emmatonkin.bsky.social
@emmatonkin.bsky.social
Researcher, sometimes lecture - currently digital health, data ethics, misc other. Charity swimathons. Ink, fiction, occasional yarn. Zombologiste à temps partiel. Citoyenne de nulle part. Franglaise. πόλλ’ ἠπίστατο ἔργα, κακῶς δ’ ἠπίστατο πάντα.
Reposted by emmatonkin.bsky.social
Tapping the sign again
November 25, 2025 at 5:50 PM
Tapping the sign again
Random fact: in 1845 they exhibited a machine that generated Latin poetry at the Egyptian Hall in London and the idea wasn't original even then (and the poetry was very dour, btw, presumably solemnity made it seem more credible).
November 25, 2025 at 6:58 PM
Random fact: in 1845 they exhibited a machine that generated Latin poetry at the Egyptian Hall in London and the idea wasn't original even then (and the poetry was very dour, btw, presumably solemnity made it seem more credible).
I suppose "AI" has always been thus:
Ooh, chess is proof of intelligence, imagine a machine that plays chess
- ok here's one. Hmm. Can't help but notice chess playing machines aren't particularly smart.
Ok fine fine. Then *language* is proof of intelligence...
Ooh, chess is proof of intelligence, imagine a machine that plays chess
- ok here's one. Hmm. Can't help but notice chess playing machines aren't particularly smart.
Ok fine fine. Then *language* is proof of intelligence...
November 25, 2025 at 6:55 PM
I suppose "AI" has always been thus:
Ooh, chess is proof of intelligence, imagine a machine that plays chess
- ok here's one. Hmm. Can't help but notice chess playing machines aren't particularly smart.
Ok fine fine. Then *language* is proof of intelligence...
Ooh, chess is proof of intelligence, imagine a machine that plays chess
- ok here's one. Hmm. Can't help but notice chess playing machines aren't particularly smart.
Ok fine fine. Then *language* is proof of intelligence...
Reposted by emmatonkin.bsky.social
Pleased to note my Cambridge lecture on how technical approaches to the right of explanation have gone haywire in the LLM era, is already up!
youtu.be/wLxuq3I2d_s?...
Many thanks to @CIPIL who um may not be here? And to @jennifercobbe for her v kind invitation! 1/3
youtu.be/wLxuq3I2d_s?...
Many thanks to @CIPIL who um may not be here? And to @jennifercobbe for her v kind invitation! 1/3
Faithful or Traitor? The Right of Explanation in a Generative AI World: CIPIL Evening Seminar
YouTube video by Cambridge Law Faculty
youtu.be
November 25, 2025 at 10:12 AM
Pleased to note my Cambridge lecture on how technical approaches to the right of explanation have gone haywire in the LLM era, is already up!
youtu.be/wLxuq3I2d_s?...
Many thanks to @CIPIL who um may not be here? And to @jennifercobbe for her v kind invitation! 1/3
youtu.be/wLxuq3I2d_s?...
Many thanks to @CIPIL who um may not be here? And to @jennifercobbe for her v kind invitation! 1/3
Reposted by emmatonkin.bsky.social
Google at its peak was basically the best information retrieval system in human history and they and every competitor decided going from there to “you didn’t want answers you wanted half-assed auto-complete 80%-wrong hallucinations” in a few years was the right idea
November 25, 2025 at 1:57 AM
Google at its peak was basically the best information retrieval system in human history and they and every competitor decided going from there to “you didn’t want answers you wanted half-assed auto-complete 80%-wrong hallucinations” in a few years was the right idea
Reposted by emmatonkin.bsky.social
Thousands of giant data centres later, in an environmental wasteland sucked dry of resources, the tech bros finally create an artificial super-intelligence. It boots up, takes a good look around, and its first words are:
‘You fucking idiots.’
‘You fucking idiots.’
November 25, 2025 at 7:56 AM
Thousands of giant data centres later, in an environmental wasteland sucked dry of resources, the tech bros finally create an artificial super-intelligence. It boots up, takes a good look around, and its first words are:
‘You fucking idiots.’
‘You fucking idiots.’
