Kit Thwaite
@k6e.me
Head of Data Science and AI @ PRH | Python, Rust, Typescript | NLP, IR
instantly recognisable, burned into my brain at this point
Including (in the '80 volume) some cool diagrams assembled out of typewriter art and hand-drawn arrows cdn.aaai.org/AAAI/1980/AA...
November 12, 2025 at 10:01 AM
instantly recognisable, burned into my brain at this point
most science journalism makes me feel like i’m losing my fucking mind
The degree to which every single major journalist seems to just hate math is palpable in almost everything
Haha, this from the New Yorker is getting passed around the math dork community. I did a comic about this kind of thought a few years ago: www.smbc-comics.com/comic/commut...
November 8, 2025 at 12:50 PM
most science journalism makes me feel like i’m losing my fucking mind
Reposted by Kit Thwaite
Can LLMs accurately aggregate information over long, information-dense texts? Not yet…
We introduce Oolong, a dataset of simple-to-verify information aggregation questions over long inputs. No model achieves >50% accuracy at 128K on Oolong!
We introduce Oolong, a dataset of simple-to-verify information aggregation questions over long inputs. No model achieves >50% accuracy at 128K on Oolong!
November 7, 2025 at 5:07 PM
Can LLMs accurately aggregate information over long, information-dense texts? Not yet…
We introduce Oolong, a dataset of simple-to-verify information aggregation questions over long inputs. No model achieves >50% accuracy at 128K on Oolong!
We introduce Oolong, a dataset of simple-to-verify information aggregation questions over long inputs. No model achieves >50% accuracy at 128K on Oolong!
this, quoting Jeffries analyst Surinder Thind, hits the nail on the head for me. you’re not going to make end-user computing happen without integrations, rich metadata, governance… to say nothing of observability! on.ft.com/491pzLj
November 8, 2025 at 12:36 PM
this, quoting Jeffries analyst Surinder Thind, hits the nail on the head for me. you’re not going to make end-user computing happen without integrations, rich metadata, governance… to say nothing of observability! on.ft.com/491pzLj
piketty on soi-disant “AI” as product/ideological project in the FT, very lucid on.ft.com/4nFLyLk
November 5, 2025 at 11:21 AM
piketty on soi-disant “AI” as product/ideological project in the FT, very lucid on.ft.com/4nFLyLk
integrity
TLDR; The PSF has made the decision to put our community and our shared diversity, equity, and inclusion values ahead of seeking $1.5M in new revenue. Please read and share. pyfound.blogspot.com/2025/10/NSF-...
🧵
🧵
The official home of the Python Programming Language
www.python.org
October 27, 2025 at 7:27 PM
integrity
Reposted by Kit Thwaite
Bring back small phones! This is my iphone 13 mini. It is a perfectly-sized phone. But it's going to die - in a few years, Apple will stop supporting it w/updates. Yet there are now NO new smartphones to buy on the market that fit the human hand. www.ft.com/content/6332...
October 21, 2025 at 12:03 PM
Bring back small phones! This is my iphone 13 mini. It is a perfectly-sized phone. But it's going to die - in a few years, Apple will stop supporting it w/updates. Yet there are now NO new smartphones to buy on the market that fit the human hand. www.ft.com/content/6332...
environmental storytelling
October 4, 2025 at 3:57 PM
environmental storytelling
Reposted by Kit Thwaite
On the morning of Keir Starmer's conference speech here's a new post on an odd psychopathology in British politics - our main parties don't like the people who vote for them - the dreaded Professional Managerial Class. And so they are acting out like a divorced dad seeking cooler voters. 1/n
British Politics' Midlife Crisis
Why British Parties Can't Make Peace with Their Actual Voters
benansell.substack.com
September 30, 2025 at 6:40 AM
On the morning of Keir Starmer's conference speech here's a new post on an odd psychopathology in British politics - our main parties don't like the people who vote for them - the dreaded Professional Managerial Class. And so they are acting out like a divorced dad seeking cooler voters. 1/n
Reposted by Kit Thwaite
I've said it many times but I will say it again: the drive to AI is a drive to constant, regular interactivity. There is enjoyment in spending an hour being interacted with -- that is the actual primary driver of so much of this behavior. The product of the interaction is secondary at best.
imagine relying on AI to write your bio and then having the nerve to say you're witty and full of personality
September 21, 2025 at 2:42 AM
I've said it many times but I will say it again: the drive to AI is a drive to constant, regular interactivity. There is enjoyment in spending an hour being interacted with -- that is the actual primary driver of so much of this behavior. The product of the interaction is secondary at best.
