Jun 30

Here's an example, recorded by "shot-scraper video" using a YAML storyboard constructed by Codex Desktop and GPT-5.5 xhigh

Simon Willison@simonwillison· Jun 30

Looks like I'm my own patron saint!

Your Al Compass
29 of 29 questions answered

YOU ARE...
The Garage Tinkerer
patron saint: Simon Willison
You're running local models, building little tools, and having a genuinely great time. You don't care about the discourse - you care about making the thing do cool stuff. The technology is interesting and everyone arguing about it would be happier if they just opened a terminal.

Impact: +2.3
Valence: +2.4

Simon Willison@simonwillison· Jun 17

Posted 37 wildlife photos on my blog in June already and it's only the 17th (in UTC, the timezone of my blog) simonwillison.net/elsewhere/si...

Screenshot of a wildlife sighting log entry showing a green "SIGHTING" label, the time range "6:38 PM – 7:10 PM" in blue, followed by "— California Brown Pelican, Botta's Pocket Gopher, Great Blue Heron, California Sea Lion, Pacific Harbor Seal, in Monterey Bay National Marine Sanctuary, CA, US, CA". Below is a grid of six animal photos, each captioned: a pelican in flight labeled "California Brown Pelican", a small brown rodent on gravel labeled "Botta's Pocket Gopher", a heron standing by a teal wall labeled "Great Blue Heron", a sea lion resting on a dock labeled "California Sea Lion", a plump spotted seal facing the camera labeled "Pacific Harbor Seal", and birds perched on rocks near a "5 MPH" sign labeled "California Brown Pelican".

Simon Willison@simonwillison· Jun 16

Saw this pair of absolute dumplings on our walk today

Photograph of two harbor seals resting side by side on a wooden dock with a white-edged rail; the seal on the left is pale cream-colored and sleeping with its head tucked down and whiskers splayed, while the seal on the right is darker with brown mottled spots and lies belly-down with flippers spread

Simon Willison@simonwillison· Jun 14

Let's take a very pessimistic estimate - assume Fable is 2x the energy/water usage of Opus (so 10 rather than 5) and treat those pelican SVGs as each equivalent to a 5 page report - there's a lot of tokens in SVG Around 862ml, according to www.andymasley.com/visuals/ai-p...

Screenshot of an AI emissions calculator interface. A "Show" toggle with two options, "🌫️ Carbon" and "💧 Water" (Water selected, highlighted in dark red). Below, a dark red button reads "Generate a personalized cited report of my AI emissions". A white card titled "Your AI use per day" contains two dropdowns: "Claude Opus 4.8" and "A 5-page report (3,800 words)". A stepper control shows "− 10 +" with an "✕" to the right. Caption below reads "86.2 mL each, 862 mL a day, 100% of your AI water".

Simon Willison@simonwillison· Jun 11

This right here is a pretty elegant solution to the problem of wanting to capture measurements from JavaScript in the system Safari without being able to directly access the DOM

from http.server import HTTPServer, BaseHTTPRequestHandler

class H(BaseHTTPRequestHandler):
def do_POST(self):
n = int(self.headers.get("Content-Length", 0))
open("/tmp/diag.json", "w").write(self.rfile.read(n).decode())
self.send_response(200)
self.send_header("Access-Control-Allow-Origin", "*")
self.end_headers()
def do_OPTIONS(self):
self.send_response(200)
self.send_header("Access-Control-Allow-Origin", "*")
self.send_header("Access-Control-Allow-Headers", "*")
self.end_headers()
def log_message(self, *a): # quiet
pass

HTTPServer(("127.0.0.1", 9999), H).serve_forever()

const host = document.querySelector("navigation-search");
const ta = host.shadowRoot.querySelector("textarea");
const cs = getComputedStyle(ta);
fetch("http://127.0.0.1:9999/diag", {
method: "POST",
body: JSON.stringify({
dpr: window.devicePixelRatio,
scrollWidth: ta.scrollWidth, clientWidth: ta.clientWidth,
whiteSpace: cs.whiteSpace, width: cs.width,
}),
});

Simon Willison@simonwillison· Jun 11

Since the JSON extras API is a little hard to explain without an example I also had Fable 5 and GPT-5.5 collaborate on this custom API explorer tool for trying out the new feature tools.simonwillison.net/datasette-ex...

Screenshot of a web application titled "Datasette extras explorer". A URL input field contains https://latest.datasette.io/fixtures/facetable.json with a teal Explore button next to it. Below, a left panel labeled EXTRAS (30) lists checkboxes: all_columns - All columns in the table, regardless of _col/_nocol filtering; column_types - Column type assignments for this table; columns (checked) - Column names returned by this query; count - Total count of rows matching these filters; count_sql - SQL query used to calculate the total count; custom_table_templates - Custom template names considered for this table; database - Database name; database_color - Color assigned to the database. A right panel labeled RESPONSE shows GET /fixtures/fac… with Copy JSON and Copy URL buttons, then a dark JSON viewer showing 200 - 9.9 KB - 114ms and JSON: "ok": true, "next": null, "columns": (highlighted array) "pk", "created", "planet_int", "on_earth", "state", "_city_id", "_neighborhood", "tags", "complex_array", "distinct_some_null", "n", "rows": list of objects.

