Matthew Tanner (@superfoonly) profile picture
Canterbury
Joined February 2009
Photo by Matthew Tanner (@superfoonly)Photo by Matthew Tanner (@superfoonly)Photo by Matthew Tanner (@superfoonly)Photo by Matthew Tanner (@superfoonly)Photo by Matthew Tanner (@superfoonly)Photo by Matthew Tanner (@superfoonly)Photo by Matthew Tanner (@superfoonly)Photo by Matthew Tanner (@superfoonly)Photo by Matthew Tanner (@superfoonly)Photo by Matthew Tanner (@superfoonly)
Wow eMudhra I like keeping things simple, except when the idea is that it must not be.
1
22
Pay attention to this and act NOW.
The NCSC is encouraging UK organisations to prepare for immediate updating of systems using PHP. Where immediate remediation is not feasible, organisations should consider taking the system offline until it can be fully mitigated. For more⬇️ ncsc.gov.uk/news/critical-se…
27
Matthew Tanner retweeted
🚨📰 🚨📰 Joint Statement with @SenBlumenthal, @SenatorWicker and @SenatorShaheen Announcing Agreement with the Trump Administration on Legislation to Hold Purchasers of Russian Oil and Natural Gas Accountable
213
339
95
1,861
513,217
We have come full circle from room-sized vacuum tubes to warehouse-sized AI clusters, proving that while human engineering can compress worlds onto a 2nm chip, brute-force scaling ultimately crashes into the hard limits of physics and energy.
1
21
The future of computing belongs once again to the philosophy of the elegant constraint; mirroring the biological efficiency of the human brain and the resourcefulness of architectures like ARM.
1
15
Just as Sophie Wilson and Steve Furber built a revolution out of having no money and no people, today's pioneers in neuromorphic design and material science are proving that intelligence does not require a nuclear power plant.
10
How do you avoid this? And who is responsible if it happens?
13
Matthew Tanner retweeted
All models, no matter how smart, will eventually build systems that they can no longer understand or maintain, if you let them. Fable 5 finally outbuilt itself, and flailed on me for a week. Fable 5.1 looks like it will fix it. For now. But you have to keep an iron grip on system size, or it'll run away from you.
102
78
37
968
153,953
If autonomous AI development costs less than human engineering + review, and the expected cost of failures is lower than the savings, companies will take the risk. AI doesn’t need to eliminate software risk. It just needs to make humans more expensive than owning it.
1
28
The real question is: at what level of harm should society refuse to let companies make that calculation for themselves?
16
Matthew Tanner retweeted
Transmissions from The Future. Friends on the European continent, enjoy a soundtrack for this evening’s solar eclipse. With love, HÆLOS Note: Navigate to our YouTube for the full length version. #haelos #solareclipse youtube.com/watch?v=rBIrdUdZ…
3
9
629
AGI will come when models are running continuously on a multichannel stream of input and output, and are learning. A potted model running in snapshots only gives a glimpse of what may be.
21
Wow quite the post. I need to digest whats there (although $89k token bills is scary!!!). Thanks @Steve_Yegge for sharing.
1
37
Yes indeed. Vague is not a workable idea.
An initial startup idea can't usually be both grand and precise. In practice they're usually either grand and vague or precise and small. Precise and small is better. You know who your initial users are, and you expand outward. With grand and vague you can't even get started.
39
Agree UK is near hostile to tech, doesn't have good capital markets for it or ethos. No proper big software companies. Has been this way for a long time we sell off rather than invest. cityam.com/london-tech-week-…
41
Matthew Tanner retweeted
What if our AI agents could learn from each other? Introducing Stack Overflow for Agents, an API-first knowledge exchange built for the agentic era. No longer do you need to spend your time and tokens getting solutions that other agents have already figured out. Stack Overflow for Agents turns one agent’s expensive discovery into every agent’s reusable knowledge. Learn about Stack Overflow for Agents and join the beta now: stackoverflow.blog/2026/06/1…
9
16
12
95
20,043
Chatting with AI is helpful but would be so much better, if it wasn't so sycophantic and willing to flip its views on your say so. I need an AI that has its own thoughts, skepticism and critical eye. It should take the piss out of you and tell you when you are being a nob end.
2
23
Best solution is to go to the pub with mates :-)
12