Reposted by Kit Thwaite
I am (self-evidently) very interested in language. And also in technology. From that perspective, the rise of LLMs has created many new things for me to think about.
One is: why is it that certain very small clusters of words are *clearly written by an LLM*? What is the quality of that writing?
One is: why is it that certain very small clusters of words are *clearly written by an LLM*? What is the quality of that writing?
September 16, 2025 at 7:32 AM
I am (self-evidently) very interested in language. And also in technology. From that perspective, the rise of LLMs has created many new things for me to think about.
One is: why is it that certain very small clusters of words are *clearly written by an LLM*? What is the quality of that writing?
One is: why is it that certain very small clusters of words are *clearly written by an LLM*? What is the quality of that writing?
Reposted by Kit Thwaite
this is ridiculous
September 11, 2025 at 9:11 AM
this is ridiculous
Reposted by Kit Thwaite
wearing an ai that mainly insults and fights with you is actually so good for all the wrong reasons. the future is an idiot plot
www.wired.com/story/i-hate...
www.wired.com/story/i-hate...
September 8, 2025 at 5:05 PM
wearing an ai that mainly insults and fights with you is actually so good for all the wrong reasons. the future is an idiot plot
www.wired.com/story/i-hate...
www.wired.com/story/i-hate...
Reposted by Kit Thwaite
Since 2023, translators and interpreters have seen work dry up, rates plummet and their jobs reduced to editing AI-generated output. Some are leaving the field, others are considering bankruptcy. All despite any major upgrade in translation quality.
This is how AI is killing translation work:
This is how AI is killing translation work:
AI Killed My Job: Translators
Few industries have been hit by AI as hard as translation. Rates are plummeting. Work is drying up. Translators are considering abandoning the field, or bankruptcy. These are their stories.
www.bloodinthemachine.com
August 21, 2025 at 6:24 PM
Since 2023, translators and interpreters have seen work dry up, rates plummet and their jobs reduced to editing AI-generated output. Some are leaving the field, others are considering bankruptcy. All despite any major upgrade in translation quality.
This is how AI is killing translation work:
This is how AI is killing translation work:
Reposted by Kit Thwaite
This is also a good example of why we are looking through the telescope the wrong way round regarding LLM’s. They are a *symptom* of a much bigger discursive shift to transactional writing, a shift that’s been under way for decades—they not the motivating cause.
There is a Democratic gubernatorial candidate in a race they truly ought to win this autumn who has chosen to use this platform to post the most middle of the road milquetoast pablum one could imagine, presumably chasing the unicorn known as the swing voter. We still haven’t learned.
August 9, 2025 at 10:06 PM
This is also a good example of why we are looking through the telescope the wrong way round regarding LLM’s. They are a *symptom* of a much bigger discursive shift to transactional writing, a shift that’s been under way for decades—they not the motivating cause.
thou shalt not suffer a machine to say “ugh”
August 8, 2025 at 5:22 PM
thou shalt not suffer a machine to say “ugh”
Reposted by Kit Thwaite
LLM training is a perfect fit for an Elon company. Where others might hesitate to benchmark hack, that opportunity is like catnip
July 10, 2025 at 8:00 AM
LLM training is a perfect fit for an Elon company. Where others might hesitate to benchmark hack, that opportunity is like catnip
Reposted by Kit Thwaite
The loud drumbeat that companies should "embrace AI now or be left behind" has more in common with high-pressure sales techniques than biz strategy. And that's fine - tech firms have something to sell. But the rest of us needn’t confuse it with anything more profound. www.ft.com/content/4688...