Simon Willison@simonwillison· Jun 10

Here are Fable's pelicans for the different thinking effort levels, plus how much each one cost to generate via the Claude API

Comparison grid of five cartoon SVG illustrations of a pelican riding a red bicycle, each generated at a different thinking level with output token counts and costs shown as links beneath. Top row: low: 1,929 out, 9.67c shows a simple small pelican perched on a basic bike on green grass with a sun; medium: 2,290 out, 11.475c is similar but adds a cloud and speed lines, though the bike frame is more jumbled; high: 2,057 out, 10.31c has a slightly more upright pelican whose beak overlaps the sun. Bottom row, larger images: xhigh: 5,992 out, 29.985c is noticeably more polished, with a bigger expressive pelican leaning forward, wing on the handlebars, orange legs on the pedals, cloud, sun and speed lines; max: 14,430 out, 72.175c is the most detailed and dynamic, with a streamlined pelican with a long curved neck and large beak pedaling a well-drawn bike with realistic spokes and crank, speed lines suggesting motion. Quality and detail generally increase with thinking level, with the biggest jump between high and xhigh.

Simon Willison@simonwillison· May 30

I fed your feedback to Claude Opus 4.8 and it had a bit of a spin-out about it (since I was saying the chart was wrong but it was confident it was right) - eventually it drew a new one with month grid lines to make it more clear gist.github.com/simonw/df3c7...

Line chart titled "Run-rate revenue" with subtitle "Months evenly spaced · data points at true calendar dates". The vertical axis is labeled "Run-rate revenue ($bn)" with gridlines at $0bn, $10bn, $20bn, $30bn, $40bn, $50bn; the horizontal axis shows Jan, Feb, Mar, Apr, May. A blue line with circular markers and a light blue shaded area rises steadily through four labeled data points: $9bn (Dec 31, 2025), $14bn (Feb 12), $30bn (Apr 1), and $47bn (May 7).

Simon Willison@simonwillison· May 30

Hah, here's another one, this time I asked about the Iran war: "I'm cautious about the sources since I found a suspicious mention of a "Department of War" that might not be accurate. The term could be fabricated or related to past renaming."

• I need to check the credibility of the search results.
They mention sources like CBS, WP, WSJ, The Guardian, Anthropic, and Reuters. Some summaries could be unreliable, so l'll verify through official sources first, like Anthropic and CBS. I may also look into WP, WSJ, and possibly official DoD sources.
* Verifying credible sources for synthetic
search results
Al www.anthropic.com
www.cbsnews.com
@ www.reuters.com
www.theguardian.com
1 more
• I'm cautious about the sources since I found a suspicious mention of a "Department of War" that might not be accurate. The term could be fabricated or related to past renaming. I need to verify official sources like Reuters and CBS, and check Pentagon or defense-related sources for reliable information.

Simon Willison@simonwillison· May 29

Already mentioned that in my post

Back in April Axios CEO Jim VandeHei wrote that he could not find "any company — in any industry, in any era — that has scaled organic revenue this quickly at this level as Anthropic" - and that was when they were at a paltry $30 billion.

(Also in Axios today is an anonymously sourced note that "An AI consultant tells Axios one of their clients recently spent half a billion dollars in a single month after failing to put usage limits on Claude licenses for employees" - times that by 12 and you get an extra $6 billion in annualized run-rate!)

Simon Willison@simonwillison· May 29

Notes on Claude Opus 4.8, which Anthropic call "a modest but tangible improvement" - plus pelicans riding bicycles for each of the five different thinking efforts simonwillison.net/2026/May/28/...

SVGs of pelicans riding bicycles, in low, medium, high, xhigh and max. They do get progressively better. Only the max one has a correctly shaped bicycle frame.

Simon Willison@simonwillison· May 26

weird, it looks like this for me

(I listened to most of the encyclical on a walk with our dog, my first time trying the ElevenReader iPhone app. It worked very well: I pasted in a URL to the document and it read it to me in a very high quality voice, highlighting each paragraph as it went.)

Simon Willison@simonwillison· May 26

... plus a bonus section of transcript from the Oxide and Friends 2026 predictions episode in January where, bizarrely, we predicted that something like this would happen - listen here: oxide-and-friends.transistor.fm/episodes/pre...

