Pinned
After Intelligence - Things That Were Not There Ch. 1 Coordinates compossible.blog/2026/10/07/c...
Peter Howitt
@peterhowitt.bsky.social
AI optimist (Iain M Banks / James Lovelock variety). Full time lawyer, part-time poet and fumbling philosopher - "That is not what I meant, at all.” https://compossible.blog/
3.3K followers7.9K following5.8K posts
Pinned
After Intelligence - Things That Were Not There Ch. 1 Coordinates compossible.blog/2026/10/07/c...
Peter Howitt reposted
"One person who really understands AI will outperform an entire startup team In 5 years, the people who build with AI will run whole companies with a tenth of the headcount" Sir Demis Hassabis
Peter Howitt reposted
The International Criminal Court has been awarded the European Union’s 2026 Sakharov Prize, the bloc’s top human rights honor.
Peter Howitt reposted
Peter Howitt reposted
Without a testable basis for consciousness, it's all assumption. Confident assumptions about who can feel pain, e.g. newborns, animals, black people have been wrong. The question isn't whether AI is conscious. It's that we have no objective test & until we do functional equivalence is the best avail
1/2 Three reasons why AI is not conscious - and perhaps never will be. New in @nature.com with Ned Block @neddo.bsky.social and Giulio Tononi www.nature.com/articles/d41...
Peter Howitt reposted
Peter Howitt reposted
Replying to @timkellogg.me
i personally have been loving DSH, it’s so incredibly extensible that you can replace even the agent loop or sandbox without forking also, i’m starting to get sold on web UI for coding agents bsky.app/profile/timk...
DeepSeek Harness v0.1 the idea is everything is a plugin — models, tools, sessions, skills, sandboxes, loops, … github.com/deepseek-ai/...
Peter Howitt reposted
Remember OpenAI published all those maths ‘proofs’? Turns out (at least) one of them is pretty ground breaking - the proof of the quasi-Riemann hypothesis. It’s being discussed in the maths community but not much wider cut through yet. Here’s a write up www.quantamagazine.org/ai-proved-th...
Peter Howitt reposted
Sorry I just need to keep making your timeline worse. They ruled it justified that police shot and killed a 1 year-old in his mother's lap because of suspected shoplifting of diapers. RIP Kohen Wiley. www.mississippifreepress.org/police-shoot...
Peter Howitt reposted
"Food insecurity is at emergency levels for 200,000 Palestinians, which means immediate humanitarian supplies are needed to prevent deaths from hunger, Hughes said. This is IPC phase four, one step away from famine conditions."
European companies face “devastating” retaliation from Washington if Brussels presses ahead with plans to restrict American rivals such as Elon Musk’s SpaceX from parts of the European market, the US ambassador to the EU has warned.
"'He’s gone full Carney' is the verdict of one top European. That is certainly a risky strategy. But Trump has not left the Ukrainians with many better options."
Peter Howitt reposted
Israeli Firm Exploited Online Ads to Spy on Al Jazeera and Palestinian Sites, Researchers Say 🏴☠️🏴☠️🏴☠️🧵 www.haaretz.com/israel-news/...
Peter Howitt reposted
How Israeli firm’s ‘zero-click’ malware attack poses global security risk A new ‘zero-click’ hack uses ads to target victims without any action needed on their part. aje.news/lgplg7
Peter Howitt reposted
Friends — It's time to defund Elon Musk. https://robertreich.substack.com/p/defund-musk
Peter Howitt reposted
The president of Brazil has won a court case against Meta, after accusing them of restricting his posts and algorithmically favoring his opponent, Flavio Bolsonaro www.nbcnews.com/world/south-...
Anthropic thinks its AI might be conscious. It has profound consequences for your life
Wildlife photographer of the year 2026
Peter Howitt reposted
Peter Howitt reposted
Google steps up in the AI race with v4 Argon: It's a heavy weight workhorse with greater accuracy on technical requirements and lower hallucination rates than the best Anthropic has but still loses on general intelligence scores (53) vs Opus 5.5 (58)
Fascinating - I had just finished including something similar for Book 2 of After Intelligence when I read this: Anthropic's team demonstrated that smaller models (around 9B to 30B parameters) can independently figure out how to set up and run inference software on low-security consumer hardware.
Peter Howitt reposted
You received a gift article from Haaretz 'Even the Lowest-Ranking Target Justified Civilian Casualties': Israeli Intelligence Soldiers Detail Gaza Policy Shift www.haaretz.com/israel-news/...
💥💥💥 Even a low-ranking target justified harm to civilians: Intelligence soldiers describe changes in the rules of engagement during the war. Intelligence soldiers and drone operators revealed in interviews with Haaretz how the IDF’s rules of engagement changed after October 7, ⚠️🏴☠️🧵