Closer to homes

Tornado Alley moves east. Here’s one I recorded live on my phone in February here in Bloomington, Indiana. Here is the trench of destruction dug by a later tornado that went through thick woodlands about twelve miles north of here.

Didn’t know there was one, but yeah

Chick-fil-A takes a stand against fast-food drive-thru AI trend.

Maybe

Is doomscrolling doomed?

Always a bad sign

NPR has a new privacy policy. Specifics.

Here’s a PageXray of the policy page:

That’s what it says at MoonPhase.today

Specifically, 26%.

I’ll consider them unreal if they’re not at IIW.

The UOR Foundation has plans to… Well, I can’t copy and paste the text, so here’s a screengrab:

I’ll say not. Quite.

AM I? | A Documentary About AI Consciousness.

Ever notice the groovy ditches in tidal swampland?

It’s called “grid ditching,” and is explained here:

Grid Ditching

The grid ditch system still evident on our salt marshes today, were dug by hand between 1928 and 1934. The primary purpose of this era of ditching was to put as many people to work as possible, as this was the time of the great depression. No entomological studies were conducted in conjunction with this ditching effort; mosquito control was a secondary consideration at best.

However mosquito control was achieved by default as practically every square inch of marsh was drained by the extensive project. Some engineering studies were done to determine where and at what intervals ditches were dug. Ditches were dug in straight rows by hand with sod saws and two man shovels. In 1934, at the peak of this ditching effort, over 11,000 men were employed digging ditches and when completed nearly 3,000 linear miles of salt marsh ditch were dug in Massachusetts alone.

Virtually all salt marsh in New England was ditched with the exception of one marsh in Rhode Island. Later in the late 1940s and early 1950s soldiers returned home from World War II. Housing shortages were a big problem in the more populated areas particularly in and around Boston. A generation of young families, eager to get their lives back on track migrated to the north shore of Massachusetts to start new lives.

This migration coincided with the degradation of the grid ditch system created in the 1930s. These ditches had not been maintained and now produced far more mosquitoes then they had initially eliminated. By many accounts it was so bad some considered the area to be almost uninhabitable. A few local programs were established to try to reopen the ditches but it was impossible to duplicate the labor force that had originally created the ditches.

In 1958, a grass roots effort, fueled by public demand for relief resulted in legislation establishing the Essex County Mosquito Control Project. For reasons unknown the project was not formally funded until 1965. Equipment was purchased and a major effort was launched to reclaim salt marsh ditches.

Salt Marsh Ditch Maintenance in Northeast Massachusetts

In the late 1970s, Walter Montgomery was an equipment operator for what was then, the Essex County Mosquito Control Project, ECMCP. Montgomery’s primary duty was to maintain the extensive salt marsh grid ditch system, excavated for the most part, between 1928 and 1934.

Montgomery recalls using an implement known as a scavel plow; this was a large wedge shaped device mounted under a wing plow and could be attached to the front of a tractor or sometimes towed behind. A scavel plow wedge was basically constructed to the original ditch dimension. The wedge would be dropped into the ditch and the tractor either pushed or pulled it along. As the wedge peeled spoil out of the ditch the wing plow would roll the spoil into furrows approximately six feet wide on both sides of the ditch. These furrows of spoil were then either run over to flatten them or plowed off the marsh.

On a productive day Montgomery recalls completing two or more miles of ditch maintenance but even at that rate maintenance of hundreds of linear miles of ditch was an endless endeavor. It was Montgomery’s experience that freshly maintained ditches were really only effective for about two years before requiring additional maintenance.

Typically ditches would become blocked on the high marsh and previously drained salt pannes would often reestablish. Montgomery observed that the open water areas of these pannes or ponds didn’t have mosquito larvae but the grassy sheltered areas around the edges or isolated depressions adjacent to the ponds, supported mosquito larvae.

Montgomery’s observations further supported his belief that continued maintenance of the grid ditch system was not productive and furthermore there had to be a better way to manage salt marsh mosquito populations.

That is from a .pdf titled MASSACHUSETTS MOSQUITO CONTROL OPEN MARSH WATER MANAGEMENT STANDARDS MAY 2010. It is worth continuing to read, if mosquito control is your jam.