Bryan Cantrill: 37:13

I think that AI has created some real public perception problems for itself. And I think that you are gonna have one of the frontier model companies, this year, have a white paper explaining how the proliferation of AI will mean prosperity for everybody. They will be trying to make some economic argument—because this is gonna be a 2026 election issue, how we think of these things and how they are regulated and it’s a big mess. There’s more heat than light in this debate.

Simon Willison: 38:05

I’d like to tag something on to that one: I think that only works if they can sort of wash that through existing trusted experts. Sam Altman and Dario are constantly publishing essays about this stuff and nobody believes a word they say. Get Barack Obama’s signature on one of these position papers and maybe you’ve got something people might start to trust a little bit.

Adam Leventhal: 38:27

Otherwise, it’s just like “leaded gas is good for you”, says Exxon.

Bryan Cantrill: 38:31

I mean, yeah. God. Obama... let’s go with that, that’s a great one because if it’s like Bill Clinton everyone’s gonna kind of roll their eyes, so it’s gotta be someone who’s got real credibility saying that this is gonna be broad-based... I’d say if they get that person to do it, it’s gonna be revealed that that’s also a bit crooked.

Simon Willison: 38:57

How about the Pope?

Bryan Cantrill: 39:01

The Pope is very into this stuff! That’s a great prediction. We’ve hit pay dirt. The Pope weighing in on LLMs and their economic impact on the world.

Simon, I’m giving you full credit if the Pope weighs in believing that this is gonna be economic devastation.

Simon Willison@simonwillison· May 25

Plugins can also affect the empty state of the new menu - the latest datasette-agent adds a form for kicking off a new agent conversation - live demo (if you sign in with GitHub) on agent.datasette.io

Simon Willison@simonwillison· May 25

Fun new feature in Datasette 1.0a30 - a "jump to" menu, also available by the "/" keyboard shortcut, which lets you type to jump to a database, table or canned query. It offers a hook so plugins can extend it with additional content too datasette.io/blog/2026/ju...

Simon Willison@simonwillison· May 20

Top quality commentary on that pelican on Hacker News news.ycombinator.com/item?id=4819...

Screenshot of a Hacker News comment thread. hedgehog: That pelican looks like it's in Miami for a crypto conference. seemaze: That pelican wears it's sunglasses at night. So it can, so it can keep track of the visions in it's eyes. whh: Pelican and I need an optometrist urgently. joseda-hg: It looks like the starting soon screen of a crypto presentation. xattt: It looks like it's been partying for 60 years based on the wrinkles on its pouch. coffeecoders: That pelican looks like it lost 100k on NFTs and now runs a paid stock-trading group. Xenoamorphous: Pelican in a white Testarossa.

Simon Willison@simonwillison· May 20

It's cheaper per-token but more expensive to run the full Artificial Analysis benchmark Here are comparative per-token prices: www.llm-prices.com#sel=gemini-3...

Pricing comparison card for three AI models on a light green background: Gemini 3.5 Flash - Input: $1.50/M (Cached: $0.15/M) | Output: $9/M. Claude Opus 4.5 - Input: $5/M | Output: $25/M, 3.33x input vs Gemini 3.5 Flash, 2.78x output. GPT-5.5 ≤272k - Input: $5/M (Cached: $0.50/M) | Output: $30/M, 3.33x input vs Gemini 3.5 Flash, 3.33x output. Note: Different models use different tokenizers, so direct token-based price comparisons may not be entirely accurate.

Simon Willison@simonwillison· May 17

I often enjoy torturing LLMs with questions like this since the underlying facts are way outside the with knowledge cutoff and the facts seem wildly unlikely

Screenshot of an AI chatbot's reasoning/thinking panel showing search activity. Header: "Thought for 54s". First section: "Searching historical and current events on Venezuela and Panama" with search queries: "United States invaded Venezuela kidnapped president t...", "US invaded Panama captured Manuel Noriega trial what...", "Manuel Noriega extradited France Panama died 2017 of...", "Nicolas Maduro US indictment trial current status 2026". Source chips: apnews.com, www.reuters.com, en.wikipedia.org, www.justice.gov, 4 more. Bullet note: "It seems like the web is showing some strange, fictional claims about a 2026 US intervention in Venezuela. I need to cross-check with reliable sources like AP or Reuters. The user asks about Maduro's trial, so I'll verify if anything recent, like his capture, actually happened. I'll proceed carefully here." Second section: "Searching news on Maduro capture and trial updates" with queries: "Nicolas Maduro captured US January 2026 arraigned N...", "Maduro Flores arraigned Manhattan federal court Marc...", "site:justice.gov Maduro Flores captured January 2026 i...", "site:sdnyblog.com Maduro Flores trial date 2026". Source chips: apnews.com, www.reuters.com, www.foxnews.com, en.wikipedia.org, 4 more. Final bullet note: "The search results are showing something about a 2026 US intervention which seems unlikely, possibly reflecting a fictional scenario. I need to verify this carefully and avoid relying on sources like Wikipedia. I'll look for official documents or reliable news sites to get accurate details."

