Huw Swanborough

Huw Swanborough

@huwroscience.bsky.social

Cognitive neuroscientist at ETH Zurich. Primary interests in the neuroscience of affect processing/perception! Outside of science: Hiking, huts, snow, bluegrass, folk, and all things creative. ------- Everything's a signal, man.... Everything's a signal

524 followers432 following4K posts

I have been saying the exact same. AI models seems to be built with the goal of impressing non-knowledgeable people. Vibe coders are amazed by the fact that something runs and it made so many lines!

This post is unavailable.

I'm somehow also meant to believe this is the tech that users are clamouring for, that needs massive buildout... Is artificially inflating usage something people do for a successful tech with servers under strain?

Litbowl@litbowl.bsky.social

Holy shit, hadn't seen this over here (makes sense cause the ad didn't run here, but wish @bookshop.org would post more here, the place where people love books!). Great, firm response. So dumb and insidious of Meta.

Ok it's been enough time I can voice my least popular movie take to date. Sinners was fine. Like, pacing was off for me and it couldn't commit to either exposition or just running with the concept, but it was fun. It was fine! A solid 6/10. Fun cinema experience, no great substance. Ok film!

I said this on Reddit when I was unaware of it's reception (I'd not seen any hype or reviews, I went spontaneously and hadnt heard of it) and I got angry comments for literally weeks after lol.

Everynow and then I go back to check the state of the art for AI and code. How is anyone getting anything done with it? The best gemini model fails to capture basic instructions. Fails to retain any context. I can't make it use regexp, it hard sticks to "fullfile" in matlab, breaking everything.

View full thread

I swear, they're tuning LLMs to be impressive to vibe coders and not to people who can code. Because I am giving it the information that should be easiest for an LLM. Continue patterns sort of stuff. Giving it structure of what the code should do at each step and asking it to fill the blanks.

Everynow and then I go back to check the state of the art for AI and code. How is anyone getting anything done with it? The best gemini model fails to capture basic instructions. Fails to retain any context. I can't make it use regexp, it hard sticks to "fullfile" in matlab, breaking everything.

It makes design choices that make no sense if you're coding as part of a project. It routinely just overrules instructions. It's output is disorganised if it works, the code is a nightmarishly organised mess And in some cases it generates WHOLE SCRIPTS that are hallucinations

I asked it for a basic as hell "go through my folder structure, save a text file for the folder names and contents" so I can at a glance my file organisation. Dead simple code. And it gave me code for a wholely different task. Like... 0 overlap.

Exactly. And if you haven't motivated the analysis model, why do I care about power? Answer: I don't, at all. I care about your ability to estimate parameters, not what your posthocs might say. And if your model can accurately estimate parameters, how much do I care about power? Also not much.

Suzanne Segerstrom@segerstrom.bsky.social

Statistical power is a property of an analysis, not a study. Thank you for coming to my TED talk.

How are people doubling down on takes that like... This isn't a "Oh I gotta think about the sate of my soul" sort of deep philosophical topic. This drastically fails the sniff test of being good folk. I can't answer or address any topics/questions this brings up but it's clear as day "eww, dont".

This post is unavailable.

One of the worst things about the "Ai Revolution" is what it's showing me about other people, and man... If ever did something to make myself think "does this have analogues to something morally reprehensible" I'd have at least *some* reservation and modesty about how I discuss it.

"oh it's like being a slaver... Oh I dont think art shoudl be challenging. Oh I don't beleive creatives should be paid. Why should I care about how this impacts other people" are all things I've seen people just declare with 0 introspection. Absolutely no pause to have a think.

Absolutely love people who talk to you like they're expecting confrontation. Dude tells me I'm in the street where bikes aren't allowed. There's no signs so I say "oh sorry, thank you for letting me know" And he's now angrier than before for depriving him of the argument or something? Lads.

"but you should already know" Switzerland has a weird cultural thing where things are/aren't allowed in certain places but also a vehement rejection of signs that indicate this.

I love the idea of fMRI analyses being containerised or part of an all-in-one pipeline that does it all automatically. However, I've never had an fMRI analysis run that didn't go like. "everyone has exactly the same data format, dimensions, and size... But subject 19, session 2, run 1 broke it"

Scientists are absolutely awful at writing manuals and tutorials lol. This manual for an MRI denoising toolbox tells you what it can do and all that jazz, but nothing at all about HOW TO WRITE THE FUNCTION SO IT DOES IT!

I will have limited success debugging your code for an error that is 10 layers deep. I gave it all the right data as far as the manual says, but something missing and that's between the toolbox and god!

Huw Swanborough reposted

KIM KITSURAGI - Well... except that one time when he stopped to sing karaoke. Which, by the way -- I have to disagree with you, Mr. Vicquemare -- was a valiant effort. He really sang his heart out.

Everything I see this i can't decide if the person doesn't engage with any fiction at all, or engaged with the wealth of fiction saying why this is bad, and felt so personally attacked that they went all in on the idea.

Bugs 🔜 MC Atlanta@bugsmaytrix.bsky.social

maybe i'm old fashion but the coolest thing about video games is being taken to new worlds that i hadn't thought up and experiencing something exciting that the artists behind the game want to show me

This should just be crime. Like, in what world can one accidentally under deliver by 20billion? The promise of that revenue was to hype investors to invest more. It was fraud.

Financial Times@financialtimes.com

FT Exclusive: The AI group's annualised revenue is about $20bn less than the company has previously signalled, according to financial documents shared with investors, a massive gap likely to damp optimism about the growth of AI demand. ft.trib.al/IC1ECgv

Genuinely, is there any hope of an independent web at all in the age of genAI? Every major company now has the resources to DDOS any small competitor and defend it under the guise of "model training" or just an oopsie. And why would they choose to not do that?

Look at the paramount-warner merger, between them and Disney, they can literally stop any small creative website gaining a foothold with ease. Not even getting into the possibility of suppressing their presence in a search engine/LLM indexing...just 'legal' ddosing.

Reminiscing on my funniest block. Mark Lamarr: My favourite country singer is X, his music influenced the genre so much. Me: He is excellent, also a huge fan of Y, I think he might have had more influence in terms of music. ML: I mean in terms of voice. Me: I know! Agreed. ML: wasting my time.

Me playing Ace Combat 8 - Boy I sure am glad I fixed that Tonitrus Spear problem, that could really have come back to bite us in the ass. Every 3 missions following: 🫨

"take AI consciousness seriously" No. It's not a serious question unless we go "look, imagine we invent an entirely new model totally different to anything we have now". Which you could also have asked in 1986, so... No, I won't treat it seriously. It doesn't deserve it on any merits.

So uh, wiring lights in the new flat and I have this hydra of wires. Anyone know what this might mean in Switzerland? These don't match the colours of any other wires in my flat lol.

Corrupting the youth of science by showing them how to simulate studies with assumed effects, allowing them to understand power in a realistic way, and being responsible statisticians and experiment designers. Mwuhahahaha