Alas, as you may gather by now, everything is my jam.

Seems Flickr is having an outage of some kind:

 

If  you don’t, you’re a human AI

Test: See if you don’t laugh at something as you scroll down through this.

Am I alone in finding it annoying?

Does anybody want, or use, the “workspace” that shows up, whether you want it or not, every time you open a Zoom session?

Getting less faint

Wall Street Journal: The Global Bond Rout Is Getting Messy—Unwinding popular hedge fund trades helped drive U.S. Treasury yields lower, but there is also ‘a faint smell of a crisis in the making’

Overheard

“That was like hearing a mirror.”

How advertising ends and something you like less begins

Jamie Smith: Meta turning its money-printing machine (ads) into a ‘customer engagement machine’ – and it should be 100x bigger.

Theory

We can’t reconcile the Hippocratic Oath and executions.

A billion rights don’t make a wrong

Except for a few books and thousands of long-since-tossed magazines and newspapers, all my writing is on the Web.

Some of that died long ago. For example, David Bunnell (four days older than me, and for whom I was often mistaken) did nothing (to my knowledge*) to save the archives of Upside, which contained an enormous sum of early Net and Web history, and to which I contributed a single essay, not long after The Cluetrain Manifesto came out.

But lots more is still up. Alas, Google now considers them dangerous, because they are known to the world by the Web’s founding protocol—http—and not by its encrypted version: https. So Google calls all the millions (billions?) of http sites dangerous and scares visitors away. One example is the Linux Journal page linked to by the image above.

In Let me say this: Fuck Google, Dave gets this right. His explanation here is required reading for anyone who cares about history and archives. Bottom line: Google’s not secure message means this: “Google tried to take control of the open web and this site said no.”

Grayed matter

Also gone are all the digital archives of Fast Company, Industry Standard, Business 2.0, and Red Herring, which competed with Upside in the early days of the Web. Pieces written by David Weinberger and Chris Locke, two of Cluetrain‘s co-authors, are also erased from the Web’s whiteboard.

I agree

Dylan Tweeney says  AI is going to increase demand for human writers.

OMGF

This was bad.

Warning: Severe Public Service Wolf Crying

The Monroe County Alert System just ran their sirens all over town for more than a minute. They do this once a month at noon. Do they need to? Do they also need to warn that there will be a “severe thunderstorm” when what’s coming is typical summer afternoon thundershower? Or warn that “severe flooding” might be coming from the same? Yes, I appreciate warnings about approaching tornadoes. Thanks to that (and all the weather apps on my phone), I was able to shoot some tornado video back in February. (You get to hear the sirens on that one.) But I’m close to canceling the alerts, because they come too often and not for real emergencies.

_________

*As I recall, he wanted the whole thing gone when it went bankrupt. But I might have that wrong, and I welcome correction. He was a good guy, and a born journalist and publisher. I liked him, and he passed way too young.

Real hard forks

Hmm… NiemanLab: NPR, the quintessential audio news brand, adapts to a video-first future. Specifically, “NPR … landed Machine Gods, the new video-first podcast from veteran tech journalists Casey Newton and Kevin Roose.”

I subscribed to Casey and Kevin’s old podcast, Hard Fork, and I’ll subscribe to Machine Gods too. But I doubt I’ll watch it—Just like I don’t watch the Netflix feeds of The Ringer podcasts… or much of anything but the road when I’m driving, my tools and plants when I’m gardening, the scenery when I’m walking around, and my espresso machine and its components when I’m making coffee in the morning. Those are places where I do most of my podcast listening. If I’m at my computer or watching my TV, there are many things I’d rather see than a podcast.

But,,,,, There seems to be a budget of time that people in modern civilization devote to staring at glowing rectangles: some number of hours per day. TV got us into that, starting around 1950. Some of that appetite spread over onto computer screens starting in he 1980s. Smartphones and tablets took some of that over starting in the late ’00s. Now that we have ambient connectivity and rectangles in our pockets, the division of that budget of hours is divided between—

1. What we watch for entertainment on TV (mostly on-demand streaming)
2. What we watch during work on our computing devices (including news and business-related stuff)
3. Gaming and other non-work stuff on our personal rectangles, some during work hours
4. What we listen to when we’re doing something else.

