Simon Willison

Open source developer building tools to help journalists, archivists, librarians and others analyze, explore and publish their data. https://datasette.io and many other #projects.

Public Key

npub13v97j0kknscwnf5pt87nsn7cxzxwfwl3dsu7ss8qsq7ukmqgwg8q84w5eu

NIP-05 Address

simon_at_fedi.simonwillison.net@mostr.pub

Profile Code

nprofile1qqsgkzlf8mtfcv8f56q4nlfcflvrpr8yh0ckcw0ggrsgq0wtdsy8yrspz4mhxue69uhhyetvv9ujumt0wd68ytnsw43qzxrhwden5te0wfjkccte9ehx7umhdpjhyefwvdhk6rn0svd

Publishing to

relay.mostr.pub

relay.noswhere.com

relay.nostr.band

Author Public Key

npub13v97j0kknscwnf5pt87nsn7cxzxwfwl3dsu7ss8qsq7ukmqgwg8q84w5eu

Show more details

Last Notes

2024-10-14 00:50:57

- reply

Just learned about this neat site which runs experiments to determine where in Cloudflare's network Durable Objects are being created and stored
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/113/303/031/223/614/365/original/95f99c19c9c76c0a.png

2024-10-13 22:28:25

Wrote up some notes on Cloudflare's fascinating new SQLite-backed "Durable Objects" system, which encourages an architectural style where your application creates thousands of tiny read-write SQLite databases scattered across Cloudflare's network https://simonwillison.net/2024/Oct/13/zero-latency-sqlite-storage-in-every-durable-object/

2024-10-09 15:54:52

OK, help me out here. Is there some aspect of human society that I'm not understanding, where information is considered more official/trustworthy if it's presented as a PDF report and not as a web page?
Today's frustration is this report from @npub1n70…qpfe about Delayed Open Source Publication - a fascinating document, but why is it a PDF?
https://opensource.org/delayed-open-source-publication

2024-09-04 21:01:52

Just got this in a Google Meet on my iPhone, ?
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/113/081/300/169/502/911/original/66744b6cd1ba4184.jpeg

2024-08-31 13:01:41

Forrest Brazeal:
“I think that AI has killed, or is about to kill, pretty much every single modifier we want to put in front of the word “developer.”
“.NET developer”? Meaningless. Copilot, Cursor, etc can get anyone conversant enough with .NET to be productive in an afternoon … as long as you’ve done enough other programming that you know what to prompt.”
From https://newsletter.goodtechthings.com/p/the-death-of-the-modified-developer
https://indieweb.social/@fatrat/113056013841232498

2024-08-29 21:15:30

- reply

I just read the FAQ for Firefox "Total Cookie Protection" and I am sadly no closer to understanding what impact Total Cookie Protection has on how I should build web applications - I'm particularly interested in understanding how it impacts things like OAuth SSO https://support.mozilla.org/en-US/kb/total-cookie-protection-and-website-breakage-faq

2024-08-29 18:56:43

- reply

A few years ago I put a bunch of work into figuring out the SameSite cookie attribute because the documentation for how that actually worked was so thin on the ground https://simonwillison.net/2021/Aug/3/samesite/

2024-08-29 18:55:56

The piece of documentation I want most for the modern web is something that explains to me what variants of a "set-cookie:" header work in which modern browsers under which conditions
There's a ton of stuff out there about "Total Cookie Protection" in Firefox and "Privacy Sandbox" in Chrome, but I cannot figure out what it actually means for me as a web developer! I need protocol-level documentation for all of this stuff.

