Ahas, Breadcrumbs, Coding Epiphanies
The transparency obligations in Article 50 of the EU AI Act started to apply on August 2, 2026. That day, the European Commission account on X posted: A quick fact check confirmed: the claim “The label says it all” is, at best, simplified. The Commission’s own page says that not all AI-generated or manipulated content needs to be labelled. The icons are optional, and using one does not establish c…
While testing new UI responsiveness changes in MyChatty, I ran a prompt that I know will trigger web search and for which I know the answer: “Who is Nils Durner?” Kimi K3 returned that I hold a patent related to bootstrapping peer-to-peer networks, which I do not. The ingredients for this error come from my ORCID record, a public profile for linking researchers to their publications and other prof…
Mainstream media coverage of OpenAI models hacking into Hugging Face read like science fiction this week: Austria’s ORF headlined it “AI breaks out and hacks another company’s platform”, Deutsche Welle said a model “went rogue”, TIME asked how OpenAI “lost control” of a model. And Fortune had the models “autonomously decide” to hack another company. Dr. Heidy Khlaaf, chief AI scientist at the AI N…
Reuters reports that “Amazon cut jobs in its artificial general intelligence group”: "We’re sharpening our focus on the initiatives that matter most for customers, so we can move faster on what counts. That focus means some difficult decisions, including eliminating some roles within parts of our AGI organization." "We’re sharpening our focus on the initiatives that matter most for customers…
Kimi K3, the newest model from Chinese lab Moonshot AI, has been released (blog post). It’s a big model - 2.8T parameters. Artificial Analysis has it at GPT-5.6 Terra/Sol capability level, but it’s also priced that way - making it the most expensive Chinese model: The only reasoning mode currently offered is “max”, with additional reasoning modes planned. Accordingly, Simon Willison’s Pelican test…
If you’re not ready to shell out ~20€ each month for AI access, your options are limited. There are free tiers of ChatGPT or Claude AI, but the models that power these entry levels are never the most powerful ones. What I wanted for myself: - diverse providers, so I can select whatever’s leading currently diverse providers, so I can select whatever’s leading currently - pay-per-use pricing (or fre…
OpenAI’s coding & work tool Codex is not tied to OpenAI as a first-party model provider. In particular, it allows OpenRouter to be used - a hub to currently 480 models offered by independent inference providers. These inference providers offer models not found in the OpenAI like OpenAI’s own open-weights models gpt-oss-20b or gpt-oss-120b, or third party models including Gemini, Claude, Grok or GL…
The house style of frontier LLMs still seems rather obnoxious at times, which I particularly noticed with GPT-5.5 in Codex. Anthropics new Fable 5 does best on the Creative Writing EQ-Benchmark (2229.6 score vs 2028.5 with GPT-5.5), but Ethan Mollick calls it “overwrought language” that cannot be prompted away: Peter Gostev confirms limited steerability for GPT-5.6 too: “Sol feels quite difficult…
GPT-5.6 is slated for general (and international) release on July 9. The model family comprises of three sizes: Sol, Terra and Luna. The model card is here. A “select group of trusted partners and organizations” had preview access, including Ethan Mollick, Derya Unutmaz and Peter Gostev. GPT-5.6 is said to “feel” part of the GPT-5 family while targeted against Anthropic’s Fable 5 as an intermediat…
The first European Digital Identity Wallet hackathon for the German implementation has concluded. My submission is IIRY - an iPhone/iPad app + small backend that lets, for example, a parent challenge a request for money by a child over WhatsApp. It builds on - and extends - two relevant Content Credentials standards: C2PA and CAWG VC+VP. Important boundary: IIRY does not prove that the WhatsApp ac…