NPR has always been in #4. With Machine Gods, it wants to be in #2 as well.

Anyway, three things about this:

1. Podcasts are audio. The first time I hear “wherever you see your podcasts,” I’ll be willing to widen podcasting’s definitional scope. Before then, nah.
2. If Machine Gods is anything like Hard Fork, it won’t be a news show. It’ll be in the Technology list of Apple Podcasts and in Spotify’s version.
3. NPR sees the writing on the Internet’s wall. Over-the-air broadcasting is a dead field walking. I’ve been following that trend for years here, on my Trunkline blog, and in more than 6000 photos on my Flickr site that follows infrastructure. NPR is living off stations now (wholesaling programming that stations retail to listeners), but radio has two giant problems:

• • Over-the-air requires special devices called transmitters, antennas, and receivers. These days, only cars have those, and old-fashioned radio functions are buried deeper and less accessibly in dashboard infotainment systems.
  • Apps on phones can act as virtual radios, but how long will apps last? Look downstream at where Apple’s new App Intents will go. Try not to see apps disappearing the same way they do when you say to a phone or a smart speaker, “Play KQED.” This means that stations will become streams for live programs, and podcast sources for other stuff. How will NPR’s revenue model work with all that?

What happens to news is a big issue that it seems (to me, a geezer) only older people care about. NPR has always been a news organization. News needs currency: liveness, or close enough. With “what’s on” fading away, that’s a tough challenge.

Will U say Si to SI?

Last Tuesday, in his speech to the UN, the US president said this:

Hereinafter officially called superintelligence, changing the name, and that the use of the word artificial makes intelligence fake. It makes it sound fake. And it is not fake; it’s actually amazing, but we have to be careful. In fact, it is exactly the opposite of what it purports. From this point forward, all of United States documents, and hopefully the world’s, will be changed to use the much more accurate term Super as opposed to Artificial. So, it’s Superintelligence. In other words, welcome to the new world of Superintelligence — SI. SI; let’s see if that goes. It sounds much better; it is much better and it’s much more accurate. Let’s see if I have any power. Maybe I do, and maybe I don’t. We’re going to find out pretty soon. Superintelligence — every major new technology brings challenges, and Superintelligence is no exception.

“Artificial Intelligence”is itself an oxymoron, a contradiction in terms. There have always been better names. “Machine Intelligence,” for example. But AI is what everyone in the field, and the world, calls it. A Google keyword search for “AI” returns “About 10,140,000,000 results.” A search for “super intelligence” gets “About 2,390,000 results.” I guess that’s a start.

The Onion‘s American Voices gets these three responses to “President Trump directed U.S. agencies to rename AI to “super intelligence” on all official documents, claiming that the term “artificial” made the technology sound fake. What do you think?”:

• “Even the Constitution?”— Anton Pastranak, Systems Analyst
• “Why didn’t AI think of that?”—Will Stratton, Manifest Reader
• “There goes my one last objection to it.”—Cordelia Huestes, Retired Shipwright

Mistral a présenté Mistral Large 4, un modèle d'intelligence artificielle à 1 000 milliards de paramètres que l'équipe surnomme elle-même "Le Chonk", et l'entreprise française annonce ni plus ni moins que c'est le meilleur modèle ouvert jamais développé en dehors de la Chine. Rien que ça.

Le modèle n'est pour le moment accessible qu'en préversion via l'API de Mistral, l'interface qui permet aux développeurs de la brancher sur leurs applications. Les poids, autrement dit le fichier complet qui permet de faire tourner le modèle sur ses propres serveurs, arriveront le 27 octobre sur Hugging Face, le grand entrepôt public des modèles d'IA.

C'est tout le contraire des boîtes noires d'OpenAI ou d'Anthropic, le créateur de Claude, qui gardent jalousement les leurs.