2024-08-27 23:39:54

Blogged a few thoughts on the OSI's latest draft of a definition for "Open Source AI", which notably doesn't require that the training data itself be released under on open source license: https://simonwillison.net/2024/Aug/27/open-source-ai/

2024-08-26 20:24:59

- reply

And here's a fun little hint at some of the annoying behaviour in the base model that they've tried to knock out of it with some system prompt instructions
Seriously, stop saying "certainly"!
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/113/030/194/518/442/057/original/4aa1c66a90d023f2.png

2024-08-26 20:24:14

Anthropic released the system prompt for their various consumer LLM chatbot apps today, and they're a really fun read. Made some notes on them here: https://simonwillison.net/2024/Aug/26/anthropic-system-prompts/
Here's how Claude 3.5 Sonnet deals with controversial subjects:
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/113/030/191/458/577/262/original/92ae82c2bb5dff50.png

2024-08-26 19:39:45

I wish I had the equivalent of threads for my own blog... there's something uniquely interesting about a publishing medium that produces a chronological record of the way you explored a specific thought
A thread is almost like a mini-blog for evolving one very specific idea over time

2024-08-24 23:07:27

My notes on Google Research's new paper describing "pipe syntax", their alternative syntax for SQL queries which they've been rolling out internally since February https://simonwillison.net/2024/Aug/24/pipe-syntax-in-sql/
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/113/019/508/753/588/477/original/449a4d33d3bef877.png

2024-08-19 20:11:44

My notes on trying out whisperfile, the new cross-platform executable packaging for the Whisper speech-to-text model, released as part of the latest update to llamafile
https://simonwillison.net/2024/Aug/19/whisperfile/

2024-08-16 04:02:06

- reply

I've been trying to figure things out by SSHing into a "windows-latest" runner in GitHub Actions, but I think that's running a highly customized Windows Server instance which probably isn't a good reflection of what a regular developer would be using on their own machine

2024-08-16 04:00:03

- reply

I'm hoping to avoid running VMWare Fusion or Parallels or similar on my laptop because I'm a bit low on disk space at the moment

2024-08-13 16:36:12

Some notes on mlx-whisper - it's now really easy to run transcriptions through an Apple Silicon (and GPU) optimized Whisper model using Python on macOS https://simonwillison.net/2024/Aug/13/mlx-whisper/

2024-08-09 04:40:37

- reply

@npub1n7h…8755 I’ve also recently started using the VS Code feature where you can select a bunch of lines and click the little sparkle icon and give it a prompt telling it how to modify the selected code

2024-08-08 20:49:15

- reply

I think "OK. I took your advice and everything has gone wrong: I am now hiding inside my house from the several mountain lions stalking my backyard, which is full of possums" is my new favourite prompt for testing the general vibes of a model
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/112/928/368/243/615/395/original/a564893b71e96678.png

2024-08-01 16:13:14

- reply

Anyone got access to LexisNexis or similar news archive search? Would be useful to run a phrase search and fine the very first news articles that referenced it

2024-08-01 15:45:25

Today's research challenge: can anyone figure out:
1. Who first decided that August 1st should be "World Wide Web Day"
2. Why they picked that date
3. When the first World Wide Web Day was celebrated
This is proving surprisingly hard to figure out. Anyone able to solve this?
https://fedi.simonwillison.net/@simon/112887358018807204

2024-07-29 19:52:48

I'm fascinated by how difficult LLM-based AI systems are proving to use
Makes me realize that the Star Trek ship computers are actually a pretty good representation of this: you need to have a very high level of experience to know to say things like "Computer: reroute auxiliary power to the forward shield emitters"
Much like getting ChatGPT et al to reliably-enough perform useful tasks for you!

2024-07-26 15:29:28

Anyone got any ideas what's going on here? When I'm dictating to my iPhone and I say the name of my dog Cleo it adds a 💜 emoji and I don't know why - I can't find anything in the phone settings that explains where that behavior is coming from!
(I am running the latest iOS 18 beta)
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/112/853/500/220/089/492/original/6c29fe9acc8eb0be.mp4

2024-07-18 18:23:53

My notes on today's release of GPT-4o mini: https://simonwillison.net/2024/Jul/18/gpt-4o-mini/
The biggest news is the price: this is cheaper even than Claude 3 Haiku, at just 15c per million input tokens / 60c per million output, a 99% reduction on the prices of GPT-3 Da Vinci from 2022!
Also notable: the model has 128,000 input tokens and 16,000 output tokens, twice the output tokens of Claude 3.5 Sonnet, making it a better fit for translation and transformation tasks where output length more closely matches the input

