@matthewmorris
A sometime mathematician and software engineer who should arguably have done something else instead. A convert to Bluesky's block-heavy approach: "But the little children, to save any bother, Let it in at one ear and out at the other."
The Indian plate is part of the Indo-Australian plate, which underlies the Indian Ocean and Bay of Bengal, and is moving north-east at an average of 60 mm/a (0.075 in/Ms).
I can overlook millimetres per year, but what utter anarchy of a unit is inches per megasecond
I've seen a lot of posts recently about whether LLMs can be conscious. I think not, due to their static nature. This paper argues my position very well. To the many who will find it hard going, I recommend asking ChatGPT to summarise its arguments: it does an excellent job.
arxiv.org/pdf/2512.12802
I will unfollow, mute or block rather than scream or yell. But ultimately it’s just very hard for me to take anyone who thinks this seriously.
Many have probably blocked or muted you TBH. I use LLM technology heavily, have a pretty good grasp of how it works, and I don’t see how anyone with that knowledge can claim this technology can yield consciousness. To me it’s like sticking googly eyes on a rock and then conversing with it.
Go have a look at the data and then opine. This is very weak. epi.org.uk/publications...
I read the article. I can’t see anywhere where she actually says “bourgeois support”? It seems like an interpolation by the subeditor. I think Labour policy sucks, but this feels like bait by the Times rather than Mahmood actually telling people like me to eff off from the Labour Party.
Just read the article. Can’t see anywhere where she actually says “bourgeois support”? Seems like an interpolation by the subeditor. I think Labour policy sucks balls this feels like bait by the Times rather than Mahmood actually telling people like me to eff off from the Labour Party.
I’ve worked in finance all my life and am a mathematician-cum-engineer by trade, but this obsession with quantifying the ultimate worth of everything in terms of market value drives me around the twist. Lazy, reductive, harmful.
There is a lot of simplistic bilge out there
Agentic coding is clearly going to eliminate a whole bunch of repetitive dreck, but that’s nowhere near the same thing as wholesale replacement of programmers. For me, it’s pretty much all upside, but that’s going to vary hugely depending on the company & nature of the software being developed.
Submitting to ChatGPT doesn’t put information in the “public domain”. But it does, by default, allow this data to be used for training the models - so that data may resurface later, in whole or in part. That means it is a bad idea to submit private or sensitive information to public LLM services.
It’s a gigantic ball of weirdness. I read it once every 10 years or so and am struck by a different facet of insanity each time.
any chance of a repost for our unique comics anthology currently on kickstarter @chadbourn.bsky.social ? thx
www.kickstarter.com/projects/tri...
This is just that there are 2 definitions of “politics” surely? He’s terrible at governing (on definition) but he’s v good at manoeuvring to get particular positions accepted as frames and achieving his aims as a result (another definition). The latter obvs being vv important, whatever you call it.
The book is a bunch of overwrought nonsense involving lots of people behaving really badly. I really like it! Sounds like the film has taken a vast number of liberties, must admit I’m tempted to check it out
Time to retell my worst first date story:
We're having drinks and a little while in she asks if I know who Andrew Wakefield is?
I launch into a real rant: call him a liar, list off all the errors in his work, his patents and other conflicts of interest, anyway, long story short he's her godfather.
They are probably copying this approach from what people are successfully doing with computer coding (generation followed by iterated automated reviews of various kinds). So I guess some of the painful re-learnings will be on the various fundamental differences between code and prose…
I remember a bunch of schizophrenics found each other in the 90’s when the web was young. They had a forum, literally swapping tips on tinfoil hats. “Mind Control Foundation”? That definitely seemed genuine. Wouldn’t be surprised if there were some disinfo experiments going on too though…
Yes this, there’s a certain kind of “left” arsehole who I remember identifying back in the 80’s: online has enabled them to find and validate each other’s brain worms: like the Flat Earth Society, but for wankers
I think one point in favour of a language is if it has relatively consistent idioms in the LLM’s training set. C++ is an example of a language that doesn’t fit this bill so well: lots of shifts in good practice over the last 20 years.
I totally agree with you on that score. People are acting as if LLMs are about to give us super-intelligent general AI: *not* going to happen. And don’t even get me started on the transhumanists… There’s a whole wing of tech who don’t seem to even like being human, and they frankly give me the ick.
I don’t think it *is* easily transferable, and I think that you are right to be suspicious. Helping with coding plays to the strengths of LLMs, which are basically sophisticated pattern matching engines. They don’t have intentionality or personalities and can’t be creative.
So you spend a lot more time on the irreducibly human parts of the process. Eg for prototyping a new system you can spend far more time (as a proportion of the whole) interacting with your representation of reality (the “domain model”) and adjusting it to make it more useful.
This is all stuff that an experienced worker knows how to do, but it can require a lot of time to acquire the necessary task-specific context. How well this works depends partly on the technology - some computer languages are better suited to LLMs than others - there is a lot of nuance.
LLMs shine at routine-yet-time-consuming work, e.g. first drafts off unfamiliar API documents, root cause analysis of errors from logs in systems with millions of lines of code, refactorings to change database backends. There is a lot of that kind of work in coding!
LLMs can be very effective for coding (with some heavy caveats, you’ve got to know what you’re doing or you’ll eventually drown in a sea of vibe-coded slop) but then (tech) people over-generalise from that. The space of potential coding solutions is very constrained compared to prose.
Agentic coding’s valuable, but - if it’s not a one-shot closed-end project - you still need to review the output and guide the process. Replace “does my job better than I do” with “it’s a multiplier of expertise and judgement - both for good and for bad” and you’re far closer to the truth.
One problem here: the many people who are apparently incapable of understanding just how simple and small the idea-space of coding up solutions to engineering problems is - compared to the bewilderingly, incomprehensibly, scarily, vast range of possibilities in any field of artistic endeavour.