Nick Lothian

@dschwarz26 Yes! And the whole Rastafarian thing is so interesting too

Ethiopian history is really something. Heads of government taking night classes on how Taiwan and South Korean developed, a federalist system of government, an agricultural revolution, constructions of huge dams. Two massive civil wars with humanitarian catastrophes, terrible military corruption, and borderline autocracy. Also only country to actually win a war against European colonists and preserve self-rule. It's like a melange of a bunch of other countries, almost out of place in Africa

My son has been learning SMID in C using Google Highway and wrote a pretty nice and fast multi-threaded Mandelbrot set viewer. Supports AVX2, AVX512, NEON (and multithreading) github.com/a-lothian/Mand… He's finishing CompSci this year and looking for internships/work! Pls RT!

This take is accurate. The benchmarked languages were *designed* to be hard to use (Brainfuck should give you a clue..)! Nice language = good performance.

For example I saw Anjunadeep at Savaya last year. I agree it's a spectacular setting but 6th best club in the world seems a stretch! Great show though

I'm not exactly a big clubber or something but somehow it turns out I've been to a number of the top DJ Mag 100 clubs djmag.com/top100clubs and ummm.. WTF are these rankings?

@EliasBiz Also depends on the use-case. For coding open source models are Sonnet 4.5 level at best, which is useful but a long way from the Opus 4.5 revolution. Other agentic workflows probably hit the same capability gap. If there are hard secure reqs it might make sense.

@nlothian Good point. I’m also reflecting on all these cloud services I pay for to host sites that can also now just be hosted on a home office server. Open weight models and local hosting two related but different trends

The AI narrative is stuck on shorting software. The real shift is underneath: less cloud, more local compute. Why move data around when models can run next to it? Physics > cloud margins. Cost wins.

@kellabyte Still distill.pub

Who makes the slickest looking diagrams in the blogosphere? I need to whip something up and would love some inspiration.

Strong agree with this. The terminal is not the end of the road for coding environments.

@jeremyphoward I'm surprised the story in the Australian was positive TBH!

The combination of skepticism of anything containing the word "AI" plus skepticism of anything containing the word "Australia" is pretty tough... ;)

Proud with @UNSWRNA to have been involved & making the mRNA-LNP for Rosie. There are nuances here that the thread below misses but nevertheless, the intersection of RNA technology, genomic & AI poses an opportunity to change the way do medicine and make access more equitable 1/8

@DynamicWebPaige @omnimagazine @internetarchive In the same issue: - Lets see if it is CFCs causing Ozone Depletion - An article on the brand new MIT Media Lab Stewart Brand

I request that you all please stop what you're doing and read this miracle of a 1987 @omnimagazine interview with Claude Shannon on @internetarchive: - "I am always building totally useless gadgets just because they're fun to make. They have no commercial value but may be amusing." - "I visualize a time when we will be to robots what dogs are to humans, and I'm rooting for the machines." - "I have got a file upstairs of unfinished papers! Ha-ha-ha! But that's true of most of the good scientists I know. Just knowing for ourselves is probably our main motivation." - Omni: "You once created quite a stir by juggling while riding a unicycle through the corridors of Bell Labs!" Shannon: "Yes, I did! Those people are very far-out, but this was something that had never happened in the halls before." - Omni: "Can you imagine a robot president of the United States?" Shannon: "Could be, but I think by then you wouldn't speak of the United States anymore. The world will have a totally different organization."

This is from Shannon's 1948 paper "A Mathematical Theory of Communication" where he builds a small language model by hand.... people.math.harvard.edu/~ctm/home/text…

@DynamicWebPaige @omnimagazine @internetarchive It's like reading Feynman when he was at (the original Thinking Machines) and wishing he could play with what we have now.

Swapping Mexico and Greenland in Mercator Projection. Work by @neilrkaye

There is a wonderful interactive guide presented as "Figure 3"

This is *not* a good explanation of differential privacy. But the sound effects are amusing...

Hey @bcherny why doesn't /review pick up the same things /simplify does? /review is pretty weak atm

@itsandrewgao Common issue, but METR tests are human equivalent hours. So Opus at 15 hours means a task it takes an expert human 15 hours to do.

u could have made Cursor. i literally had my agent vibecode a working Cursor/Windsurf clone (0:40) 99.99% of devs are SEVERELY underutilizing agents and haven't pushed their limits. METR says opus 4.6 works for 18 hrs but ur not even using 5% of that. full prompt + tips below

Levels and Locks as Agent Orchestration primitives nicklothian.com/blog/2026/02/m…