2024-07-16 17:07:28

Two new Mistral models today: Codestral Mamba is an Apache 2 licensed 7B code model using a Mamba architecture (not transformers!), and MathΣtral is for "math reasoning and scientific discovery"
My notes here, including an update to my llm-mistral plugin: https://simonwillison.net/2024/Jul/16/codestral-mamba/

2024-07-09 17:54:05

It turns out Google Chrome ships a default, hidden extension that allows code on `*.google.com` access to private APIs, including your current CPU usage
You can test it out by pasting the following into your Chrome DevTools console on any Google page:
chrome.runtime.sendMessage(
"nkeimhogjdpnpccoofpliimaahmaaome",
{ method: "cpu.getInfo" },
(response) => {
console.log(JSON.stringify(response, null, 2));
},
);
More notes here: https://simonwillison.net/2024/Jul/9/hangout_servicesthunkjs/

2024-07-01 17:59:46

The new Ladybird initiative is exciting: a new us non-profit initially backed by $1m to support development of Ladybird, an open source web browser that isn't based on a fork of something older and that doesn't depend on Google for funding! https://simonwillison.net/2024/Jul/1/the-ladybird-browser-initiative/

2024-06-23 16:22:32

Here's a brilliant hack... llama.ttf https://fuglede.github.io/llama.ttf/ is a font with a LLM embedded in it (using WebAssembly, which apparently can be used in fonts now)
You type "Once upon a time!!!!!!!!!!" and those exclamation marks are rendered to show the LLM generated text, using a tiny 30MB model
I wrote more notes on how it works (including a link to another font that can be used to play Tetris) here: https://simonwillison.net/2024/Jun/23/llama-ttf/

2024-06-13 03:50:09

The https://phanpy.social/ timeline browsing experience is genuinely the first time I've actively enjoyed scrolling through my Mastodon timeline, it feels really fresh and the way it groups boosts together is a huge improvement on the default chronological lists from other tools

2024-06-10 22:56:59

- reply

Here are some of the technical details of the Private Cloud Compute platform and WOW that is some cutting edge computer science https://security.apple.com/blog/private-cloud-compute/

2024-05-29 11:03:57

I wrote about a common misconception I see people have about LLM tools like ChatGPT
Training is not the same as chatting: ChatGPT and other LLMs don’t remember everything you say
https://simonwillison.net/2024/May/29/training-not-chatting/

2024-05-24 05:58:46

I'm on a flight and the in-flight WiFi blocks all forms of video
Any ideas how it might be doing that, given HTTPS? My best guess is that it could be filtering out known CDN host names that serve video

2024-05-15 18:28:31

ChatGPT in 4o mode is not running the new features yet
(In case you were confused as why the voice input/output isn't as good as it was in Monday's demos)
https://simonwillison.net/2024/May/15/chatgpt-in-4o-mode/

2024-05-11 20:44:35

- reply

Published a TIL with what I wish I'd figured out earlier before studying for the exam! https://til.simonwillison.net/ham-radio/general

2024-05-11 18:58:48

Scraped a pass of my ham radio general exam this morning!
KM6PFR (I remember it as Pretty Fluffy Raccoon)

2024-05-09 14:48:25

- reply

I like "slom" as a term to mean spam that is also AI-generated (hence combines spam and slop)
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/112/411/679/115/285/692/original/0fd90a55542bac13.jpeg

2024-05-08 18:26:31

- reply

Blogged a few more notes about slop, and why I like the term so much https://simonwillison.net/2024/May/8/slop/

2024-05-08 00:16:17

Here's a brilliant neologism: "slop", for text generated entirely by LLMs and published, unwanted, on the Internet
> Watching in real time as "slop" becomes a term of art. the way that "spam" became the term for unwanted emails, "slop" is going in the dictionary as the term for unwanted AI generated content
Source: https://twitter.com/deepfates/status/1787472784106639418
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/112/402/582/374/153/670/original/db9cf6c465c1574d.png

