Tim Cook steps down as CEO of Apple. John Ternus taking over
Yes, it was health issues
Ternus is SVP over most hardware products: iPhone, Mac, Airpods, etc. He’s engineer-minded and innovative
techcrunch.com/2026/04/20/t...
Posts by William Neal
Closing the Strait is just triffin rebalancing with extra steps.
*I went down the DGX spark route for my budding local agentic setup. Whilst calls are free, there is a different limit I am optimising for, prefill-heavy work with small decode bandwidth
Multiable passes, esembles as default, speculative exploration, continuous background processing etc... become doable when calls are effectively free*. Whether these patterns produce better outputs is an open question (but it's one where external API economics can make it expensive to investigate)
Self-hosting LLMs unlocks architectural patterns that aren't available when optimizing for per-call costs on external APIs. When calls are metered you either optimize for minimum calls or you watch token costs balloon. Local first inverts this. 1/?
Nice! Still a shame he ain’t on here
Hence part of the appeal of offshore wind is promise of being able to side step local politics
In the UK we call them lifts but in the US they call them elevators, because we’re raised differently.
I'm going to try on my dgx spark at the weekend. I am not going to hold my breath though!
Nice one!
These 'safe haven' articles make me think of Brad's post and I wonder to what extent it applies to other 'safe haven' events other than 2008 www.cfr.org/articles/for...
🫂 It will get better!
😅
I read it the other way round. The 'ending of civilisation' was a cover for the temporary ceasefire
🫂 related to concerns around latest models being held back due to security concerns?
Jokes aside, this is an incredibly important day for freedom and democracy and, most importantly, for basic common sense and human dignity, in the spirit of 1989.
Another podcast worth a listen. A corrective to Trump as a Russian stooge narrative podcasts.apple.com/be/podcast/t...
If you haven't already, you might like to check out github.com/searlsco/pro... and github.com/obra/superpo.... I'm experimenting with them and I am liking what I am seeing so far
An episode of LoM about dollar dominance and international currencies that is worth a listen podcasts.apple.com/be/podcast/t...
Forcing the rest of Europe to confront its own complicity with Moscow’s resource politics has also driven a clearer divide in the nature of the choice being made. This is why, in so many ways, it has always been Europe’s fight. And why winning is so crucial to the project as a whole.
It was my duff patched code. I'll get round to it sometime this week.
Thanks. The thought did cross my mind that the privacy angle might have its own value, particularly in a European context...
bsky.app/profile/sung... also seems significant. My bet is that you can approximate "looped language model" with a clever custom harness of some kind. Much like CoT started as a type of structured prompting and was later incorporated into models, so too with looping...
Of course! Once I am sure the tool calling issue is my duff vLLM config rather than my duff patched code I will make a PR
Thanks. I've decided to go down the route of wrangling open llms and working with/around their limits and learning something new. I'm hoping that I will be able to recreate some magic with interaction between an ensemble of models. Lets see...
brb just googling Jask...
I forked open-strix (h/t @timkellogg.me) and patched it to work with my local model. I am still working through some issues with tool calling with QWEN and vLLM but I can see in the strix logs that it kinda worked so I am going to take the win github.com/tkellogg/ope...
I'm hiding from the cluster fook that is international affairs by serving my own open source model on my own hardware. I went for the big boy that is OSS GPT but I quickly ran into arm64 and CUDA13 issues. I've had more success serving a QWEN 32bn coding model
Thinking out loud: is there a Baumol/differential price changes aspect to rising social atomisation?
In that: the relative price of, say, going to the pub vs spending time online/playing computer games/accessing streamed media has changed quite a lot over the last 2-3 decades?