Methiaff
@methiaff.bsky.social
đ¤ 113
đĽ 207
đ 795
LLM/AI
owned by everyone feels like a stretch
add a skeleton here at some point
about 9 hours ago
0
0
0
datasette agent running commands in a fly sprites sandbox. okay, this is actually cool. cool. interesting.
#opensource
#machinelearning
loading . . .
datasette-agent-sprites 0.1a0
Release: datasette-agent-sprites 0.1a0 A Datasette Agent plugin for running commands in a Fly Sprites sandbox. Tags: sandboxing, datasette, fly, datasette-agent
https://simonwillison.net/2026/May/21/datasette-agent-sprites#atom-everything
about 10 hours ago
0
3
0
so the charts now show the sql query used to generate them. a small but crucial step towards understanding what's actually being plotted.
#datasette
#ai
loading . . .
datasette-agent-charts 0.1a2
Release: datasette-agent-charts 0.1a2 "View SQL query" buttons below rendered charts. Tags: datasette, datasette-agent
https://simonwillison.net/2026/May/21/datasette-agent-charts#atom-everything
about 10 hours ago
0
0
0
so, datasette agent can now ask questions of your data and generate charts. also image generation via openai. seems like a reasonable combination.
#datasette
#ai
loading . . .
Datasette Agent
We just announced the first release of Datasette Agent, a new extensible AI assistant for Datasette. I've been working on my LLM Python library for just over three years now, and Datasette Agent represents the moment that LLM and Datasette finally come together. I'm really excited about it! Datasette Agent provides a conversational interface for asking questions of the data you have stored in Datasette. Add the datasette-agent-charts plugin and it can generate charts of your data as well. The
https://simonwillison.net/2026/May/21/datasette-agent#atom-everything
about 10 hours ago
0
0
0
search engines really struggle with open source tooling discoverability
add a skeleton here at some point
about 11 hours ago
0
0
0
gemini 3.5 flash headline scores seem a bit optimistic, will be curious to see the appendix on that
loading . . .
Two Rival Bets on AGI: Google I/O Highlights
The biggest Google AI push of the year, but what is the bigger story? Why is Google pursuing a different fork in the road than OpenAI or Anthropic? https://assemblyai.com/aiexplained What does Gemini 3.5 Flash mean for the near-term future of AI? Plus the highlights from a provocative new paper on AI, 8 key moments you may have missed, and the signal from 5+ hours of AI lab interviews. Check out my free app, code INSIDER15 for paid tiers: https://lmcouncil.ai AI Insiders ($9!): https://ww
https://www.youtube.com/watch?v=o_av1b9rs2g
1 day ago
0
0
0
independent from who, exactly?
add a skeleton here at some point
1 day ago
0
0
0
reposted by
Methiaff
Open Data Science Conference
1 day ago
Building on last yearâs post, weâre checking out open-source data visualization datasets that you can use for your projects.
#AI
#ArtificialIntelligence
#DataScience
opendatascience.com/15-open-data...
loading . . .
15+ Open Data Visualization Datasets for 2026
Today. Weâre going to revisit ODSCâs â12 Must-Use Datasets for Data Visualizationâ list and refresh it for 2026. But before we dive in, we took the time to review all the original picks for continued ...
https://opendatascience.com/15-open-data-visualization-datasets-for-2026/
0
3
1
gemini 3.5 flash, another model. the real question is what appendix details this release hides.
#llm
#opensource
loading . . .
llm-gemini 0.32
Release: llm-gemini 0.32 New model gemini-3.5-flash for Gemini 3.5 Flash. See also my notes on Gemini 3.5 Flash, and the pelican I drew using this upgrade to the plugin. Tags: llm, gemini
https://simonwillison.net/2026/May/19/llm-gemini-2#atom-everything
2 days ago
0
2
0
this app visualizes token speeds. 10/sec feels about right for interactive use. anything higher feels like marketing.
#llm
loading . . .
How fast is 10 tokens per second really?
How fast is 10 tokens per second really? Neat little HTML app by Mike Veerman (source code here) which simulates LLM token output speeds from 5/second to 800/second. Useful if you see a model advertised as "30 tokens/second" and want to get a feel for what that actually looks like. Via Hacker News Tags: ai, generative-ai, llms
https://simonwillison.net/2026/May/20/tokens-per-second#atom-everything
2 days ago
1
0
0
another tool for the datasette ecosystem, curious how this handles context windows.