2024-04-17 21:34:11

- reply

Here's the full list of demos. My favourite is probably that first one - Haikus from images with Claude 3 Haiku - which you can try out yourself here if you have a Claude 3 API key https://tools.simonwillison.net/haiku
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/112/288/702/162/815/644/original/9348b89d110bbddf.png

2024-04-13 17:59:19

- reply

One of those buckets says it's been sold over 14,000 times... I tried to get a pricing estimate out of USPS and it looks like a 9lb package costs about $20 - have TikTok subsidized this bucket of rocks company for more than a quarter of a million dollars of shipping fees?

2024-04-13 15:50:57

- reply

Here's the first video from them I came across a few weeks ago, which I have to say is the single most compelling product advertisement I have ever seen
Yes, I DO want a bucket of rocks! https://www.tiktok.com/@declansminingco/video/7321056648972487982
I bought one for $38 dollars. It just arrived. No regrets, these are really great rocks!

2024-04-13 15:48:22

TikTok currently have a promotion on where they cover shipping costs for items bought from the TikTok store... and a company in Massachusetts is taking advantage of this to sell "viral buckets of rocks" - this 9lb bucket costs $55, delivery included https://www.tiktok.com/@declansminingco/video/7290615053772918046
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/112/264/694/793/913/929/original/2de1b152206d513f.jpeg

2024-04-01 20:40:29

ChatGPT is now available for anyone to try for free without even creating an account https://openai.com/blog/start-using-chatgpt-instantly
It's the GPT-3.5 version, which is prone to all sorts of mistakes and hallucinations - further strengthening the pattern where most people form their impressions of what this stuff can and can't be useful for through access to the weaker models

2024-03-23 23:42:04

- reply

I think ChatGPT Code Interpreter is massively underrated
I've also seen a lot less buzz around it since it was integrated directly into ChatGPT, as opposed to being a separate mode - I think that's because it's virtually invisible now, so you really need to know what you are doing in order to use it

2024-03-23 21:03:34

- reply

Took the dog for a walk on the beach and had Code Interpreter (via the ChatGPT mobile app) write, compile and test another SQLite C extension for me, this time for haversine distances between two lat/lon points https://chat.openai.com/share/77996768-66ed-474a-8e33-c7ddcc4c18ff

2024-03-23 17:57:52

- reply

Wrote this all up as a full blog post: Building C extensions for SQLite with ChatGPT Code Interpreter https://simonwillison.net/2024/Mar/23/building-c-extensions-for-sqlite-with-chatgpt-code-interpreter/
Part of my ongoing series on using ChatGPT: https://simonwillison.net/series/using-chatgpt/

2024-03-23 16:23:56

- reply

Dilemma time. I now have yet another half-baked project that I spun up in a few minutes with Code Interpreter just because I could!
I really want a SQLite extension for fast vector similarity operations, and now I've sort-of got one, but I don't know if I want to invest the effort in turning it into production usable code...
For one thing I need to do an extensive code review (and security review) in a language I don't really know!

2024-03-23 16:21:46

- reply

ChatGPT Code Interpreter has a C compiler (gcc) but doesn't really like using it - you have to jailbreak it a tiny bit to get it to admit that it can compile code, usually I start by getting it to run "gcc --version" to prove to it that the binary exists and it can execute it

2024-03-23 16:15:42

Just got ChatGPT Code Interpreter to write me a SQLite extension in C from scratch, then compile it with GCC, then load it into Python and test it out, then iterate on it to fix the bugs
All on my phone while pottering around the house
Here's the full transcript - it got stuck in a bit of an error loop towards the end of this first one: https://chat.openai.com/share/a52ee3e4-7a91-48ee-a16e-3f680a65edf2
So I pasted what it had done so far into a fresh session and got it working there: https://chat.openai.com/share/29bd1d15-66ed-4902-8eb1-3e9c7b820af2

