Arthur Clune
@clune.org
Geek. Likes bikes, climbing and tech
Work: IT at University of Sheffield
Work: IT at University of Sheffield
Reposted by Arthur Clune
My econ lecturing buddies don't tell any tales like this. Issues are 1) restructuring assessment back to exams to cope with ChatGPT 2) making best use of face to face lectures when, as a legacy of the pandemic, there is an archive of online video lectures.
An issue we're seeing at all levels of university is that many students are simply refusing to do *anything*. They aren't reading the syllabus, aren't following assignment guidelines, aren't engaging with material, ignoring deadlines. And this might seem like old news, but it truly has ramped up.
November 29, 2025 at 11:33 AM
My econ lecturing buddies don't tell any tales like this. Issues are 1) restructuring assessment back to exams to cope with ChatGPT 2) making best use of face to face lectures when, as a legacy of the pandemic, there is an archive of online video lectures.
Some of the replies to this thread get a bit tied up around AI, but I don't think it's the core/only issue. How do you learn agency when you can fill every waking moment with an endless scroll of mindless entertainment that is designed to addict? Add in the pandemic and you get this. 1/
An issue we're seeing at all levels of university is that many students are simply refusing to do *anything*. They aren't reading the syllabus, aren't following assignment guidelines, aren't engaging with material, ignoring deadlines. And this might seem like old news, but it truly has ramped up.
November 29, 2025 at 7:43 AM
Some of the replies to this thread get a bit tied up around AI, but I don't think it's the core/only issue. How do you learn agency when you can fill every waking moment with an endless scroll of mindless entertainment that is designed to addict? Add in the pandemic and you get this. 1/
Sigh. All the stuff I read growing up. Asimov turns out to be a groper, Clarke a pedophile and Heinlein I stopped reading after 'Time Enough for Love'.
A book I am reading made such casual reference to Arthur C. Clarke being a pedophile that it sent me digging a little, and... yikes. 😬
www.patreon.com/posts/yes-ar...
www.patreon.com/posts/yes-ar...
Yes, Arthur C. Clarke was likely a pedophile | Jason Sanford's Genre Grapevine
Get more from Jason Sanford's Genre Grapevine on Patreon
www.patreon.com
November 29, 2025 at 7:36 AM
Sigh. All the stuff I read growing up. Asimov turns out to be a groper, Clarke a pedophile and Heinlein I stopped reading after 'Time Enough for Love'.
After a bit over a month, I've written a new blog post! This time on Anthropic's new suggestions for managing context windows with Claude Code
clune.org/posts/anthro...
clune.org/posts/anthro...
Context Engineering for Claude Code
On Tuesday I did a talk to our internal AI group, along with talks by excellent colleagues, on Context Engineering. The very next day, Anthropic released a blog post on Effective harnesses for long-ru...
clune.org
November 27, 2025 at 3:42 PM
After a bit over a month, I've written a new blog post! This time on Anthropic's new suggestions for managing context windows with Claude Code
clune.org/posts/anthro...
clune.org/posts/anthro...
Ok. We're done. AGI is here. '2025-final-v2' just needs to become '2025-final-v2-fixed'
claude has been clauding on some pdf stuff tonight and i've noticed that it's starting to get kinda funky with the file names as we iterate
first we had '2025-processed'
then '2025-corrected'
then '2025-corrected-fix'
then '2025-final'
we're up to '2025-final-v2'
first we had '2025-processed'
then '2025-corrected'
then '2025-corrected-fix'
then '2025-final'
we're up to '2025-final-v2'
November 27, 2025 at 10:53 AM
Ok. We're done. AGI is here. '2025-final-v2' just needs to become '2025-final-v2-fixed'
This is very impressive - serious mathematicians seeing real gains from LLMs in research.
Via @brendannyhan.bsky.social I haven't linked the original thread because the comments are so wrong ('they can't be using LLMs' etc)
Via @brendannyhan.bsky.social I haven't linked the original thread because the comments are so wrong ('they can't be using LLMs' etc)
Early science acceleration experiments with GPT-5
AI models like GPT-5 are an increasingly valuable tool for scientists, but many remain unaware of the capabilities of frontier AI. We present a collection of short case studies in which GPT-5 produced...
arxiv.org
November 26, 2025 at 4:32 PM
This is very impressive - serious mathematicians seeing real gains from LLMs in research.