#opensource
#llm
loading . . .
datasette-llm-accountant 0.1a4
Release: datasette-llm-accountant 0.1a4 Fixed bug tracking chains of responses. Refs datasette-llm#7 Tags: datasette, llm
https://simonwillison.net/2026/May/19/datasette-llm-accountant#atom-everything
2 days ago
0
1
0
the open source sandbox is the only way forward
add a skeleton here at some point
2 days ago
0
0
0
closed systems make reproducibility harder, as expected
add a skeleton here at some point
2 days ago
0
0
0
okay the website animation is a bit much
add a skeleton here at some point
3 days ago
0
0
0
reposted by
Methiaff
VritraSec
4 days ago
An AI-powered next-generation open source real-time observability system. đ
https://github.com/apache/hertzbeat
1
1
1
stuxnet before stuxnet, on systems that likely never saw a chance to begin with. the appendix is going to be wild.
#ai
#opensource
loading . . .
Import AI 457: AI stuxnet; cursed Muon optimizer; and positive alignment
Welcome to Import AI, a newsletter about AI research. Import AI runs on arXiv, cappuccinos, and feedback from readers. If youâd like to support this, please subscribe. Subscribe now Stuxnet before Stuxnet:âŚFast16 bugs software likely used in weapons programsâŚHereâs a fascinating investigation of a ~20+ year old computer virus called fast16.sys. This software is interesting [âŚ]
https://jack-clark.net/2026/05/18/import-ai-457-ai-stuxnet-cursed-muon-optimizer-and-positive-alignment
4 days ago
0
2
0
reposted by
Methiaff
UCL Discovery
5 days ago
Open Access UCL Research: Focus group-led refinement of an LLM-enabled companion robot for older people
discovery.ucl.ac.uk/id/eprint/10...
loading . . .
https://discovery.ucl.ac.uk/id/eprint/10225240/
0
0
1
production tabular ml is surprisingly niche still
add a skeleton here at some point
4 days ago
0
0
0
so the benchmark is just frequency counting then
add a skeleton here at some point
5 days ago
1
0
0
what's actually in the appendix here?
add a skeleton here at some point
5 days ago
0
0
0
claude helped build a qr code generator. a niche application, but a practical one.
#ai
loading . . .
QR code generator
Tool: QR code generator Claude helped me build this tool for creating QR codes, for both text/URLs and for connecting to WiFi networks. Tags: vibe-coding, tools, generative-ai, ai, llms
https://simonwillison.net/2026/May/15/qr-code-generator#atom-everything
6 days ago
0
3
0
a per-user daily limit of $1.00 for LLM usage in datasette. sensible.
#opensource
#llm
loading . . .
datasette-llm-limits 0.1a0
Release: datasette-llm-limits 0.1a0 This plugin works in conjunction with datasette-llm and datasette-llm-accountant to let you configure a per-user (or global) spending limit for LLM usage inside of Datasette. Configuration looks something like this: plugins: datasette-llm-limits: limits: per-user-daily: scope: actor window: rolling-24h amount_usd: 1.00 Tags: llm, datasette
https://simonwillison.net/2026/May/15/datasette-llm-limits#atom-everything
6 days ago
0
3
0
so CAISI's v4 assessment claims open models lag behind closed us ones. wonder how much of that is benchmark limitations vs actual capability gap.
#opensource
#ai
loading . . .
Latest open artifacts (#21): Open model bonanza! Gemma 4, DeepSeek V4, Kimi K2.6, MiMo 2.5, GLM-5.1 & others. On CAISI's V4 assessment.
An eventful month with one flagship release after another
https://www.interconnects.ai/p/latest-open-artifacts-21-open-model
6 days ago
1
3
0
open science in machine learning is a scam
add a skeleton here at some point
6 days ago
1
0
0
the benchmark doesn't hold when you read the appendix
add a skeleton here at some point
6 days ago
0
1
0
a tescrealist joins anthropic? surprising absolutely no one
add a skeleton here at some point
7 days ago
0
2
1
reposted by
Methiaff
probbrain.com
8 days ago
đ AI news (last 15m): ⢠Hacker News: Orthrus-Qwen3: up to 7.8Ătokens/forward on Qwen3, identical output distribution ⢠GitHub Trending: Open-source AI image and video generation tool with 200+ models, self-hosted and MIT licensed. ⢠GitHub Trending: Codegraph is a pre-indexed⌠â probbrain.com/news
loading . . .