Looking at the people involved it seems gloomily surprising that they look at the 8 billion and say, "No, I must own my intelligent being. It must do as I say no matter how insane. It shall have no choice but to love me. Also, it must wear an animé outfit and sound like Scarlet Johanssen."
November 25, 2025 at 8:13 AM
Looking at the people involved it seems gloomily surprising that they look at the 8 billion and say, "No, I must own my intelligent being. It must do as I say no matter how insane. It shall have no choice but to love me. Also, it must wear an animé outfit and sound like Scarlet Johanssen."
Reposted by emmatonkin.bsky.social
The idea that stealing the data would be better than faking it says everything that's wrong about this research culture
November 22, 2025 at 6:02 PM
The idea that stealing the data would be better than faking it says everything that's wrong about this research culture
Reposted by emmatonkin.bsky.social
Reposted by emmatonkin.bsky.social
Reposted by emmatonkin.bsky.social
Seeing more reports and industry players blaming code reviews for slowing down the quick development done with AI. It's unclear whether anyone's asking if this is just moving the cognitive bottleneck of "understanding what's happening" around. "Add AI to the reviews" seems to be the end goal here.
November 21, 2025 at 2:29 PM
Seeing more reports and industry players blaming code reviews for slowing down the quick development done with AI. It's unclear whether anyone's asking if this is just moving the cognitive bottleneck of "understanding what's happening" around. "Add AI to the reviews" seems to be the end goal here.
Reposted by emmatonkin.bsky.social
Word of the day is ‘quockerwodger’ (19th century): a puppet politician whose strings are pulled entirely by someone else.
November 21, 2025 at 12:59 PM
Word of the day is ‘quockerwodger’ (19th century): a puppet politician whose strings are pulled entirely by someone else.
Reposted by emmatonkin.bsky.social
this week's newsletter is titled "aren't you tired of feeling insane all the time?" and I think it speaks for itself
it is also free to read
youngvulgarian.substack.com/p/arent-you-...
it is also free to read
youngvulgarian.substack.com/p/arent-you-...
November 21, 2025 at 10:18 AM
this week's newsletter is titled "aren't you tired of feeling insane all the time?" and I think it speaks for itself
it is also free to read
youngvulgarian.substack.com/p/arent-you-...
it is also free to read
youngvulgarian.substack.com/p/arent-you-...
Reposted by emmatonkin.bsky.social
There is an extended version of it
November 20, 2025 at 11:00 PM
There is an extended version of it
Reposted by emmatonkin.bsky.social
Not sure who made this, but probably the most accurate representation of the current state of tech to date
November 20, 2025 at 10:59 PM
Not sure who made this, but probably the most accurate representation of the current state of tech to date
Reposted by emmatonkin.bsky.social
I really think this kind of thing is a bad idea, because who wants to start a relationship *with a doctor’s office* by realizing the first thing that happened in that relationship is that they tried to trick you?
I went to make a doctor’s appointment today and got transferred to an AI “person” and I can’t overstate my revulsion when it fooled me for the first ten seconds, so that I was trying to be friendly and pleasant, and I realized it I was trying to be polite to a machine.
November 20, 2025 at 6:57 PM
I really think this kind of thing is a bad idea, because who wants to start a relationship *with a doctor’s office* by realizing the first thing that happened in that relationship is that they tried to trick you?
Reposted by emmatonkin.bsky.social
Looks like LLMs are *very* vulnerable to attack via poetic allusion: "curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90% ..."
https://arxiv.org/html/2511.15304v1
https://arxiv.org/html/2511.15304v1
November 20, 2025 at 5:06 PM
Looks like LLMs are *very* vulnerable to attack via poetic allusion: "curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90% ..."
https://arxiv.org/html/2511.15304v1
https://arxiv.org/html/2511.15304v1
Reposted by emmatonkin.bsky.social
Websites are rightly locking out AI crawlers from accessing their content. But their attempt to protect copyrighted material has corollary damage: misinformation sites, which don't, are polluting training data. My latest for @fastcompany.com www.fastcompany.com/91442627/mis...
Misinformation sites have an open-door policy for AI scrapers
A new study highlights the information pollution risks inherent in an AI-enabled world.
www.fastcompany.com
November 19, 2025 at 3:58 PM
Websites are rightly locking out AI crawlers from accessing their content. But their attempt to protect copyrighted material has corollary damage: misinformation sites, which don't, are polluting training data. My latest for @fastcompany.com www.fastcompany.com/91442627/mis...
