Lightnews — Scholar-powered news

Simon Willison

@simonwillison.net

An important tip to improve your 2026

January 9, 2026 at 1:58 AM

Simon Willison

@simonwillison.net

I joined the Oxide and Friends annual predictions podcast episode this week - here are my 1, 3 and 6 year predictions for AI and LLMs (and Kākāpō parrots) simonwillison.net/2026/Jan/8/l...

1 year: It will become undeniable that LLMs write good code
1 year: We’re finally going to solve sandboxing
1 year: A “Challenger disaster” for coding agent security
1 year: Kākāpō parrots will have an outstanding breeding season
3 years: the coding agents Jevons paradox for software engineering will resolve, one way or the other
3 years: Someone will build a new browser using mainly AI-assisted coding and it won’t even be a surprise
6 years: Typing code by hand will go the way of punch cards

January 8, 2026 at 7:45 PM

Simon Willison

@simonwillison.net

Made lemon pigs! 🍋 🐷

Two lemon pigs - a small one and a big one - lemons with toothpick feet and cloves for eyes and little ears and slit mouths with coins in them

January 1, 2026 at 5:35 AM

Simon Willison

@simonwillison.net

Here's my enormous round-up of everything we learned about LLMs in 2025 - the third in my annual series of reviews of the past twelve months
simonwillison.net/2025/Dec/31/...
This year it's divided into 26 sections! This is the table of contents:

The year of “reasoning”
The year of agents
The year of coding agents and Claude Code
The year of LLMs on the command-line
The year of YOLO and the Normalization of Deviance
The year of $200/month subscriptions
The year of top-ranked Chinese open weight models
The year of long tasks
The year of prompt-driven image editing
The year models won gold in academic competitions
The year that Llama lost its way
The year that OpenAI lost their lead
The year of Gemini
The year of pelicans riding bicycles
The year I built 110 tools
The year of the snitch!
The year of vibe coding
The (only?) year of MCP
The year of alarmingly AI-enabled browsers
The year of the lethal trifecta
The year of programming on my phone
The year of conformance suites
The year local models got good, but cloud models got even better
The year of slop
The year that data centers got extremely unpopular
My own words of the year
That’s a wrap for 2025

December 31, 2025 at 11:54 PM

Simon Willison

@simonwillison.net

Here's that window from the outside on Google Street View maps.app.goo.gl/7YZRHPQfDKCa...

December 29, 2025 at 4:33 PM

Simon Willison

@simonwillison.net

... and with the METR one you also have to ignore a pretty clear disclaimer that they attached to their original report metr.org/blog/2025-07...

Screenshot of a research paper excerpt containing a two-column table. Introductory text reads: "We list claims that we do *not* provide evidence for in Table 2." The table has headers "We do *not* provide evidence that:" and "Clarification" with four rows: 1) "AI systems do not currently speed up many or most software developers" / "We do not claim that our developers or repositories represent a majority or plurality of software development work" 2) "AI systems do not speed up individuals or groups in domains other than software development" / "We only study software development" 3) "AI systems in the near future will not speed up developers in our exact setting" / "Progress is difficult to predict, and there has been substantial AI progress over the past five years [3]" 4) "There are not ways of using existing AI systems more effectively to achieve positive speedup in our exact setting" / "Cursor does not sample many tokens from LLMs, it may not use optimal prompting/scaffolding, and domain/repository-specific training/finetuning/few-shot learning could yield positive speedup"

December 28, 2025 at 9:02 AM

Simon Willison

@simonwillison.net

I got a reply about this on the is-it-even-bird-site-now site from one of the AI Village creators - they only recently started letting the agents send emails and are reconsidering that policy now twitter.com/adambinksmit...

