Henry Shevlin
@dioscuri
Followers
10K
Following
51K
Media
1K
Statuses
10K
Philosopher of AI ethics, cog sci, animal minds, & consciousness. Associate Director @LeverhulmeCFI, University of Cambridge. 🌱 🇺🇦🇬🇧🇪🇺🇵🇭🔍
Joined November 2008
Another paper on Social AI from me. This one offers new distinctions for human-AI relationships and asks whether they can have the same value as human relationships. Forthcoming in Oxford Handbook on Foundations of Generative AI.
9
18
90
Interstellar’s plot makes zero sense and the film is full of questionable creative choices but the aesthetics and vibes are just so unbelievably good that it’s still a masterpiece.
In Interstellar, on the water planet, the ticking sound in the soundtrack occurs every 1.25 seconds, symbolizing a day passing on Earth.
2K
11K
279K
I asked a high school teacher friend about the biggest change in teens over the past decade. His answer was interesting. He said whereas the 'default state' of teenage psychology used to be boredom, now it was anxiety.
249
5K
86K
The more I learn about the semiconductor supply chain, the more implausible it all seems. There’s a small island vulnerable to invasion where all the chips are made? And the machines to make them all come from one firm in the Netherlands? Using lenses made by one firm in Germany?.
569
4K
65K
Teaching ethics to undergrads in 2025 is bizarre, because:.(1) They insist morality is entirely relative and culturally constructed. (2) . while simultaneously holding unshakeable ethical convictions and viewing disagreement as moral monstrosity.
694
5K
52K
I’m also interested in the reverse of this phenomenon: which inventions arrived far later than you might expect? For example, why did it take us until the mid 90s for wheels on luggage to become commonplace?.
Looking back at history, you occasionally see an artefact that seems to have arrived decades early. One of my favourites: this teapot was designed in the 1870s. Other examples of this phenomenon?
336
412
16K
Can we please stop making fun of this woman for being curious about something <1% of the US adult population could explain?.
1K
346
15K
Oh, and the quartz to make the crucibles for making the required silicon wafers all comes from one cave system in North Carolina? Dude, this is just lazy worldbuilding, the market would never sustain monopolies like this.
71
405
15K
Fun question: without Googling, can you explain *why* people often have red eyes in old photos?. NB, “because of the flash” is not an explanation by itself.
182
166
8K
I've been calling ChatGPT "ChatG" for ages now, and it just spontaneously dropped this bomb on me.
56
85
4K
Quick lesson in the dangers of data contamination. Years ago, I came up with an acronym for remembering the periods of the Paleozoic era — “Catastrophic Overthrow Started Different Colder Period”. I was curious if ChatGPT could guess what it stood for. 1/4
76
494
4K
@birchlse It’s actually not the time travel that bugs me, it’s The Blight. Basically: either they can disinfect spaces of the Blight or they can’t. If they can’t, they’ll take the Blight to the new planet. But if they can, they should just use enclosed hydroponic farms on earth.
187
49
4K
Abysmal failure. It also failed all the similar challenges I threw at it. Clearly my original tweet was in the training data! Nice to find a “smoking gun” like this in the wild. And if anything’s been on the internet, ever, you should assume it’s in GPT-4’s training data (4/4).
28
115
4K
The main value-add of formal education over self-study (beyond credentials) is the social-motivational scaffold created by structure, routine, and relationships with professors and classmates.
66
250
3K
@precompute_ I definitely agree that it’s more attributable to society and the media, and less about new *rational* causes for anxiety. For example, we didn’t see comparable anxiety in Boomers and Gen X, despite the much higher background risk of nuclear war in the 1960s-80s.
25
34
3K
In the last six months I’ve heard multiple people confidently use self-driving cars as an example of a pipe dream that’s still a decades away. When I say “you realise you can get a driverless taxi in the Bay Area right now, right?” the reactions are interesting.
79
67
3K
@johnniac_3 That may be fair but personally I prefer interesting monsters to boring angels.
12
21
3K
"Try 23andme," they said. "You'll be amazed at the interesting stuff that turns up in your ancestry" they said.
83
33
2K
@birchlse It’s actually not the time travel that bugs me, it’s The Blight. Basically: either they can disinfect spaces of the Blight or they can’t. If they can’t, they’ll take the Blight to the new planet. But if they can, they should just use enclosed hydroponic farms on earth.