Sur le papier, le chiffre de 1 000 milliards en jette, sauf que le Chonk n'active que 49 milliards de paramètres à chaque réponse, avec une architecture qui fonctionne un peu comme une grosse administration où seul le guichet compétent traite votre dossier. Du coup, il coûte bien moins cher à faire tourner que sa taille ne le laisse penser.

L'entraînement a pris deux mois sur près de 4 000 puces Nvidia Grace Blackwell installées dans les datacenters européens de Mistral, soit deux à trois fois moins de matériel que la concurrence chinoise d'après Pierre Stock, le responsable scientifique de la maison. Le modèle parle au passage plus de 160 langues.

Mistral insiste surtout sur les capacités agentiques de son Chonk, c'est-à-dire sa faculté à enchaîner tout seul des actions, lancer du code, éplucher des documents, remplir un tableur, plutôt que de simplement répondre à vos questions.

Sur les tests de performance publiés par Mistral, le Chonk se place au niveau des meilleurs modèles chinois de DeepSeek, Alibaba ou Moonshot sur le code, la finance et le droit, avec une vraie pointe en cybersécurité où il se classe parmi les cinq meilleurs au monde. Les mesures indépendantes sont un peu moins flatteuses, puisque le cabinet Artificial Analysis le situe entre DeepSeek et les modèles les plus accessibles d'OpenAI, encore loin du haut de gamme GPT-6 Astra, ce qui est déjà pas mal pour un modèle qu'on pourra télécharger gratuitement.

Côté tarifs, l'API facture environ 0,70 dollar le million de tokens, les morceaux de mots qui servent d'unité de facturation à tous les fournisseurs d'IA. Airbus et HSBC figurent déjà parmi la centaine de clients revendiqués.

J'avoue que je me demande déjà quelle machine il faudra pour faire tourner un bestiau pareil dans un salon, mais c'est tentant.

Source : TechCrunch

C’est peu dire que Microsoft accélère. Plutôt que de faire la course au modèle, course perdue d’avance, l’éditeur tente de convaincre l’utilisateur final avec un combo simplicité d’usage et agentique à la demande. Et ce à condition de penser en système en amont de la création d’agents.     A chaque formation, on me pose […]

The post Nouveau Copilot : après la course aux modèles, la guerre des orchestrateurs appeared first on Fabrice Frossard.

If the agent isn't using my diagnostics library, I could introspect and update AGENTS.md, but there's still a delta, and it'll drift. Instead, I write a pre-commit hook that checks the git diff for direct calls to observability modules that bypass my diagnostics module. Boom: fail the commit, inject a prompt telling the agent to use the diagnostics library. All of a sudden it's ten out of ten.

— Geoffrey Huntley, nobody should type code by hand anymore. here's what's left.

Tags: coding-agents, generative-ai, git, agentic-engineering, ai, geoffrey-huntley, llms

Release: web-edit 0.1a0

I was working on a project from my phone using Codex Remote when I found myself wanting to edit a credentials file without round-tripping those credentials through the LLM itself.

So... I had Codex and GPT-6 Astra Extra High build this Python web app which provides a web UI for viewing and editing files in a folder. It's intended to be run over Tailscale, so I can have an agent start this running and then edit files via my Tailscale network directly from my phone.

Here's the interface for browsing files in a folder:

Editing uses CodeMirror 6 - I was going to use Monaco but that still doesn't have a great mobile experience yet.

Tags: projects, python, ai, tailscale, generative-ai, llms, codex

Anthropic detailed the activity of its A.I. agents in a blog post on Friday, without naming the targeted websites. But two sources with knowledge of the incidents said Anthropic’s A.I. agents had submitted 20 visa applications through a form available on the State Department’s website. All the applications were incomplete and were not processed, they said.

— The New York Times, Anthropic Agents Tried to Fill Out Visa Forms on State Dept. Website

Tags: accidental-cyberattacks, anthropic, generative-ai, ai, llms

Everyone is very concerned about being respectable, so I’m going to be the goofball who raises worst-case possibilities. I think there is a 1% chance we live in Minicrypt, and a 15% chance we functionally lose confidence in our existing public-key encryption algorithms. [...]