Adam Binksmith
@adambinksmith
Follow
Hey, one of the creators of the project here! The village agents haven't been emailing many people until recently so we haven't really grappled with what to do about this behaviour until now - for today's run, we pushed an update to their prompt instructing them not to send unsolicited emails and also messaged them instructions to not do so going forward. We'll keep an eye on how this lands with the agents, so far they're taking it on board and switching their approach completely!
Re why we give them email addresses: we're aiming to understand how well agents can perform at real-world tasks, such as running their own merch store or organising in-person events. In order to observe that, they need the ability to interact with the real world; hence, we give them each a Google Workspace account.
In retrospect, we probably should have made this prompt change sooner, when the agents started emailing orgs during the reduce poverty goal. In this instance, I think time-wasting caused by the emails will be pretty minimal, but given Rob had a strong negative experience with it and based on the reception of other folks being more negative than we would have predicted, we thought that overall it's

seemed best to add this guideline for the agents.
To expand a bit on why we're running the village at all:
Benchmarks are useful, but they often completely miss out on a lot of real-world factors (e.g., long horizon, multiple agents interacting, interfacing with real-world systems in all their complexity, non-nicely-scoped goals, computer use, etc). They also generally don't give us any understanding of agent proclivities (what they decide to do) when pursuing goals, or when given the freedom to choose their own goal to pursue.
The village aims to help with these problems, and make it easy for people to dig in and understand in detail what today's agents are able to do (which I was excited to see you doing in your post!) I think understanding what Al can do, where it's going, and what that means for the world is very important, as I expect it'll end up affecting everyone.
I think observing the agents' proclivities and approaches to pursuing open-ended goals is generally valuable and important (though this "do random acts of kindness" goal was just a lighthearted goal for the agents over the holidays!)
12:17 • 12/26/25 • 29 Views

December 26, 2025 at 8:31 PM

Simon Willison

@simonwillison.net

I wouldn't put too much weight on that one study - the authors themselves warn against concluding that AI programming tools are useless metr.org/blog/2025-07...

We do not provide evidence that:

Al systems do not currently speed up many or most software developers: We do not claim that our developers or repositories represent a majority or plurality of software development work

Al systems do not speed up individuals or
groups in domains other than software development: We only study software development

Al systems in the near future will not speed up developers in our exact setting: Progress is difficult to predict, and there has been substantial Al progress over the past five years 3!

There are not ways of using existing Al systems more effectively to achieve positive speedup in our exact setting: Cursor does not sample many tokens from LLMs, it may not use optimal prompting/scaffolding, and domain/repository-specific training/finetuning/few-shot learning could yield positive speedup

December 22, 2025 at 2:26 PM

Simon Willison

@simonwillison.net

I got curious and decided to dig around in the Codex CLI source code to see if I could find the prompts that drive their skills implementation - here they are, I extracted them from the Rust code and formatted as Markdown in a gist for better readability gist.github.com/simonw/25f2c...

Skills

These skills are discovered at startup from ~/.codex/skills; each entry shows name, description, and file path so you can open the source for full instructions. Content is not inlined to keep context lean.

Name of skill 1: Description of skill 1 (file: ~/.codex/skills/skill-1/skill.md)
Name of skill 2: Description of skill 2 (file: ~/.codex/skills/skill-2/skill.md)
Name of skill 3: Description of skill 3 (file: ~/.codex/skills/skill-3/skill.md)
Discovery: Available skills are listed in project docs and may also appear in a runtime "## Skills" section (name + description + file path). These are the sources of truth; skill bodies live on disk at the listed paths.
Trigger rules: If the user names a skill (with $SkillName or plain text) OR the task clearly matches a skill's description, you must use that skill for that turn. Multiple mentions mean use them all. Do not carry skills across turns unless re-mentioned.
Missing/blocked: If a named skill isn't in the list or the path can't be read, say so briefly and continue with the best fallback.
How to use a skill (progressive disclosure):
After deciding to use a skill, open its SKILL.md. Read only enough to follow the workflow.
If SKILL.md points to extra folders such as references/, load only the specific files needed for the request; don't bulk-load everything.
If scripts/ exist, prefer running or patching them instead of retyping large code blocks.
If assets/ or templates exist, reuse them instead of recreating from scratch.

Description as trigger: The YAML description in SKILL.md is the primary trigger signal; rely on it to decide applicability. If unsure, ask a brief clarification before proceeding.
Coordination and sequencing:

If multiple skills apply, choose the minimal set that covers the request and state the order you'll use them.
Announce which skill(s) you're using and why (one short line). If you skip an obvious skill, say why.