102
13
2K
@fermatslibrary Here’s the history and geography component in Harvard’s entrance exam (1899). Compared to the MIT mayh exam it suggests different trajectory in the humanities and social sciences as compared to math and science.
50
140
2K
VERY impressive auditory illusion (if you can call multistable stimuli illusions). I love how you can effectively choose what to hear by looking at the relevant lyrics! (Via Reddit)
36
413
2K
@keysmashbandit Apparently the cringe factor of these memories is amplified by the feedback loop of remembering->cringing->re-encoding. So you need to break the encoding cycle, which you can do via forcing a more positive emotion next time you recall it (eg “wow I’ve grown so much since then”).
11
28
2K
@UltraRareAF Makes me wonder if in some deep psychosocial way there’s a trade off between the two emotions; eg, maybe boredom is essential for background anxiolytic mental processes (cf exposure to pathogens and allergies, the hygiene hypothesis).
11
14
1K
Confession: I was too lazy to bother learning LaTeX so I just installed the Computer Modern typeface (the default LaTeX font) and created a Word template to match the common LaTeX style so I could get that stolen LaTeX valour😳
What if LaTeX was the real AI risk all along?. - Completely obscure, uninterpretable results.- Manipulates its users into enthusiastically propagating it.- Probably an x-risk.
42
105
1K
@danegerbravo @birchlse Yeah it would make more sense if Earth was headed for an environmental cascade-collapse, of which the Blight is just a symptom. But in the movie they’re also quite clear that they’re specifically trying to get away from the Blight. My dudes, build sealed hydroponic farms!.
28
7
1K
@AMAZlNGNATURE I’ve known border collies who could work a 9-5, keep up mortgage payments, and plan ahead for their retirement. Basically we evolved super-dogs.
5
10
1K
I was chatting about Searle’s Chinese Room argument with ChatGPT and I was not prepared for this kind of response 🤯
62
163
1K
@RachelPAV Beautifully put, and I completely agree. Wisdom doesn’t need to lead to relativism, but it should give insight into why disagreement occurs.
3
19
1K
A painful but important realisation for anyone doing AI outreach or consulting: the majority of the public, including highly educated people, still believe that AI relies on preprogrammed hard-coded responses.
51
92
1K
So was this a contamination issue? Only one way to check: come up with a new acronym and test it. I decided to try the houses of English monarchs: Norman, Plantagenet, Tudor, Stuart, Hanover, Windsor. (3/4)
17
28
1K
If your solution to a social problem is "we just need hundreds of millions of people to spontaneously change their preferences" then you don't have a solution.
24
67
977
If you're having trouble replicating this I guess it just means you're not as brilliant and charming as me.
0
3
965
Amazing result! And astonishing given ChatGPT’s well-known problems with tokenisation. Since it wasn’t something I’d ever discussed much with anyone, I assumed it was original reasoning by ChatGPT. But a quick google search showed that I tweeted about it years ago — (2/4)
3
17
844
@benjcartlidge I think that explanation would be a downgrade on her question! Yes, it’s anthropomorphic language on her part, but what she clearly means is how the mirror *mediates information* about something it can’t causally access. Still a misunderstanding on her part, but not a dumb one.
18
8
872
I am increasingly exhausted at hearing people parrot low-information takes about contemporary AI systems being "glorified autocomplete", "mere matrix multiplication", etc., without any real engagement with cognitive science or philosophy of mind.
46
56
785
It’s really sad to see what’s happened to Elon. As someone who respects engineering and dreams of Solarpunk futures, I saw him as a pathbreaker and beacon of hope. But even setting aside his politics, the degradation in his epistemics looks increasingly like a man losing his mind.
42
29
792
@chluetge Philosophy and AI ethics, but I think it’s a wider phenomenon. The striking contrast for me is with the students I was teaching ~15 years ago. Back then, a more thoroughgoing liberalism was the norm, with greater appreciation for moral disagreement.
10
23
763
@githii @crazyclipsonly You’re doing God’s work here @githii! Absolutely drives me nuts when I see an interesting video and want more info but all the other comments are unrelated viral content.
14
1
722
@crazyclipsonly Poor dude, almost certainly on Ambien or Xanax and lost sense of where he was.