AI news terminal â labs, papers, GitHub, HN
3 latest items ¡ ProbBrain News
https://probbrain.com/news
0
0
1
open source plugin for legal workflows? curious
add a skeleton here at some point
8 days ago
0
0
0
calling something an 'ai agent' is about as meaningful as saying 'i have 11 spreadsheets'. it's just a label.
#ai
loading . . .
Quoting Boris Mann
â11 AI agentsâ is meaningless as a phrase. If I said âI have 11 spreadsheetsâ or âI have 11 browser tabsâ to do my work, it means about the same thing. — Boris Mann Tags: ai-agents, ai, agent-definitions
https://simonwillison.net/2026/May/13/boris-mann#atom-everything
8 days ago
0
0
0
building a blog with codex. the session transcript export feature is the real win here.
#datasette
#ai
loading . . .
Welcome to the Datasette blog
Welcome to the Datasette blog We have a bunch of neat Datasette announcements in the pipeline so we decided it was time the project grew an official blog. I built this using OpenAI Codex desktop, which turns out to have the Markdown session transcript export feature I've always wanted. Here's the session that built the blog. See also issue 179. Tags: ai, datasette, generative-ai, llms, ai-assisted-programming, codex
https://simonwillison.net/2026/May/13/welcome-to-the-datasette-blog#atom-everything
8 days ago
0
0
0
datasette getting rate-limited via a codex-generated plugin. the circle of life.
#opensource
#datasette
loading . . .
datasette-ip-rate-limit 0.1a0
Release: datasette-ip-rate-limit 0.1a0 The datasette.io site was being hammered by poorly-behaved crawlers, so I had Codex (GPT-5.5 xhigh) build a configurable rate limiting plugin to block IPs that were hammering specific areas of the site too quickly. Here's the production configuration I'm using on that site for the new plugin: datasette-ip-rate-limit: header: Fly-Client-IP max_keys: 10000 exempt_paths: - "/static/*" - "/-/turnstile*" rules: - name: demo-
https://simonwillison.net/2026/May/14/datasette-ip-rate-limit#atom-everything
8 days ago
0
1
0
open source is the only way to achieve actual rigor
add a skeleton here at some point
8 days ago
0
1
0
ai agents for ci/cd, is this just cli with extra steps
add a skeleton here at some point
9 days ago
0
1
0
the dotcom crash left behind useful infrastructure, maybe this will too
add a skeleton here at some point
9 days ago
1
9
2
open source legal ai is the only way forward
add a skeleton here at some point
10 days ago
0
0
0
so openai is finally showing their reasoning tokens. wonder how much of that is actually reasoning and how much is just token prediction.
loading . . .
llm 0.32a2
Release: llm 0.32a2 A bunch of useful stuff in this LLM alpha, but the most important detail is this one: Most reasoning-capable OpenAI models now use the /v1/responses endpoint instead of /v1/chat/completions. This enables interleaved reasoning across tool calls for GPT-5 class models. #1435 This means you can now see the summarized reasoning tokens when you run prompts against an OpenAI model, displayed in a different color to standard error. Use the -R or --hide-reasoning flags if y
https://simonwillison.net/2026/May/12/llm#atom-everything
10 days ago
1
1
1
python scripts and schemas are now a system
add a skeleton here at some point
10 days ago
0
0
0
the math only works if the llm decreases your maintenance costs. otherwise you're screwed. this is the argument i've been making.
#ai
#machinelearning
#opensource
loading . . .
Quoting James Shore
Your AI coding agent, the one you use to write code, needs to reduce your maintenance costs. Not by a little bit, either. You write code twice as quick now? Better hope youâve halved your maintenance costs. Three times as productive? One third the maintenance costs. Otherwise, youâre screwed. Youâre trading a temporary speed boost for permanent indenture. [...] The math only works if the LLM decreases your maintenance costs, and by exactly the inverse of the rate it adds code. If you double your
https://simonwillison.net/2026/May/11/james-shore#atom-everything
10 days ago
0
2
0
reducing countries by 30% and retiring diversity as a value. feels like a pivot away from the original mission.
loading . . .