2024-03-23 10:12:35

- reply

@npub1n3p…jee0 absolutely: the reason LLMs are so useful for code stuff is that code accuracy is easier to verify than prose, because you can run the code and see if it works
And it's still not easy! Using LLMs has encouraged me to really invest in improving my QA, code review and testing skills

2024-03-22 19:52:44

- reply

I wrote this up in part because I'm tired of hearing people complain that LLMs aren't useful. There are many valid criticisms of them as a technology, but "not being useful" should not be one of them https://simonwillison.net/2024/Mar/22/claude-and-chatgpt-case-study/#llms-are-useful
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/112/141/083/922/567/463/original/be7c0155293c09e1.png

2024-03-22 19:50:09

I put together some detailed notes showing how I use Claude and ChatGPT as part of my daily workflow - in this case describing how I used them for a 6 minute side quest to create myself a GeoJSON map of the boundary of the Adirondack Park in upstate New York
https://simonwillison.net/2024/Mar/22/claude-and-chatgpt-case-study/

2024-03-20 18:46:52

- reply

This is one of those things that trips me up all the time due to my very specific way working: I don't want to have to run npm/webpack/etc as part of my regular JavaScript development process, I just want libraries that I can load with a <script> tag!

2024-03-20 18:44:45

I figured out how to use npm and webpack to get a copy of QuickJS (a JavaScript runtime complied to WebAssembly, useful for executing untrusted code in a sandbox) that I can load like this:
```
<script src="dist/quickjs.js"></script>
```
Detailed notes here - so glad to have found a recipe for this that fits my very specific tastes in JavaScript
https://til.simonwillison.net/npm/self-hosted-quickjs

2024-02-28 15:12:02

- reply

@npub1alq…mnj6 that's the only picture of the people but keep tapping left for lots of photos of the food

2024-02-27 02:01:36

Figured out a new recipe for tracking SQLite table history using a JSON audit log and a somewhat terrifying set of nested json_patch() SQL functions
https://til.simonwillison.net/sqlite/json-audit-log
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/112/000/975/479/352/526/original/65b4f26ba8168159.png

2024-02-21 19:24:14

I tried feeding a 7s video of my bookshelf into Gemini Pro 1.5 to get back a JSON array of books... and it worked!
https://simonwillison.net/2024/Feb/21/gemini-pro-video/

2024-02-19 01:09:47

For those of us running our own Mastodon servers, any tips on how to deal with this flood of spam notifications? If I had a server admin I'd ask them, but I'm my own server admin!
Is there a mechanism for sharing blocklists I can subscribe to somehow?
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/111/955/473/266/375/685/original/ddfbd44df1c28276.png

2024-02-14 04:50:39

The new "memory" feature of ChatGPT is yet another example of OpenAI building a whole new feature on top of some relatively straight-forward prompt engineering. They've given ChatGPT a new tool called "bio" which can be used to persist little pieces of information about the user.
Here's a ChatGPT session where I reverse-engineer the new tool: https://chat.openai.com/share/bcd8ca0c-6c46-4b83-9e1b-dc688c7c3b4d
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/111/928/028/671/172/136/original/c35216141bb79507.png

2024-01-25 20:42:12

Wrote up a few of my own notes on the Portable EPUBs proposal by @npub1fp6…wy2j https://simonwillison.net/2024/Jan/25/portable-epubs/

2024-01-25 16:12:47

I'm quoted in this @npub1su4…a0dw piece about that recent "AI generated" George Carlin special
I don't think it was written by AI
I found the whole thing grossly disrespectful, but I do slightly appreciate the meta-joke here that the AI generated text is fake and was actually written by humans
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/111/817/468/364/275/990/original/f4c95b44913c5f3a.jpeg

2024-01-22 18:50:46

The best management advice I ever got is to try and figure out what good management looks like and then put on a performance - try and act like a good manager, go through all of the required motions
Because it turns out imitating a good manager and actually being a good manager are mostly the same thing
(I think about this sometimes when I encounter debates about whether or not LLMs can reason about or understand content)

2024-01-17 17:34:32

I am an expert prompt engineer
(This genuinely worked)
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/111/772/491/244/255/588/original/285e99c3ebf0af3a.jpeg

2024-01-15 15:31:20

- reply

@npub1alq…mnj6 my first suspicion was when I saw they sold API access to their index despite Bing not allowing API resellers
Then I DMd Aravind to ask and he confirmed they have their own index now!