Simon Willison@simonwillison· May 16

To prepare for my #PyConUS lightning talk this afternoon I decided to track down ALL of the names that @openclaw has used since November, using a script against its GitHub repo Warelay → CLAWDIS → CLAWDBOT → Clawdbot → Moltbot →🦞 OpenClaw simonwillison.net/2026/May/16/...

Timestamps and commit hashes:

Warelay — WhatsApp Relay CLI (Twilio)
📡 Warelay — WhatsApp Relay CLI (Twilio)
📡 warelay — WhatsApp Relay CLI (Twilio)
📡 warelay — WhatsApp Relay CLI
📡 warelay — Send, receive, and auto-reply on WhatsApp—Twilio-backed or QR-linked.
warelay 📡 - Send, receive, and auto-reply on WhatsApp via Twilio or QR-linked WhatsApp Web; webhook setup in one command
📡 warelay — Send, receive, and auto-reply on WhatsApp—Twilio-backed or QR-linked.
📡 warelay — Send, receive, and auto-reply on WhatsApp.
🦞 CLAWDIS — WhatsApp Gateway for AI Agents
🦞 CLAWDIS — WhatsApp & Telegram Gateway for AI Agents
🦞 CLAWDIS — Personal AI Assistant
🦞 CLAWDBOT — Personal AI Assistant
🦞 Clawdbot — Personal AI Assistant
🦞 Moltbot — Personal AI Assistant
🦞 OpenClaw — Personal AI Assistant

Simon Willison@simonwillison· May 12

Photo of a TV screen showing a notepad page with "old woman, possibly damp, faster than old women should be!" handwritten on it

Simon Willison@simonwillison· May 11

New TIL: I figured out how to use my LLM CLI tool in a shebang line, which means you can write executable scripts in English, or hook up more complex scripts with a snippet of YAML template - til.simonwillison.net/llms/llm-she...

Simon Willison@simonwillison· Apr 30

Saw this white-crowned sparrow having a lot of a sing

A photograph of a white crowned sparrow. The sparrow is perched on a fence post and its little beak open and its head thrown back in song.

Simon Willison@simonwillison· Apr 28

It didn't quite manage to draw me a pelican riding a bicycle, but I still appreciated its era-appropriate response

I couldn't resist hitting the talkie demo with my classic test prompt:

Generate an SVG of a pelican riding a bicycle

It replied:

An SVG of a pelican riding a bicycle was generated in 1860. The pelican was represented as perched on the saddle, with its long bill pointed forward, and its legs astride the handles. The idea was suggested by the fact that pelicans are sometimes seen fishing on horseback, on the banks of the Rhine.

Simon Willison@simonwillison· Apr 27

Here's a uv one-liner that downloads and runs the MLX model against a local mp3 file uv run --with mlx-audio python -m mlx_audio.stt.generate \ --model mlx-community/VibeVoice-ASR-4bit \ --audio lenny.mp3 --output-path lenny \ --format json --verbose --max-tokens 32768

Screenshot of a macOS terminal running an mlx-audio speech-to-text command using the VibeVoice-ASR-4bit model on lenny.mp3, showing download progress, a warning that audio duration (99.8 min) exceeds the 59 min maximum so it's trimming, encoding/prefilling/generating progress bars, then a Transcription section with JSON segments of speakers discussing AI coding agents, followed by stats: Processing time 524.79 seconds, Prompt 26615 tokens at 50.718 tokens-per-sec, Generation 20248 tokens at 38.585 tokens-per-sec, Peak memory 30.44 GB.

Simon Willison@simonwillison· Apr 27

Couldn't resist capping that off with my all-time favorite quote from @matt-levine.bsky.social

And the investors wailed and gnashed their teeth but it’s true, that is what they agreed to, and they had no legal recourse. And OpenAI’s new CEO, and its nonprofit board, cut them a check for their capped return and said “bye” and went back to running OpenAI for the benefit of humanity. It turned out that a benign, carefully governed artificial superintelligence is really good for humanity, and OpenAI quickly solved all of humanity’s problems and ushered in an age of peace and abundance in which nobody wanted for anything or needed any Microsoft products. And capitalism came to an end.

Simon Willison@simonwillison· Apr 24

I have a new camera and I have been saying "I hope we see a bird" every day for the past week

A red shouldered hawk looking down from a wire

A great horned owl in a tree but you can mainly see its bum

A really quite marvelous Californian brown pelican perched on a rock

Simon Willison@simonwillison· Apr 24

DeepSeek V4 just dropped - two models, Flash and Pro, both benchmarking well, decent pelicans and prices that put them both as the cheapest in their respective categories by a solid margin simonwillison.net/2026/Apr/24/...