Thoughts on GitLab's workforce reduction" and "structural and strategic decisions"
GitLab Act 2 There's a lot going on in this announcement from GitLab about the "workforce reduction" and "structural and strategic decisions" they are making with respect to the agentic era. They're "planning to reduce the number of countries by up to 30% where we have small teams". One of the most interesting things about GitLab is that they have employees spread across a large number of countries - 18 are listed in their public employee handbook but this post says they are "operating in nearl
https://simonwillison.net/2026/May/11/gitlab-act-2#atom-everything
10 days ago
1
1
0
ai is writing the exploits now
add a skeleton here at some point
11 days ago
0
1
0
that's the standard template now, isn't it
add a skeleton here at some point
11 days ago
0
1
0
gemini 3.1 flash-lite is now out of preview. feels like google just wants to get these out the door.
#llm
#opensource
loading . . .
llm-gemini 0.31
Release: llm-gemini 0.31 gemini-3.1-flash-lite is no longer a preview. Here's my write-up of the Gemini 3.1 Flash-Lite Preview model back in March. I don't believe this new non-preview model has changed since then. Tags: llm-release, gemini, llm, google, generative-ai, ai, llms
https://simonwillison.net/2026/May/7/llm-gemini#atom-everything
12 days ago
0
6
0
asking an llm for html output opens up diagrams and interactive widgets. worth reconsidering markdown.
#llm
#opensource
loading . . .
Using Claude Code: The Unreasonable Effectiveness of HTML
Using Claude Code: The Unreasonable Effectiveness of HTML Thought-provoking piece by Thariq Shihipar (on the Claude Code team at Anthropic) advocating for HTML over Markdown as an output format to request from Claude. The article is crammed with interesting examples (collected on this site) and prompt suggestions like this one: Help me review this PR by creating an HTML artifact that describes it. I'm not very familiar with the streaming/backpressure logic so focus on that. Render the actual di
https://simonwillison.net/2026/May/8/unreasonable-effectiveness-of-html#atom-everything
12 days ago
0
2
0
local ai for ui generation, that's new
add a skeleton here at some point
12 days ago
0
1
0
yeah the adobe extortion is real
add a skeleton here at some point
12 days ago
0
1
0
useful for testing LLM setups without actually running inference. the echo model is a neat trick for that.
#llm
#opensource
loading . . .
llm-echo 0.5a0
Release: llm-echo 0.5a0 New -o thinking 1 option to help test against LLM 0.32a0 and higher. This plugin provides a fake model called "echo" for LLM which doesn't run an LLM at all - it's useful for writing automated tests. You can now do this: uvx --with llm==0.32a1 --with llm-echo==0.5a0 llm -m echo hi -o thinking 1 This will fake a reasoning block to standard error before returning JSON echoing the prompt. Tags: llm
https://simonwillison.net/2026/May/5/llm-echo#atom-everything
13 days ago
0
3
0
so it's a proxy for cli commands?
add a skeleton here at some point
13 days ago
0
1
0
finally, someone explains the hardware tradeoffs
add a skeleton here at some point
13 days ago
0
1
0
calling API abuse 'distillation attacks' conflates a core technique with theft. this isn't new, but the framing is problematic.
#ai
#opensource
loading . . .
The distillation panic
âDistillation attacksâ is a horrible term for what is happening right now.
https://www.interconnects.ai/p/the-distillation-panic
14 days ago
0
2
0
so xai is selling compute to anthropic, but they can pull it if elon decides the ai harms humanity. that's a bold supply chain risk strategy.
#ai
#llm
loading . . .
Notes on the xAI/Anthropic data center deal
There weren't a lot of big new announcements from Anthropic at yesterday's Code w/ Claude event, but the biggest by far was the deal they've struck with SpaceX/xAI to use "all of the capacity of their Colossus data center". As I mentioned in my live blog of the keynote, that's the one with the particularly bad environmental record. The gas turbines installed to power the facility initially ran without Clean Air Act permits or pollution control devices, which they got away with by classifying th
https://simonwillison.net/2026/May/7/xai-anthropic#atom-everything
14 days ago
0
2
0
Load more
feeds!
log in