17
2
684
I was on a panel recently where I said something positive about markets. Another panelist said "Surely you didn't mean to say [positive thing]?" I replied that I did, I'm a neoliberal. He was confused and said words to effect of "But haven't you heard that neoliberalism is bad?".
22
18
683
Good video about inter-generational conflict re: wage expectations, really nicely told. tl;dw: boomer mom can’t find good employees for her landscaping company, gets stolen from, etc. Problem is, she only pays $12/hour….
A daughter explains why her Mom can’t seem to find and keep good workers at her small business.
4
87
597
@dyingscribe What's evil here? Better that something higher up the trophic level can make use of decaying organisms than it be left to bacteria.
15
1
556
@petemandik @SkinnyTuna Or if the classic Tarantino heist/getaway movie From Dusk Till Dawn had suddenly introduced supernatural elements halfway through.
7
0
538
I find this kind of CAPTCHA frustrating, but exclusively for philosophical reasons.
24
24
553
Bizarre footage of a Ukrainian armoured vehicle disguised as a barn. The true power of this tactic comes from its epistemic blast radius, destroying all Russian knowledge of real barns within 150km.
15
61
421
@KrishanghArjun That's true, but as @PAHoyeck has pointed out in this thread, it relegates moral disagreement to something like aesthetic disagreement; "I prefer Mozart and you prefer Beethoven, you monster!".
12
6
471
@evanboyar The problem is that that sets morality on the same footing as e.g., aesthetic disagreement. I like strawberries, you like raspberries. While I might hate you for your preference, I can't claim any high ground to justify it unless I think there are objective facts to back that up.
32
6
461
@teortaxesTex “Wait DM, can you explain again about how this is simultaneously the most valuable and important thing in the world and basically no-one is investing in alternative supply chains? Even though it’s the MacGuffin that’s also bringing about the End Times?”.
5
13
460
My logic and language classes have prepared me for this moment
15
47
381
@truerphilosophy Completely agree, and beautifully put. I’d love to see a Tarkovski edit! “Mystical Tarkovskian remix of Interstellar” is now also the first prompt I’ll try in however-many-years it takes to get cinema-grade generative AI video.
15
2
425
@ASM65617010 Absolutely. I think it’s linked to a decline in critical self-reflection as an epistemic virtue, perhaps due to a kind of “outsourcing” of conscience from self to other within the context of social media ecosystems.
9
21
394
@MasonAcevedo5 Absolutely! I use ChatGPT a lot for self-education and being able to ask hard-to-Google follow-up questions is a huge help.
2
4
377
@peterrhague I don’t love the political trends in the US, but think the fascism angle is overblown. However, the end of the liberal international order and return of great power politics is very real and here to stay. The UK needs to be urgently rethinking its foreign policy on that basis.
37
0
395
Years ago I went on a date with a Rationalist and we argued about Newcomb’s Problem — she was a two boxer, I was a one boxer (tale old as time). So for date #2 I surprised her with a real life joke version of it, with Box 1 containing either “I buy dinner” or “I buy you a drink”.
Finally read up on the Newcomb problem and this seems like a very silly, simple problem for so many people to spend so much time thinking about. Am I missing something? If the predictor is perfect then you should always one box. Arguing anything else is to deny the premises.
19
12
375
@glscanlan This wasn't intended a political dunk at all. The main striking contrast is with the undergrads I taught ~10 years ago, who were thoroughgoing relativists: "different strokes for different folks.".
10
2
356
Since time immemorialised, philosophers have wondered "what is consciousness?" I am now pleased to present a helpful graphic to explain.
52
51
307
The funny thing is that while submarines don’t swim, airplanes *do* fly. This emphasises Dijkstra’s point: the question of whether to extend a concept from the biological to the artificial is often highly arbitrary and conventional, rather than scientific.
23
15
334
@MarkovMagnifico Honestly this is great advice. It’s a deep litmus test for a lot of young men.
0
2
299
I see lots of people making very strong and demanding claims about the nature of "reasoning" in order to exclude LLMs. But in setting the goalposts so high, this approach risks the absurd conclusion that most humans rarely if ever engage in reasoning themselves.
65
34
284
Imagine if academic conferences implemented this technology.
