A observation about safety culture as applied to programming. Not every risk can or should be prevented. Where possible, at reasonable expense and without imposing on people, risks should be mitigated. But what really needs to be done is prevention or redemption of substantive harms.
Michael Witbrock ✓
@witbrockU of Auckland Professor. Founder TransAxon.bio @TranzAxon, nz.com, curiouscat.cc, @AI4Good; BCI, AI, NLP, ML; ex: IBM, CMU. AI that reasons & Improved brains
If you use its voice dictation mode, don't install ChatGPT Version 26.1027.30824 on the Mac. It's had an extreme regression in transcription accuracy, and is currently dangerously unusable. @OpenAIDevs @OpenAI
Why are we using the word "recursive" to describe tail recursion, or, in other words, a loop? Related, what would "real" recursive self improvement loop like? It would have to involve subsidiary improvement processes that serve to enhance the ones above them.
The entire argument, in the limit, for investment in vast data centres by AI companies is that the scale will enable them to deliver AI at lower cost than sovereign or on-prem models. There is a strong case for both, but that case is not cost saving.
If Andy Burnham wanted to signal, clearly, Labour's competence and lack of capture, he'd find a way to abolish the UK's absurd 5 second imposition on every website visitor to satisfy the absurd idea that every person needs to be repetitively interrogated on their preference about cookies. A mild repudiation, and retirement from public influence of the people who thought that they should impose this on the rest of us might also be widely appreciated. Of course, I'm not British, so my opinion on this should hold no particular weight.
AI coding agents can turn a diagnosis into code, tests and permanent architecture in minutes. The “reliability ratchet” makes systems locally defensible but globally less adaptable: additions accumulate, while almost nothing gets removed. In the worst case leads to systems that can never have their intended function. michaelwitbrock.com/2026/08/…
michaelwitbrock.comThe Reliability RatchetAI coding can turn one plausible diagnosis into code, tests, documentation and permanent architecture within minutes. When every visible failure adds machinery and almost nothing removes it, reliab…
Typical operating configuration for an @Apple MacBook power supply. Design and engineering genius!
It's deleted 92000 lines and added 36000. I hope it knows what it's doing.
For the very few who will care, in a relatively rare occurrence these days, Gemini (or whatever is behind "AI Mode" confabulates when it comes to David Hume's typographic practices in 1758. archive.org/details/bim_eigh…
archive.orgPhilosophical essays concerning human understanding. By the author of the Essays moral and politi...Philosophical essays concerning human understanding. By the author of the Essays moral and political. 1748..Digitized from IA40310303-32.Previous issue: bim_eighteenth-century_twenty-five-sermons-p...
Is the Apple Macintosh charging brick the least-well-designed device ever? It blocks adjacent sockets completely. That is, until it falls out of the wall.
New Rules: processes that are compulsory ought to work. @WeWork Merchant Square London
Space, that is.
I need to go to the pace station in a PanAm space plane. @elonmusk and @SpaceX - please do the deal! @HiltonHotels needs get involcrd too. #late2001 cnn.com/travel/pan-am-return…
www.cnn.comCan you bring a legendary airline back to life? Pan Am is about to find out | CNNThe latest attempt to resurrect the iconic Pan Am airline brand might be the most ambitious yet. Hotels, restaurants and airport lounges are all part of the plan.

⟳ Michael Witbrock retweeted
I'm proud we are releasing LAION-fMRI, a densely sampled 7T fMRI dataset of natural images, with very broad stimulus sampling for testing countless hypotheses & deeply exploring brain representations. It is now available at laion-fmri.hebartlab.com/ What does LAION-fMRI offer? 🧵
laion-fmri.hebartlab.comLAION-fMRI - a 7T fMRI dataset of human visionLAION-fMRI (LfMRI / LAION MRI dataset): 5 subjects, 25,052 launch-release natural images, 165 acquired 7T fMRI sessions with single-trial GLMsingle betas, retinotopy, localizers, and diffusion.
No posts match this search.