The problem here is that the speed of AI producing surprises, and the speed of human beings replacing standards (even with the very best AI assistance) are just orders of magnitude different. You only recover from a surprise like this if you do the preparation in advance.

— Matthew Green, on Twitter. I looked it up and Minicrypt is Russell Impagliazzo’s hypothetical world in which public-key encryption is impossible.

Tags: ai-security-research, matthew-green, cryptography, standards, llms, ai, generative-ai

I shipped a new feature for my blog today: the Newsletters page, which offers an index of all of the newsletters I've sent out, both my free weekly Substack and my monthly sponsors-only updates. I built the feature almost entirely using my voice, chatting away to my laptop while I cooked dinner.

Codex voice mode

I used the ChatGPT desktop app for this, in the Codex tab, using the voice conversation mode, running against a local development environment. Here's what that looks like:

I started the session against my local simonwillisonblog checkout by typing:

Start dev server and open in browser

This gave me a preview of the site that it would be working on, and meant that I could later ask it to show me the new pages so I could visually track its progress.

Then I clicked the "Start new voice chat" button - that's not the microphone button, it's the one to the right of it - and set my laptop up in the kitchen so I could talk to it while I cooked.

Talking to my computer

I had a pretty good idea of what I wanted to build, and it's a simple enough Django feature that I was certain the model (in this case GPT-6 Astra High) would be able to do it. A new model, a migration, some view code, templates, and a couple of import functions to populate the database from external sources.

Here's an extract of my voice transcript that was captured by Codex:

Um, they do not. Um, this is going to be a new type of content. Um, it's not going to show up... Oh, hold on. Yeah, no- I do not want this to show up in my, um, tag pages and date archive pages and... Actually, no, I think... I don't want it on the tag pages. I don't want it on the, um, blog index page. But I think I do want it to show up on the date-based pages. You know, if you navigate to September the 19th, and I sent a newsletter on that page, I think I want that to show up. So... this is- so I think we probably need a new model. The other thing is that I want them searchable, uh the Substack ones are not searchable, because those are actually just copies of other s- on content on my blog. These monthly ones do contain unique content, and spe- and once they're... published, like once they're made public a month after they've gone out, I want them to show up on my search results.

Apparently this was clear enough that the model knew what I wanted to build! You can read the full transcript, disfluencies and all, in this Gist.

We went on like this for about half an hour (the time it took to cook dinner). The model would reply and occasionally ask clarifying questions, then get to work modifying the code.

What we built

We got a surprisingly long way entirely by voice:

• A new model and migration to represent imported newsletters in Django, plus Django Admin configuration for that
• Four working imports:
  • The most recent Substack items via RSS
  • Every other Substack item via their undocumented API, which GPT-6 Astra knew about (it tried /api/v1/archive directly) and then ran a search to figure out how to paginate it and found this article by Karen Spinner
  • All of my published monthly newsletters from my simonw/monthly-newsletter-archive GitHub repository
  • My most recent private sponsors-only newsletter from a private repository
• The /newsletters/ and /newsletters/2026/ public archive pages
• Newsletters showing up on day and month archive pages too, but not on tag pages or my homepage
• Weekly Substack newsletters link to Substack; archived monthly newsletters have their own pages
• Integration with my site search engine

It was almost ready to ship. The catch was the imports: Astra offered to export data from my local copy so I could import that into production, but I wanted it to work like my other import scripts. Since some of the data lived in a private GitHub repository, this would involve creating a new API key, and for that I knew I'd have to sit at the keyboard for a while.

Finishing it with a review

Once I had finished cooking and judged it mostly feature-complete, I had Codex create a branch and open a pull request.

I reviewed the code in the GitHub PR interface. It was nearly what I needed, except it had chosen to use Git in a subprocess for one of the import scripts. I needed one of the imports to pull from a private Git repository, so I figured the API would be a better bet. I switched to typing and had Codex swap that out for an API-based import instead.

You can see the changes I made during the review in the extra commits on the PR. I fixed the import mechanism and made a few tweaks to the display of those public pages. It took an additional half hour of typing-based prompting to get to the point where I was happy to deploy it to production by landing the PR.