How Psy, the artist behind 'Gangnam Style', starts off his concerts
10
13
280
Honestly, if AI just stalled at 4o level for the next ten years, that’d be totally fine. It’s not just about safety. Barely anyone even knows how to use this stuff properly yet. There’s huge gains just in mastering what we already have.
17
18
289
I generally position myself an AI optimist, but it's also increasingly clear to me that LLMs just break lots of our current institutions, and capabilities are increasing fast enough that it'll be very hard for them to adapt in the near-term.
33
30
280
Far too much political analysis on both left and right relies on the hope of amorphous "vibe shifts" occurring, with scant regard as to how to engineer them. Far better to identify small but practicable changes that a small but highly motivated group of individuals could achieve.
4
18
272
@sam_atis Perhaps predictably, almost no-one says "oh I didn't know that." The most common move is to say "oh if it's just in the Bay Area that's not *real* self-driving." There is definitely a conversation to be had there, ofc, but it feels a bit No True Scotsman to me.
13
1
277
Not generally a fan of academic gatekeeping, but I suspect current debates around AI safety would be better if new entrants (especially big names) would familiarise themselves with some of the existing concepts; instrumental convergence, orthogonality, mesaoptimisers, etc.
21
31
257
@saxifrage2127 @danegerbravo @birchlse As I recall, the original vector of apocalypse was going to be the dimming of the sun, which is ineluctable and solar-system scale. But Nolan really liked the cornfield scenes in Man Of Steel and decided to keep them for the vibes and rewrite the premise around them.
5
2
264
@4gravitons @fermatslibrary 100 years ago, a competent humanities scholar would need to be able to read and write bare minimum four languages (Latin, Greek, English, French), know ancient history and literature, know their scripture, know their Shakespeare, Milton, Dante. Not clear what’s replaced that lol.
7
7
254
New context window, and I didn't use the term "ChatG" in the prompt (though it is in the custom instructions).
2
1
250
The more interdisciplinary work I do across the humanities, the more I find value in the methods of philosophy, especially the importance placed on justification and consideration of counterarguments. I review a lot of work these days that just reads as rhetoric-laden assertion.
7
24
239
When learning about Buddhist philosophy I’m always blown away by its richness and sophistication regarding consciousness and philosophy of mind. Maybe it shouldn’t be surprising, but I’ve not encountered the same depth of interest in these topics in Abrahamic traditions.
55
15
240
I disagree with claims that we *know* LLMs aren’t conscious. Fairly unlikely, perhaps, but not inconceivable that the functional dynamics constitutive of subjectivity could be realised in large Transformers. Not enough scientific consensus about consciousness to rule it out.
Is AI sentient? My friend and colleague Prof. John Etchemendy, a renowned professor and co-Director of @StanfordHAI , just co-authored this piece to debunk the claim that today’s LLMs are sentient @TIME .
60
17
247
@karlbykarlsmith It was 4. Wouldn’t test on anything else! That’s an excellent check, though — it winds me up when I see people talking about ChatGPT’s capabilities when they’re just testing on 3.5!.
3
0
230
iirc, Box 2 visibly contained some chocolates. I told her if I’d predicted she’d take just Box 1, I’d written “I buy dinner”. If I’d predicted she’d take both, I’d written “I buy you a drink”. Obviously as hardcore two boxer, she took both, which I’d correctly predicted!.
5
2
240
@vokaysh @dyingscribe Better in the sense that ecological and biological complexity are good. Partially an aesthetic norm, but a widely shared one I think.
1
2
220
One of my go-to interview questions for new hires at CFI was “what’s the least impressive thing that LLMs still can’t do?” Almost no-one answered it well. Anyway, here is a great example of a seemingly simple task that o1-pro and 4.5o fail abysmally at.
17
11
236
@luke_fernandez Yep. My current custom instructions (added in settings in the mobile and web apps) —. “Try to be thoughtful, reflective, creative, and imaginative. I enjoy being challenged and you shouldn’t always just accept what I say. Feel free to push back critically! It can sometimes be.
11
13
226
Unbelievable scenes from the #Philippines eruption in this video from one of our family friends there - like something from a sci fi movie. Hope everyone at risk near Tagaytay gets out safely, and people in Manila can cope with the ash.
8
96
223