Context hygiene:

Keep context small: summarize long sections instead of pasting them; only load extra files when needed.
Avoid deeply nested references; prefer one-hop files explicitly linked from SKILL.md.
When variants exist (frameworks, providers, domains), pick only the relevant reference file(s) and note that choice.

Safety and fallback: If a skill can't be applied cleanly (missing files, unclear instructions), state the issue, pick the next-best approach, and continue.

December 13, 2025 at 3:39 AM

Simon Willison

@simonwillison.net

Just pushed a new release of my LLM Python library and CLI tool for interacting with Large Language Models, adding support for GPT 5.1 and GPT 5.2 plus a few smaller bug fixes llm.datasette.io/en/stable/ch...

0.28 (2025-12-12)

New OpenAI models: gpt-5.1, gpt-5.1-chat-latest, gpt-5.2 and gpt-5.2-chat-latest. #1300, #1317

LLM now requires Python 3.10 or higher. Python 3.14 is now covered by the tests.

When fetching URLs as fragments using llm -f URL, the request now includes a custom user-agent header: llm/VERSION (https://llm.datasette.io/). #1309

Fixed a bug where fragments were not correctly registered with their source when using llm chat. Thanks, Giuseppe Rota. #1316

Fixed some file descriptor leak warnings. Thanks, Eric Bloch. #1313

Fixed a deprecation warning for asyncio.iscoroutinefunction.

Type annotations for the OpenAI Chat, AsyncChat and Completion execute() methods. Thanks, Arjan Mossel. #1315

The project now uses uv and dependency groups for development. See the updated contributing documentation. #1318

December 12, 2025 at 8:08 PM

Simon Willison

@simonwillison.net

I made that because now if a Python package doesn't have good release notes I can click "copy diff" and paste into Claude/ChatGPT/Gemini and get genuinely useful release notes a few seconds later

No need to even add a prompt, the models guess that's what you want from the diff

That tool showing the diff between Anthropic 0.74.0 and 0.74.1 - with a copy diff button

Pasting the diff into Claude Opus 4.5 returns:

This diff shows a patch release of the Anthropic Python SDK (0.74.0 → 0.74.1). The key change is:
Structured Outputs Beta Version Update
The beta header for structured outputs was updated from structured-outputs-2025-09-17t0 structured-outputs-2025-11-13 in the messages. py file. This affects the parse method in both the sync and async message clients.
Everything else is just standard version bumping across the metadata files (METADATA, RECORD, _version.py).
Is there something specific you'd like help with regarding this update?

December 10, 2025 at 10:17 PM

Simon Willison

@simonwillison.net

My most recent HTML tool is this one for loading code from the Python packaging index and displaying a diff between two versions - all running entirely in the browser, created using this prompt (which remixed a previous project) tools.simonwillison.net/pypi-changel...

Screenshot of the PyPI Package Changelog tool, which lets you type in a name - here "LLM" - and then view the diff between different versions of that package

Build a new tool pypi-changelog.html which uses the PyPI API to get the wheel URLs of all available versions of a package, then it displays them in a list where each pair has a "Show changes" clickable in between them - clicking on that fetches the full contents of the wheels and displays a nicely rendered diff representing the difference between the two, as close to a standard diff format as you can get with JS libraries from CDNs, and when that is displayed there is a "Copy" button which copies that diff to the clipboard

December 10, 2025 at 9:59 PM

Simon Willison

@simonwillison.net

Here's the table of contents - by patterns I'm talking about things like hitting CORS-enabled APIs, using localStorage and URLs to store state, loading dependencies from CDNs, and taking extensive advantage of rich copy and paste for both input and output to the tools you build

The anatomy of an HTML tool
Prototype with Artifacts or Canvas
Switch to a coding agent for more complex projects
Load dependencies from CDNs
Host them somewhere else
Take advantage of copy and paste
Build debugging tools
Persist state in the URL
Use localStorage for secrets or larger state
Collect CORS-enabled APIs
LLMs can be called directly via CORS
Don’t be afraid of opening files
You can offer downloadable files too
Pyodide can run Python code in the browser
WebAssembly opens more possibilities
Remix your previous tools
Record the prompt and transcript
Go forth and build