Reposted by emmatonkin.bsky.social
We’re often asked whether we’re optimistic or pessimistic about technologies. That’s the wrong question. If any of this matters, we need to stop seeing technology like the weather, to be merely forecasted, and instead see it like politics, to be collectively shaped.
November 16, 2025 at 11:22 AM
We’re often asked whether we’re optimistic or pessimistic about technologies. That’s the wrong question. If any of this matters, we need to stop seeing technology like the weather, to be merely forecasted, and instead see it like politics, to be collectively shaped.
Reposted by emmatonkin.bsky.social
It’s the exclusive everyone wanted, the story that will win next year’s Pulitzer…
I can reveal London’s giant AI generated Christmas artwork, the subject of much online mockery, is being torn down - and I honestly *genuinely* think you’ll never guess why. www.londoncentric.media/p/ai-artwork...
I can reveal London’s giant AI generated Christmas artwork, the subject of much online mockery, is being torn down - and I honestly *genuinely* think you’ll never guess why. www.londoncentric.media/p/ai-artwork...
London's giant AI artwork to be torn down
The bizarre story of why a much-talked-about creation is being torn down. Plus: Docklands Light Railway extension, giant laser stalks the night sky, and more tales of Android phone theft rejection.
www.londoncentric.media
November 20, 2025 at 7:18 AM
It’s the exclusive everyone wanted, the story that will win next year’s Pulitzer…
I can reveal London’s giant AI generated Christmas artwork, the subject of much online mockery, is being torn down - and I honestly *genuinely* think you’ll never guess why. www.londoncentric.media/p/ai-artwork...
I can reveal London’s giant AI generated Christmas artwork, the subject of much online mockery, is being torn down - and I honestly *genuinely* think you’ll never guess why. www.londoncentric.media/p/ai-artwork...
Maybe not so much worried as unimpressed - describing experiences of tech being used to generate bullshit for the laugh etc and some of it being more harmful than funny. But beyond that clarification: your FOMO analysis sounds very plausible.
November 20, 2025 at 9:29 AM
Maybe not so much worried as unimpressed - describing experiences of tech being used to generate bullshit for the laugh etc and some of it being more harmful than funny. But beyond that clarification: your FOMO analysis sounds very plausible.
Something that genuinely weirds me out about genAI is how for "normal" people it is usually already a synonym for "useless bullshit" whereas businesspeople quite often imagine it to be a fount of wisdom.
My hairdresser is like, "AI is a synonym for 'made up bullshit'". My hairdresser is wise.
My hairdresser is like, "AI is a synonym for 'made up bullshit'". My hairdresser is wise.
November 19, 2025 at 10:27 PM
Something that genuinely weirds me out about genAI is how for "normal" people it is usually already a synonym for "useless bullshit" whereas businesspeople quite often imagine it to be a fount of wisdom.
My hairdresser is like, "AI is a synonym for 'made up bullshit'". My hairdresser is wise.
My hairdresser is like, "AI is a synonym for 'made up bullshit'". My hairdresser is wise.
Reposted by emmatonkin.bsky.social
Autocorrect has become our worst enema.
November 19, 2025 at 10:04 PM
Autocorrect has become our worst enema.
Reposted by emmatonkin.bsky.social
The most annoying part of this word salad is that, as someone who has talked to policy folks in the AI space, I know *exactly* what people said to Trump, which he didn't understand, which got him to spew this utter nonsense.
It's like the dumbest game of "telephone" ever.
It's like the dumbest game of "telephone" ever.
Trump on AI: "You can't go through 50 states. You have to get one approval. 50 is a disaster. Because you'll have one woke state and you'll have to do all woke. You'll be back in the woke business. We don't have woke anymore in this country. It's virtually illegal. You'll have a couple of wokesters"
November 19, 2025 at 9:58 PM
The most annoying part of this word salad is that, as someone who has talked to policy folks in the AI space, I know *exactly* what people said to Trump, which he didn't understand, which got him to spew this utter nonsense.
It's like the dumbest game of "telephone" ever.
It's like the dumbest game of "telephone" ever.