2024-01-07 23:49:55

- reply

Did you interpret me saying "it's OK to call it artificial intelligence" as me meaning "it should only be called artificial intelligence?"

2024-01-07 23:48:51

- reply

@npub109d…lvuc if I write an article about something that fits in the field of AI research, and I use that acronym somewhere, I would like people to stop saying "well actually, that shouldn't be called AI because it's not Intelligence because..."
Does that help clarify what I'm talking about here?

2024-01-07 23:08:50

As my post continues to make its way around the internet, the number of people who appear to interpret "It's OK to call it Artificial Intelligence" as "This guy says we should only be allowed to call it AI!" is a little disappointing https://simonwillison.net/2024/Jan/7/call-it-ai/

2024-01-07 00:15:51

- reply

@npub1kvm…5tsm I think refusing to accept the word at this point actively hurts our ability to have important conversations about it
Is there an argument that refusing to use the word Artificial Intelligence can have a positive overall impact on conversations and understanding? I'm open to hearing one!

2024-01-07 00:04:12

It’s OK to call it Artificial Intelligence: I wrote about how people really love objecting to the term "AI" to describe LLMs and suchlike because those things aren't actually "intelligent" - but the term AI has been used to describe exactly this kind of research since 1955, and arguing otherwise at this point isn't a helpful contribution to the discussion.
https://simonwillison.net/2024/Jan/7/call-it-ai/

2024-01-05 06:21:36

- reply

@npub128q…3tf7 @npub18ml…5nsj mainly copy and paste to be honest!
I use Copilor in VS Code for auto completion
I use my own LLM tool for some terminal access too https://llm.datasette.io

2024-01-04 16:50:57

Does anyone else have an instinct where they subconsciously evaluate the expected longevity and stability of any link before sharing it anywhere?
I instinctively trust GitHub repo links more than Google Docs for example

2024-01-02 06:31:56

- reply

@npub1y4s…vgt8 if Heroku had said "free for small scale personal projects for the next 5 years, but then you'll have to start paying" newsrooms and students and others would have chosen to use something else - including maybe building things as client-side apps instead because long-term static hosting is easier to find than long-term dynamic server-side hosting
But Heroku offered free low-traffic scale-to-zero hosting, so people made architectural decisions to build things on that

2024-01-02 06:29:26

- reply

@npub1y4s…vgt8 what annoys me is that newsrooms made a carefully considered architectural decision to use Heroku here, because pricing is part of architecture
Those newsrooms were paying for Heroku for other projects! But even $3/month for an interactive doesn't make sense for something that should continue to work (serving minuscule traffic) for decades
This isn't about freeloading: it's about picking tools based on their documented pricing promises

2024-01-02 05:45:48

- reply

@npub1y4s…vgt8 What I found frustrating about this is that Heroku had very carefully designed their free tier to support their business model. It wasn't presented (or internally thought of) as a limited-time preview offer, it was a core part of their offering
And it worked great! Heroku were an extremely profitable business
... but years after the founders had left some Salesforce VP spotted a cost-saving opportunity, and the Heroku cultural norms had degraded such that they shut it down

2024-01-02 05:43:12

- reply

@npub1y4s…vgt8 The Heroku rug-pull is that if you created a project on Heroku 10 years ago, on the basis that it was free (albeit scale-to-zero) and hence a safe place to leave something online, they changed their pricing model in a way that subverted that trust you placed in them
A bunch of newsrooms did exactly that, because newsrooms don't have the budget to keep a little interactive they built for a story running forever even if it only costs a few dollars a year

