FeedCity

In anthropology there is a notion of high-context and low-context cultures. A high-context culture is a culture or society that communicates dominantly through the use of contextual elements, such as specific forms of body language, the status of an individual, and the tone...

Spoiler: I have data from the story in the title of this post, it's mostly what I expected it to be, I've just added it to HIBP where I've called it "Data Troll", and I'm going to give everyone a

pyx: a Python-native package registry, now in Beta Since its first release, the single biggest question around the uv Python environment management tool has been around Astral's business model: Astral are a VC-backed company and at some point they need to start making real r...

Screaming in the Cloud: AI’s Security Crisis: Why Your Assistant Might Betray You I recorded this podcast conversation with Corey Quinn a few weeks ago: On this episode of Screaming in the Cloud, Corey Quinn talks with Simon Willison, founder of Datasette and creator of LLM...

Today's links Maga's boss class think they are immune to American carnage: They're in for a surprise. Hey look at this: Delights to delectate. Object permanence: Paradox of choice (screens); Perpetual Lenovo crapware; Trump's Solicitor General: "bribery is legal"; Marvel's 10...

How Does A Blind Model See The Earth?

Fun, creative new micro-eval. Split the world into a sampled collection of latitude longitude points and for each one ask a model:

If this location is over land, say 'Land'. If this location is over water, say 'Water'. Do not say anything else.

Author henry goes a step further: for models that expose logprobs they use the relative probability scores of Land or Water to get a confidence level, for other models they prompt four times at temperature 1 to get a score.

And then.. they plot those probabilities on a chart! Here's Gemini 2.5 Flash (one of the better results):

This reminds me of my pelican riding a bicycle benchmark in that it gives you an instant visual representation that's very easy to compare between different models.

Via @natolambert

Tags: ai, generative-ai, llms, evals

Regardless of their flaws, AI systems continue to impress with their ability to replicate certain human skills. Even if imperfect, such systems were a few years ago science fiction. It was not even clear that we were so near to create machines that could understand the human ...

TL;DR Pebble Time 2 Design reveal Available in 4 colourways Final specifications How to change your pre-order from P2D to PT2… Like what you…

I just got back from the most awesome WHY2025, the tenth installment of the four-yearly Dutch hacker festival series. I’ve been attending these things since 1993, so it is a big deal for me. By being among the hacker community for nearly a week, you start seeing the world differently. I usually spend (too much) time on policy and writing stuff, and this was a much needed reset. My badge for HEU 1993, the first Dutch hacker camp

Stay safe out there folks!

simonw/codespaces-llm GitHub Codespaces provides full development environments in your browser, and is free to use with anyone with a GitHub account. Each environment has a full Linux container and a browser-based UI using VS Code. I found out today that GitHub Codespaces co...

We were recently travelling to faraway lands, doing meet and greets with gov partners, when one of them posed an interesting idea:

What if people from our part of the world could see a link through to our local resource on data breaches provided by the gov?

Initially, I was

Microsoft today released updates to fix more than 100 security flaws in its Windows operating systems and other software. At least 13 of the bugs received Microsoft's most-dire "critical" rating, meaning they could be abused by malware or malcontents to gain remote access to a Windows system with little or no help from users.

Reverse Engineering the Raspberry Pi Zero 2W Jeff Geerling August 12, 2025

Claude Sonnet 4 now supports 1M tokens of context Gemini and OpenAI both have million token models, so it's good to see Anthropic catching up. This is 5x the previous 200,000 context length limit of the various Claude Sonnet models. Anthropic have previously made 1 million t...

It’s an adversarial question for LLMs, but it’s not unfair.

On the tension between self-preservation and self-sabotage

I think the most amusing comment I had during this live stream was one to the effect of expecting me to have all my tech things neat and ordered. As I look around me now, there are Shellys with cables hanging off them all over my desk, the keyboard I&

I think there's been a lot of decisions over time that proved pretty consequential, but we made them very quickly as we have to. [...]

[On pricing] I had this kind of panic attack because we really needed to launch subscriptions because at the time we were taking the product down all the time. [...]

So what I did do is ship a Google Form to Discord with the four questions you're supposed to ask on how to price something.

But we got with the $20. We were debating something slightly higher at the time. I often wonder what would have happened because so many other companies ended up copying the $20 price point, so did we erase a bunch of market cap by pricing it this way?

— Nick Turley, Head of ChatGPT, interviewed by Lenny Rachitsky

Tags: chatgpt, discord, generative-ai, openai, llm-pricing, ai, llms

I shipped LLM 0.27 today, adding support for the new GPT-5 family of models from OpenAI plus a flurry of improvements to the tool calling features introduced in LLM 0.26. Here are the annotated release notes. GPT-5 New models: gpt-5, gpt-5-mini and gpt-5-nano. #1229 I w...