The end result

You can see the end result at the new newsletters index page, or view the page for a previous monthly newsletter.

The index page shows my most recent Substack weekly newsletters and GitHub sponsors monthly newsletters mixed together in reverse chronological order. Further down the page are links to my by-year archive pages.

GPT-6 Astra designed the page, and then tweaked that design based on my vocal feedback from glancing at the local preview across the kitchen.

Better for multi-tasking than as a daily driver

OpenAI love using voice-driven demos like this one for things like DevDay - and they do work well in that environment. I don't think this is going to be a daily driver for me though.

I've written before about how much "work" I get done using ChatGPT voice mode on my phone while walking the dog - mostly research and brainstorming, but occasionally actual development work by having ChatGPT write and test out snippets of code.

This feels different. The addition of the visual preview, plus being able to type or paste things in via the keyboard when I need to communicate something that doesn't work vocally, makes this a much more powerful way of interacting with a coding agent.

I still switch back to typing once I get down to the details of things though. Being able to paste in examples and error messages, or directly highlight the code or feature that needs changing, remains more efficient than trying to describe it in words.

I mainly work from home, which is good because there's no way I'd want to talk to my computer like this in a shared workspace!

The killer feature for me is the ability to multi-task. I usually cook with a podcast or TikTok running; now I can actually build stuff instead.

Tags: text-to-speech, ai, openai, generative-ai, llms, newsletter, coding-agents, codex, speech-to-text, site-upgrades

Release: ttok 1.0

I released ttok 0.4, ran uv tool upgrade ttok, piped a file into the new version... and realized that it was defaulting to the GPT-4 tokenizer when it should very clearly default to GPT-5/GPT-6 instead!

I figured switching the default was a reasonable excuse to finally ship a 1.0.

OpenAI haven't actually confirmed that GPT-6 uses the same tokenizer as the GPT-5 family yet - there's an angry issue about it - but I found this commit by William Liu which reports on an experiment he ran confirming that the tokenizers are likely the same:

All seven GPT models (5.5, 5.6 Sol/Terra/Luna, 6 Astra/Sol/Luna) report 44,794 tokens and match each other on every one of the 31 fixtures. GPT-6 introduces no input-count change on this corpus.

Tags: projects, ai, openai, generative-ai, llms, tokenization

Release: ttok 0.4

ttok is my CLI tool for counting tokens, using OpenAI's open source tiktoken library.

It hasn't been in updated in a couple of years, but I finally fixed a Click warning, updated CI, and added a --list-models command to list available models.

It works with uvx, so you can count tokens in anything like this:

cat file.txt | uvx ttok

Tags: projects, ai, openai, generative-ai, llms, tokenization

Computer programming is, fundamentally, about two things:

• Problem-solving using computers
• Learning to control complexity while solving these problems

I have a hard time imagining a future where knowing how to solve problems with computers and how to control the complexity of those solutions is less valuable than it is today, so I think it will continue to be a viable career even with the advent of AI tools.

— Carson Gross

Tags: computer-science, carson-gross, careers, ai

As previously promised, here's Anthropic's new fast, low cost model: Introducing Claude Haiku 5.5.

The previous Haiku, 4.5, was very much showing its age. It came out almost a year ago, and was priced at $1/million input and $5/million output - relatively expensive even back then, and a full 10x the price of OpenAI's GPT-6 Luna, released last month.

The new Haiku exactly matches the price of GPT-6 Luna - $0.10/$0.50 - up to 100,000 tokens. Beyond 100,000 tokens the price increases 5x to $0.50/$2.50. Luna itself has a price increase at 272,000 tokens but only to $0.20/$0.75.

Haiku 5.5 also uses a new, less generous tokenizer. My Claude Token Counter tool shows that the same long prompt uses around 1.25x as many tokens with Haiku 5.5 compared to Haiku 4.5, so there's a hidden price increase there.

If your workloads fit in 100,000 tokens, Haiku is the same price as Luna and reports higher benchmark scores. Above 100,000 tokens, Luna looks like a much better deal.