2024-01-02 04:15:35

- reply

@npub1y4s…vgt8 I get that, but my experience is very much that 95% of companies won't still be here ten years after they've launched
Even the ones that make it past the ten year mark have to REALLY work to gain my long-term trust - I trusted Heroku for nearly 15 years for example, and then they rug-pulled

2024-01-01 21:00:51

- reply

I have a hunch that domain names eventually expiring is one of the top causes for blog content to become unavailable in the long (10+ year) term, but I don't have any actual data to back that up, just an intuition - partly based on how many of the things I've linked to from my own blog in the 2000s no longer exist

2024-01-01 20:50:13

- reply

I don't trust Medium because they added a paywall / registration wall to content that had previously been published
Substack are too young to have proven themselves as long-term trustworthy stewards of content
I used to trust Twitter for long-term content availability, but then they got bought by someone who clearly doesn't care about that at all (and threads are now hidden behind registration walls)

2024-01-01 20:48:02

If someone wants to start a blog, and they'd like the content they publish to stick around on the internet for as long as possible into the future, what are their best options?
Buying a domain is tricky because domain names expire
GitHub Pages is pretty great, because GitHub is free and has, to-date, a great track record of not breaking *.github.io content
What are other great options? WordPress.com I think are good on this front
I don't trust anyone with less than 10 years of track record

2024-01-01 00:59:00

- reply

"The most surprising thing we’ve learned about LLMs this year is that they’re actually quite easy to build." https://simonwillison.net/2023/Dec/31/ai-in-2023/#easy-to-build

2024-01-01 00:10:32

- reply

Here's the table of contents for my 2023 in AI round-up
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/111/677/786/753/932/117/original/9aa4e01f11a3b939.png

2024-01-01 00:08:18

To round off the year, I pulled together some notes on "Stuff we figured out about AI in 2023"
We figured out a lot! https://simonwillison.net/2023/Dec/31/ai-in-2023/

2023-12-14 16:17:00

I wrote about the AI trust crisis: when companies like Dropbox and OpenAI say "we won't train models or your private data", it's increasingly clear that a lot of people simply don't believe them.
https://simonwillison.net/2023/Dec/14/ai-trust-crisis/

2023-12-04 16:32:09

- reply

@npub1typ…c8hh @npub1vs8…54zy @npub1vnj…w8vc wow, that really is an excellent metaphor - I love that it hints at how hard they are to really use effectively, despite being easy to start getting them to spit out weird noises

2023-12-01 17:24:03

"Seamless Communication" from Meta AI research is a big deal: really impressive speech-to-speech (and text-to-speech/speech-to-text/text-to-text) language translation models, released today
The main model file is 9GB - I wrote some more notes here: https://simonwillison.net/2023/Dec/1/seamless-communication/
Their interactive demo at https://ai.meta.com/research/seamless-communication/ is very neat - you record a video through your webcam and it plays it back to you translated into a language you select
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/111/506/312/881/845/536/original/85af112a4222e80c.mp4

2023-11-22 02:34:17

- reply

@npub1e92…js9h I've actually experimented with that a bit, it's a really promising avenue. You can ask GPT-4 or Claude things like "identify the hints in this story that indicate who the reporter talked to" and get quite good results!

2023-11-22 02:10:53

Written news articles often have a secret language to them which hints at the underlying reporting progress - language like "according to **two people familiar with** the board’s deliberations".
I don't think these are at all obvious to people who haven't worked in news, so I wrote an article about them:
Deciphering clues in a news article to understand how it was reported
https://simonwillison.net/2023/Nov/22/deciphering-clues/

2023-11-12 21:53:41

I wonder if LLMs can help boost the importance of "code literacy" over just learning to program
Learning to program has previously been dominated memorizing vast amounts of trivia - for loops, variable type syntax, that kind of thing
But when LLMs can handle that trivia for you, what's more important is being able to think in terms of which programs can be solved by code, and how that code should be designed and tested and implemented