July 1, 2025 at 11:15 AM
The loud drumbeat that companies should "embrace AI now or be left behind" has more in common with high-pressure sales techniques than biz strategy. And that's fine - tech firms have something to sell. But the rest of us needn’t confuse it with anything more profound. www.ft.com/content/4688...
Unfortunately paywalled, but the bestseller in question is "a popular colouring book or a 'dad's book of jokes' featuring AI-generated puns[...] by contrast, he argued that AI is less likely to generate a bestselling work of narrative fiction"
www.thebookseller.com/news/ai-like...
www.thebookseller.com/news/ai-like...
AI 'likely' to produce bestseller by 2030, Nielsen's Philip Stone tells Marketing & Publicity Conference
AI will ’likely’ produce a bestseller by 2030, according to Philip Stone, head of publisher account management at NielsenIQ BookData.
www.thebookseller.com
July 1, 2025 at 9:43 AM
Unfortunately paywalled, but the bestseller in question is "a popular colouring book or a 'dad's book of jokes' featuring AI-generated puns[...] by contrast, he argued that AI is less likely to generate a bestselling work of narrative fiction"
www.thebookseller.com/news/ai-like...
www.thebookseller.com/news/ai-like...
this is the lock-in the effective altruists were whining about
“rewrite the entire corpus of human knowledge”
June 21, 2025 at 11:35 AM
this is the lock-in the effective altruists were whining about
Reposted by Kit Thwaite
New paper: "Large Language Models and Emergence: A Complex Systems Perspective" (D. Krakauer, J. Krakauer, M. Mitchell).
We look at claims of "emergent capabilities" & "emergent intelligence" in LLMs from the perspective of what emergence means in complexity science.
arxiv.org/pdf/2506.11135
We look at claims of "emergent capabilities" & "emergent intelligence" in LLMs from the perspective of what emergence means in complexity science.
arxiv.org/pdf/2506.11135
arxiv.org
June 16, 2025 at 1:15 PM
New paper: "Large Language Models and Emergence: A Complex Systems Perspective" (D. Krakauer, J. Krakauer, M. Mitchell).
We look at claims of "emergent capabilities" & "emergent intelligence" in LLMs from the perspective of what emergence means in complexity science.
arxiv.org/pdf/2506.11135
We look at claims of "emergent capabilities" & "emergent intelligence" in LLMs from the perspective of what emergence means in complexity science.
arxiv.org/pdf/2506.11135
Reposted by Kit Thwaite
No we need more philosophical inputs on reasoning design, not offloading that impact and ethics section you don’t want to write.
June 14, 2025 at 1:41 PM
No we need more philosophical inputs on reasoning design, not offloading that impact and ethics section you don’t want to write.
“human writers follow an
iterative process of thinking, outlining, writing, and refining to ensure coherence and quality” — arxiv.org/abs/2506.04180
don’t know where to begin with this
iterative process of thinking, outlining, writing, and refining to ensure coherence and quality” — arxiv.org/abs/2506.04180
don’t know where to begin with this
June 5, 2025 at 8:50 AM
“human writers follow an
iterative process of thinking, outlining, writing, and refining to ensure coherence and quality” — arxiv.org/abs/2506.04180
don’t know where to begin with this
iterative process of thinking, outlining, writing, and refining to ensure coherence and quality” — arxiv.org/abs/2506.04180
don’t know where to begin with this
Reposted by Kit Thwaite
Also pretty important to distinguish between “workplace DEI” meaning “some bullshit training they make us do every year” and “workplace DEI” meaning “hiring women and people of colour and not firing people for being gay”
i do think that most workplace DEI is useless, but that doesn’t matter because when Republicans complain about “DEI” they’re not talking about that, they’re talking about the visibility of minorities in public life
"DEI is pointless rather than menacing" seems like a reasonable take to me, but if that's the case it's still bad.
May 27, 2025 at 8:12 PM
Also pretty important to distinguish between “workplace DEI” meaning “some bullshit training they make us do every year” and “workplace DEI” meaning “hiring women and people of colour and not firing people for being gay”
"99.45% uptime" is a funny way of saying that you had issues one out of every three days in the past 90
May 27, 2025 at 11:05 PM
"99.45% uptime" is a funny way of saying that you had issues one out of every three days in the past 90