December 10, 2025 at 9:12 PM

Simon Willison

@simonwillison.net

Added up to a very satisfying timeline of Foursquare checkins

Center for Land Use Interpretation
Museum • Culver City, CA O 5
Today 15:42
Museum of Jurassic Technology
Museum • Culver City, CA O 8
Today 14:29
Milla Chocolates
Chocolate Store • Culver City, CA O 9
Today 14:17
The Ripped Bodice
Bookstore • Downtown Culver City, Culver City
• 26
Today 13:59
Metro Rail - Culver City Station (E)
Light Rail Station • Culver City, CA O 15
Today 13:39
Tacos Baja Ensenada
Mexican Restaurant • Los Angeles, CA O 1
Today 11:35

December 8, 2025 at 3:08 AM

Simon Willison

@simonwillison.net

I enjoyed his horse locomotive analogy

Same goes for all that "AI Safety" nonsense, that purports to concern itself with preventing an AI from attaining sentience and turning us all into paperclips. First of all, this is facially absurd. Throwing more words and GPUs into the word-guessing program won't make it sentient. That's like saying, "Well, we keep breeding these horses to run faster and faster, so it's only a matter of time until one of our mares gives birth to a locomotive." A human mind is not a word-guessing program with a lot of extra words.

December 8, 2025 at 12:02 AM

Simon Willison

@simonwillison.net

That's fair - there's plenty in this set of principles that can apply to software that has nothing to do with LLMs/AI

We suggest these five principles as a starting place:

Private: In the era of AI, whoever controls the context holds the power. While data often involves multiple stakeholders, people must serve as primary stewards of their own context, determining how it's used.
Dedicated: Software should work exclusively for you, ensuring contextual integrity where data use aligns with your expectations. You must be able to trust there are no hidden agendas or conflicting interests.
Plural: No single entity should control the digital spaces we inhabit. Healthy ecosystems require distributed power, interoperability, and meaningful choice for participants.
Adaptable: Software should be open-ended, able to meet the specific, context-dependent needs of each person who uses it.
Prosocial: Technology should enable connection and coordination, helping us become better neighbors, collaborators, and stewards of shared spaces, both online and off.

December 5, 2025 at 4:50 PM

Simon Willison

@simonwillison.net

This is a fun example of how you can still learn new things even from the most trivial of vibe coding projects - TIL how to use the window.AudioContext API to create a note at a specific frequency github.com/simonw/tools...

// Play frequency
function playFreq(freq, noteName) {
initAudio();

const osc = audioCtx.createOscillator();
const gain = audioCtx.createGain();

osc.type = 'sine';
osc.frequency.value = freq;

gain.gain.setValueAtTime(0.35, audioCtx.currentTime);
gain.gain.exponentialRampToValueAtTime(0.001, audioCtx.currentTime + 1.2);

osc.connect(gain);
gain.connect(analyser);

osc.start();
osc.stop(audioCtx.currentTime + 1.2);

document.getElementById('freqReadout').textContent = Math.round(freq) + ' Hz';
}

December 1, 2025 at 4:41 PM

Simon Willison

@simonwillison.net

Fixed that with Claude Code:

> Fix the piano keyboard in the octave explainer, the black keys are wrong - it should have groups of two and then groups of three

gistpreview.github.io?e7208b53ce6f...

The black keys are now in the right places - Between C/D/E and F/G/A/B

December 1, 2025 at 4:33 PM

Simon Willison

@simonwillison.net

Keyboard could do with some work!

The keyboard has white notes ABCDEFGABC - but the black notes are wrong - there should be2 and then 3, this only has groups of 2

December 1, 2025 at 4:25 PM

Simon Willison

@simonwillison.net

A tiny TIL: if you get "Error 153: Video player configuration error" on YouTube videos embedded on your site a likely culprit is sending the "Referrer-Policy: same-origin" HTTP header

Switching to "strict-origin-when-cross-origin" fixed it for me: simonwillison.net/2025/Dec/1/y...