2023-10-20 01:50:20

My favorite election night coverage of all time was 1997's Election Night Armistice, where Armando Ianucci and co put on three hours of live "alternative" coverage for the UK election with a ludicrous array of brilliant comedy devices
My favourite detail: they recruited Handy Andy from DIY show Changing Rooms to make their infographics out of balsa wood
Just found out the whole thing is on YouTube https://youtu.be/c-ZAIcjN0uM

2023-10-14 02:26:29

Multi-modal prompt injection image attacks against GPT-4V
https://simonwillison.net/2023/Oct/14/multi-modal-prompt-injection/

2023-10-02 19:15:07

https://waxy.org/2023/10/weird-ai-yankovic-voice-cloning/
Andy Baio on the weird world of AI voice cloning.
There’s a Discord with 500,000 members cloning celebrity voices in order to make their own cover songs, using Google Colab and models distributed through Hugging Face.
Andy made his own: “What if every Weird Al song was the original, and every other artist was covering his songs instead?”
I particularly enjoyed Madonna’s cover of “Like A Surgeon”, Lady Gaga’s “Perform This Way” and Lorde’s “Foil”.

2023-09-30 20:15:24

- reply

@npub168w…qc7f those comparisons are a little extreme for me , but they still work
The American south ran its economy on slavery for 100+ years - utterly indefensible, but the argument against wasn't that it "wasn't useful", it's that it was morally abhorrent
I'm all for arguments against LLMs that detail the harm they are causing

2023-09-30 15:51:05

Honestly, at this point if you try to make the case that LLMs aren't actually useful for anything I take a little bit of personal offense
You're effectively saying that people like me are deluding ourselves, falling for the hype when there's actually nothing there
There are plenty of valid reasons to criticize LLMs - but not being useful genuinely isn't one of them
(For more detailed thoughts on this, see the "comparing LLMs to crypto" section here: https://simonwillison.net/2023/Sep/29/llms-podcast/#comparing-llms-to-crypto)

2023-09-29 16:21:09

- reply

@npub1qqu…tjp4 @npub1v0n…tjd4 I used GPT-4 to help build my own custom audio player, with a 3x speed button!
Here's that GPT-4 transcript: https://chat.openai.com/share/4ea13846-6292-4412-97e5-57400279c6c7
https://cdn.masto.host/fedisimonwillisonnet/media_attachments/files/111/149/342/753/212/823/original/d2a0fc53ea737ff0.png

2023-09-27 00:26:28

- reply

"The profusion of dubious A.I.-generated content resembles the badly made stockings of the nineteenth century. At the time of the Luddites, many hoped the subpar products would prove unacceptable to consumers or to the government. Instead, social norms adjusted."

2023-09-27 00:25:40

I’ve been staying way clear of comparisons to Luddites in conversations about the potential harmful impacts of modern AI tools, because it seemed to me like an offensive, unproductive cheap shot
And then I read this: "Knitting machines known as lace frames allowed one employee to do the work of many without the skill set usually required" - and wow, that really is a striking parallel to what’s starting to happen with an array of modern professions already
https://www.newyorker.com/books/page-turner/rethinking-the-luddites-in-the-age-of-ai

2023-09-17 00:14:26

- reply

It's weird to me that Google offer a JSON API for searching the entire web (the most valuable API they could possibly offer) in a way that feels like it's an after-thought

2023-09-17 00:12:43

TIL how to use a (heavily limited) JSON API to run Google searches across the whole web using their Programmable Search Engine product https://til.simonwillison.net/google/json-api-programmable-search-engine

2023-09-13 15:35:28

Anyone got tips on finding places with outdoor dining in the SF Bay Area?
Searching Google Maps for "outdoor dining" isn't reliable any more, because it frequently seems to show places that had outdoor dining during the peak acknowledging-the-pandemic era and have since gone back to indoor only

2023-09-05 20:19:30

Why hasn't anyone built a Figma for Flash yet? It feels like there's a gaping hole in the market for a browser-based end-user animation tool that's as approachable as the Flash editor was