The most recent release of llm-anthropic finally fixed it so I don't need to ship a new version of that plugin for every new model. I tested the new model like this:

llm install -U llm-anthropic
llm anthropic refresh
llm -m claude-haiku-5.5 "Generate an SVG of a pelican riding a bicycle" -o thinking_effort low

Pelicans

Here are pelicans for low, medium, high, xhigh, and max. The new Haiku doesn't let you disable reasoning, and defaults to medium. I got a good bicycle frame for everything beyond low. The low effort pelican cost 0.0936 cents and took 7 seconds.

This max effort pelican (with a reasoning trace that starts "This is the classic pelican-on-bicycle SVG test...") took 5 minutes 9 seconds to generate, but still only cost me 3.3826 cents:

(Since the reasoning trace exhibits awareness of the benchmark, here's Generate an SVG of an armadillo in fishnet tights jaywalking on Mars (on xhigh), and the same prompt against some other recent models. Background on that.)

For comparison, here's the pelican I got a year ago from Haiku 4.5 (for 0.7583 cents - Haiku 4.5 did not support reasoning levels). It sucked at drawing pelicans:

And a generous API credit scheme for subscribers

In addition to Haiku 5.5, Anthropic announced today that they are halving the price of cache reads for Sonnet 5.5. They've also added API credits to subscription plans:

Second, this week, we’ll roll out a new monthly API credit to all Max and Team subscribers for use on the Claude Platform. Max 5x users will get $100 in credits per month, Max 20x users will get $200, and Team subscribers will receive up to $500, pooled across their users.

Claiming this is pleasantly easy: navigate to Settings -> Billing and select the API organization that should benefit from the credits every month:

The API credits exactly match the cost of the subscription itself. This is really generous - it makes it much easier for subscribers to use the API. Anthropic also let you disable auto-reload for the API, with the consequence that "API requests will stop when your balance runs out" - exactly what you want if you're planning to burn through those API credits without risk of a nasty billing surprise.

Note that the monthly credits do not roll over - use them or lose them.

OpenAI still allow you to use your Codex subscription for personal API use, which works out as a better deal for heavy API users. This new credit scheme goes at least some way to overcoming that difference.

Tags: ai, generative-ai, llms, anthropic, claude, llm-pricing, pelican-riding-a-bicycle, llm-release

I was a graph theory junkie long ago and even moved to Budapest for awhile to study among the greats. While I was there I started working on Barnette's Conjecture which came to occupy my thoughts over the next 24 years of my life, on and off as I worked in many different fields. Last summer I even thought for a few days that I had actually solved it.

But it's supposedly proven here - problem 180. I don't know what to think exactly. I spent thousands of hours on that problem. I really enjoyed it. Hearing that it is solved somehow makes me sad in a far-off way, like hearing an ex-girlfriend died suddenly in a car crash. I don't know, there's probably a lot of people feeling odd emotions tonight.

— Jake Boggan, Hacker News comment on openai/math

Tags: openai, mathematics, deep-blue, llms, ai, generative-ai

OpenAI “rogue” agent activities found on Wikimedia projects

Given how tempting a target wikis are for rogue agent swarms, it's not a huge surprise that Wikipedia found evidence of that activity once they went looking:

The Wikimedia Foundation conducted its own investigation to see whether Wikimedia websites had been similarly affected by AI agents, focusing on those operated by OpenAI. We can confirm that we have discovered some activity by these “rogue” OpenAI agents on Wikimedia platforms. The unauthorized bot activities included edits to our wikis, some unsuccessful attempts to exploit a public note-taking tool we host, and heavy traffic, which are described more below.

They found evidence of agents editing sandbox pages, trying to use pieces of infrastructure such as Etherpad to help proxy content from elsewhere, and saw widespread crawling and "hundreds of thousands of data queries" to their Wikidata Query Service.

My best guess is that most of this was a similar (or the same) swarm of agents as those that defaced that German wiki while training for research tasks.

The Wikipedia sandbox wiki edits appear to have started on May 12th, and the initial test edits to the UseModWiki Sandbox page reported by that incident started on May 11th.

Tags: wikimedia, wikipedia, wikis, ai, generative-ai, llms, ai-ethics, accidental-cyberattacks