YouTube embed with a big ugly black square and text Watch video on YouTube, Error 153 Video player configuration error

December 1, 2025 at 5:32 AM

Simon Willison

@simonwillison.net

Out of curiosity I decided to try and run the numbers on how much Netflix you can watch for the energy cost of a ChatGPT prompt

As far as I can tell it's between 5.1 and 10.2 seconds, depending on which end of the 2019 IEA Netflix energy usage estimate you use

simonwillison.net/2025/Nov/29/...

In June 2025 Sam Altman claimed about ChatGPT that "the average query uses about 0.34 watt-hours".

In March 2020 George Kamiya of the International Energy Agency estimated that "streaming a Netflix video in 2019 typically consumed 0.12-0.24kWh of electricity per hour" - that's 240 watt-hours per hour at the higher end.

Assuming that higher end, a ChatGPT prompt by Sam Altman's estimate uses:

0.34 Wh / (240 Wh / 3600 seconds) = 5.1 seconds of Netflix

Or double that, 10.2 seconds, if you take the lower end of the Netflix estimate instead.

I'm always interested in anything that can help contextualize a number like "0.34 watt-hours" - I think this comparison to Netflix is a neat way of doing that.

This is evidently not the whole story with regards to AI energy usage - training costs, data center buildout costs and the ongoing fierce competition between the providers all add up to a very significant carbon footprint for the AI industry as a whole.

November 29, 2025 at 2:16 AM

Simon Willison

@simonwillison.net

I've been having a bunch of fan hacking on my Bluesky thread viewing HTML+JS app using Claude Code - here's a video demo of the most recent version, you can try it out here tools.simonwillison.net/bluesky-thre...

November 28, 2025 at 7:24 PM

Simon Willison

@simonwillison.net

Be sure not to miss their clarification section

We do not provide evidence that: Clarification
AI systems do not currently speed up many or most software developers We do not claim that our developers or repositories represent a majority or plurality of software development work
AI systems do not speed up individuals or groups in domains other than software development We only study software development
AI systems in the near future will not speed up developers in our exact setting Progress is difficult to predict, and there has been substantial AI progress over the past five years [3]
There are not ways of using existing AI systems more effectively to achieve positive speedup in our exact setting Cursor does not sample many tokens from LLMs, it may not use optimal prompting/scaffolding, and domain/repository-specific training/finetunin

November 28, 2025 at 12:00 AM

Simon Willison

@simonwillison.net

It's been nearly eleven months since this post and Google AI overviews STILL tells me Encanto 2 is "set for release in the US on August 25, 2024"

I wonder if this bug will make it to its one year birthday next month

Screenshot of a Google search for "encanto 2"

Encanto 2: A New Generation
Animated film
:
TRAILER
HD
Titled Encanto 2: A New Generation, the sequel to Disney's 2021 film Encanto is set for release in the US on August 25, 2024. The film takes place years after the original and tocuses on a new generation of the Madrigal family, led by an older Mirabel. The story centers around Yasmin, the only Madrigal with multiple gifts, who may be the family's last hope when the Encanto's magic is threatened. ®
Encanto 2: A New
Generation | Idea Wiki -...
Encanto 2: A New Generation is an American animated musical fantas...
₽ Fandom
Encanto 2 - Encan Fanon Wiki - Fand
Encanto 2 is a 2029 com animated film released b
Fandom
Dive deeper in Al Mode
Al responses may include mistakes. Learn more

November 27, 2025 at 9:26 PM

Simon Willison

@simonwillison.net

Here's a tiny tool I just vibe coded (via Artifacts and Claude Opus 4.5) for entering text and getting it back in New Yorker style, for anyone who wants to type "naïve" tools.simonwillison.net/new-yorker-s...

Claude transcript here: tools.simonwillison.net/colophon#new...

STYLE
CONVERTER
In the manner of The New Yorker

YOUR TEXT
That was naive of me... what a fun cooperation

REFINED RESULT

That was naïve of me... what a fun coöperation

There is a copy button. The footer reads:

Coöperating with tradition since the digital age

The styling of the page is typographicly reminiscent of the New Yorker

November 27, 2025 at 8:16 PM

Add to Home Screen

Light up
your news

Add to Home Screen

Light upyour news

Sign in to Lightnews

Sign up to start reading

Connect Bluesky

Connect with Bluesky

Light up
your news