Via @brendannyhan.bsky.social I haven't linked the original thread because the comments are so wrong ('they can't be using LLMs' etc)
Via @brendannyhan.bsky.social I haven't linked the original thread because the comments are so wrong ('they can't be using LLMs' etc)
Reposted by Arthur Clune
the math doesn't even pencil out. there are ~10^24 stars in the sky. ~10^11 humans have ever lived.
if everyone who ever lived on earth had a pocket universe in which every star was inhabited by earthlike civilization, that would be ~1 trillionth the number of ppl Grok would kill to save Elon Musk.
if everyone who ever lived on earth had a pocket universe in which every star was inhabited by earthlike civilization, that would be ~1 trillionth the number of ppl Grok would kill to save Elon Musk.
new trolley problem just dropped: Elon vs. 10^58 randomly chosen people. let's see what Grok has to say
November 22, 2025 at 6:34 AM
the math doesn't even pencil out. there are ~10^24 stars in the sky. ~10^11 humans have ever lived.
if everyone who ever lived on earth had a pocket universe in which every star was inhabited by earthlike civilization, that would be ~1 trillionth the number of ppl Grok would kill to save Elon Musk.
if everyone who ever lived on earth had a pocket universe in which every star was inhabited by earthlike civilization, that would be ~1 trillionth the number of ppl Grok would kill to save Elon Musk.
Reposted by Arthur Clune
I've reached the point on an almost-entirely vibecoded project where I need to do some major architectural refinement. Some thoughts...
November 20, 2025 at 12:29 AM
I've reached the point on an almost-entirely vibecoded project where I need to do some major architectural refinement. Some thoughts...
Can anyone recommend me a good RSS reader? I've been using The Old Reader after Google Reader died, which is basically a clone of Google Reader, but it seems abandoned now. I want:
1) Web based
2) Keyboard shortcut heavy. Ideally Google Reader shortcuts as the muscle memory is real
1) Web based
2) Keyboard shortcut heavy. Ideally Google Reader shortcuts as the muscle memory is real
November 20, 2025 at 4:06 PM
Can anyone recommend me a good RSS reader? I've been using The Old Reader after Google Reader died, which is basically a clone of Google Reader, but it seems abandoned now. I want:
1) Web based
2) Keyboard shortcut heavy. Ideally Google Reader shortcuts as the muscle memory is real
1) Web based
2) Keyboard shortcut heavy. Ideally Google Reader shortcuts as the muscle memory is real
Reposted by Arthur Clune
Survey research has been in a downward spiral for decades: LLMs are administering the coup de grâce. As someone who’s spent much of their working life using surveys … Gulp. As I keep saying, many bad things happen when the price of custom plausible bullshit falls through the floor.
new paper by Sean Westwood:
With current technology, it is impossible to tell whether survey respondents are real or bots. Among other things, makes it easy for bad actors to manipulate outcomes. No good news here for the future of online-based survey research
With current technology, it is impossible to tell whether survey respondents are real or bots. Among other things, makes it easy for bad actors to manipulate outcomes. No good news here for the future of online-based survey research
November 18, 2025 at 9:28 PM
Survey research has been in a downward spiral for decades: LLMs are administering the coup de grâce. As someone who’s spent much of their working life using surveys … Gulp. As I keep saying, many bad things happen when the price of custom plausible bullshit falls through the floor.
Reposted by Arthur Clune
new paper by Sean Westwood:
With current technology, it is impossible to tell whether survey respondents are real or bots. Among other things, makes it easy for bad actors to manipulate outcomes. No good news here for the future of online-based survey research
With current technology, it is impossible to tell whether survey respondents are real or bots. Among other things, makes it easy for bad actors to manipulate outcomes. No good news here for the future of online-based survey research
November 18, 2025 at 7:16 PM
new paper by Sean Westwood:
With current technology, it is impossible to tell whether survey respondents are real or bots. Among other things, makes it easy for bad actors to manipulate outcomes. No good news here for the future of online-based survey research
With current technology, it is impossible to tell whether survey respondents are real or bots. Among other things, makes it easy for bad actors to manipulate outcomes. No good news here for the future of online-based survey research
More on Gemini 3 and reading historical documents. With a line to make Gary Marcus hop.
Google does seem to be proving that just scaling LLMs is still working
generativehistory.substack.com/p/the-sugar-...
Google does seem to be proving that just scaling LLMs is still working
generativehistory.substack.com/p/the-sugar-...
November 18, 2025 at 8:15 PM
More on Gemini 3 and reading historical documents. With a line to make Gary Marcus hop.
Google does seem to be proving that just scaling LLMs is still working
generativehistory.substack.com/p/the-sugar-...
Google does seem to be proving that just scaling LLMs is still working
generativehistory.substack.com/p/the-sugar-...
If this analysis from EpochAI is correct then
a) model training costs (financial and environmental) are ~5-10x the final run and
b) inference costs (financial and environmental) are smaller than assumed
I'm making heroic assumptions for a). No-one outside OpenAI can answer properly
a) model training costs (financial and environmental) are ~5-10x the final run and
b) inference costs (financial and environmental) are smaller than assumed
I'm making heroic assumptions for a). No-one outside OpenAI can answer properly
November 18, 2025 at 9:32 AM
If this analysis from EpochAI is correct then
a) model training costs (financial and environmental) are ~5-10x the final run and
b) inference costs (financial and environmental) are smaller than assumed
I'm making heroic assumptions for a). No-one outside OpenAI can answer properly
a) model training costs (financial and environmental) are ~5-10x the final run and
b) inference costs (financial and environmental) are smaller than assumed
I'm making heroic assumptions for a). No-one outside OpenAI can answer properly
I'm on a train to London using AI to reformat and summarise messages from Epstein to Steve Bannon about Macon having a boyfriend. The 2020s are wierd
Prompted by today's @garbageday.email
Prompted by today's @garbageday.email
November 18, 2025 at 9:06 AM
I'm on a train to London using AI to reformat and summarise messages from Epstein to Steve Bannon about Macon having a boyfriend. The 2020s are wierd
Prompted by today's @garbageday.email
Prompted by today's @garbageday.email
Using transcription of historical handwriting as a test for general intelligence. Also a new Gemini model is coming
Has Google Quietly Solved Two of AI’s Oldest Problems?
A mysterious new model currently in testing on Google’s AI Studio is nearly perfect on automated handwriting recognition but it is also showing signs of spontaneous, abstract, symbolic reasoning.
generativehistory.substack.com
November 16, 2025 at 10:05 AM
Using transcription of historical handwriting as a test for general intelligence. Also a new Gemini model is coming
I am very taken with this argument, but I'm not sure anyone is ready for it - teach without LLMs and push heavily for LLMs in practical work but let the companies do most of the heavy lifting
The obvious counter is that it's stepping away, but unis will still have to embrace AI for research
The obvious counter is that it's stepping away, but unis will still have to embrace AI for research
University education as we know it is over
Take-home assignments are dead, "one prompt away" is one prompt too far, and what we should do next
inexactscience.substack.com
November 13, 2025 at 10:08 PM
I am very taken with this argument, but I'm not sure anyone is ready for it - teach without LLMs and push heavily for LLMs in practical work but let the companies do most of the heavy lifting
The obvious counter is that it's stepping away, but unis will still have to embrace AI for research
The obvious counter is that it's stepping away, but unis will still have to embrace AI for research
Age yourself with gaming
November 12, 2025 at 6:48 PM
Age yourself with gaming
I now understand where Elon gets it from. The whole thread is great
Erroll believed he was so good at Roulette that he had created an "almost total solution to what is called randomness" that relied on the understanding that "there are no 'random events' or chance.' All events follow the Fibonacci Sequence like the Mandelbrot Set."
November 11, 2025 at 8:07 AM
I now understand where Elon gets it from. The whole thread is great
Two new posts about MCP and context engineering that seem very linked
1) Anthropic suggesting "Code Execution with MCP" but don't give examples www.anthropic.com/engineering/... but then I also came across
2) replacing playwright with js scripts mariozechner.at/posts/2025-1...
1/
1) Anthropic suggesting "Code Execution with MCP" but don't give examples www.anthropic.com/engineering/... but then I also came across
2) replacing playwright with js scripts mariozechner.at/posts/2025-1...
1/
What if you don't need MCP at all?
Got Bash and some code interpreter? Skip MCP.
mariozechner.at
November 6, 2025 at 10:50 AM
Two new posts about MCP and context engineering that seem very linked
1) Anthropic suggesting "Code Execution with MCP" but don't give examples www.anthropic.com/engineering/... but then I also came across
2) replacing playwright with js scripts mariozechner.at/posts/2025-1...
1/
1) Anthropic suggesting "Code Execution with MCP" but don't give examples www.anthropic.com/engineering/... but then I also came across
2) replacing playwright with js scripts mariozechner.at/posts/2025-1...
1/
Reposted by Arthur Clune
Academics in Assyria in the 7th c BC complain that admin is preventing them from doing research and teaching
November 3, 2025 at 10:04 AM
Academics in Assyria in the 7th c BC complain that admin is preventing them from doing research and teaching
Reposted by Arthur Clune
Will be discussing ChatGPT with students this week - decided to populate my sidebar with a couple of silly queries for them to spot when I share the window. Have to say it knocked this one out the park - particularly point 2.
October 29, 2025 at 3:50 PM
Will be discussing ChatGPT with students this week - decided to populate my sidebar with a couple of silly queries for them to spot when I share the window. Have to say it knocked this one out the park - particularly point 2.
This is even worse than the headline makes it seem
Pretty extraordinary story here
www-bbc-co-uk.cdn.ampproject.org/c/s/www.bbc....
www-bbc-co-uk.cdn.ampproject.org/c/s/www.bbc....
China intimidated UK university to ditch human rights research, documents show - BBC News
Sheffield Hallam University apologises to Professor Laura Murphy for restricting her academic freedom.
www-bbc-co-uk.cdn.ampproject.org
November 3, 2025 at 9:04 AM
This is even worse than the headline makes it seem
Reposted by Arthur Clune
ladies and gentlemen...we got him
October 30, 2025 at 7:10 PM
ladies and gentlemen...we got him
Reposted by Arthur Clune
Genuinely delighted to download a PhD thesis from a university repository where the author has neglected to remove the words "BITCH THIS IS YOUR THESIS" from the filename.
October 30, 2025 at 8:21 AM
Genuinely delighted to download a PhD thesis from a university repository where the author has neglected to remove the words "BITCH THIS IS YOUR THESIS" from the filename.