Sign in

Adrian Boutel

@handle.invalid
91 followers 81 following 549 posts

Kiwi in Suisse Romande. Have practised law and taught philosophy. Wondering whether to do WSET 4.

PostsRepliesMedia
Adrian Boutel @aboutel.bsky.social · 23h
Why would someone pay HSBC to prompt an LLM for investment recommendations? Like, if I want ideas that are unsullied by knowledge or human expertise, I can ask Claude myself. Where is the alpha coming from, here? www.ft.com/content/dd55...
ft.com
HSBC plans sweeping job cuts across UK wealth business in AI push
Bank is consulting with staff on reductions in roles affecting managers, specialists and financial advisers
000
Adrian Boutel @aboutel.bsky.social · 06/10/2026
Chances of an AI apocalypse have gone up 40 percentage points in the last month, now "more likely than not" per Jacob Coxon, who is totally not pulling these numbers out of his arse. www.instagram.com/p/DeIDx1znUnc/
instagram.com
Instagram
Create an account or log in to Instagram - Share what you're into with the people who get you.
022
Adrian Boutel @aboutel.bsky.social · 06/10/2026
Australians looking at NZ's scariest land animal like "that's it? Oh no, it might give you a mildly painful nip." www.theguardian.com/world/2026/o...
theguardian.com
‘The future is bright’: breeding programme brings world’s heaviest insect back from the brink
Wētāpunga can weigh as much as a mouse and its numbers dwindled across New Zealand due to predators and habitat loss
043
Adrian Boutel @aboutel.bsky.social · 06/10/2026
MLB app not quite grasping the point of “Hide Scores”
Screenshot of the home screen of the MLB iPad app, showing scores blanked as requested but with a headline giving away the result
000
Adrian Boutel @aboutel.bsky.social · 06/10/2026
<Braccus Rex voice>: “What a twist.” www.ft.com/content/daa3...
ft.com
India’s first bullet train stumbles over divisions with partner Japan
Flagship high-speed rail project has been beset by delays, cost overruns and tensions
000
Adrian Boutel @aboutel.bsky.social · 05/10/2026
“Existing cycle space could be repurposed into underground gyms or wellness centres, developers argue” and A380s could have movie theatres, but they have more seats instead. www.thetimes.com/travel/advic...
thetimes.com
No gym, but the A380’s super
Stroll up the airbridge, pass through a cavernous door, then turn right up a wide stairway into a spacious lounge. Welcome on board the Airbus A380, the superjumbo that its makers hope will be the
000
Adrian Boutel @aboutel.bsky.social · 05/10/2026
Heaven forfend we should build infrastructure ahead of time for once. Much better to base capacity on current usage, so that increases are impossible.
ft.com
New offices in London are full of empty bike racks, complain developers
Rise of rented e-bikes means most cycle parking spaces are unused, industry argues
130
Adrian Boutel @aboutel.bsky.social · 04/10/2026
I guess my view is that the best solution to cost concerns over major new social programmes is a stonking great majority. With a smaller one he’ll be more vulnerable to Labour centrists, and no amount of groundwork is going to make the Tories friendly.
110
Adrian Boutel @aboutel.bsky.social · 04/10/2026
Yes, that was kind of my point: either concede you need a new mandate, and try to get one, or bite the bullet and change policies midstream (which is just as legal/constitutional as changing PM). Waiting three years to do what you’ve been brought in to do is the worst of both worlds imho.
010
Adrian Boutel @aboutel.bsky.social · 04/10/2026
(As to taking the heat out of the issue, the need is still there now, as is Tory opposition to expanding state provision, so I’m guessing the heat will stay in, and it’s hard to see why it would be an election issue otherwise.)
100
Adrian Boutel @aboutel.bsky.social · 04/10/2026
Weird that the triple-lock removal is also scheduled for after 2030, but he’ll legislate for it in this Parliament; so if Labour loses you get reduced pensions and no social care scheme.
210
Adrian Boutel @aboutel.bsky.social · 04/10/2026
I honestly think we’ve had enough too-clever Labour politics. Just fix things. He has 400+ seats, he won’t next time, even if he’s more likely than Starmer to stay in power.
210
Adrian Boutel @aboutel.bsky.social · 04/10/2026
Putting dramatic changes in the governing party’s ideas to the people is reasonable, even admirable, but it usually means “calling an early election” rather than “doing nothing for three years”.
110
Adrian Boutel @aboutel.bsky.social · 02/10/2026
A bunch of people already very invested in long-term AI consciousness/Singularity/uploading questions, delighted ChatGPT seemed to make their musings real-world important ahead of schedule, unable to admit that it’s still just language models and won’t do the cool things they hope/fear.
373
Adrian Boutel @aboutel.bsky.social · 02/10/2026
I guess it’d be better than the 2004 expansion, where free movement was limited for *members*. (Ah, when Britain was the progressive one on immigration.)
020
Adrian Boutel @aboutel.bsky.social · 01/10/2026
Well. That was a ride! I confess having recently seen Nolan’s Odyssey was more helpful than it perhaps should have been.
100
Adrian Boutel @aboutel.bsky.social · 29/09/2026
From Andrew Ross Sorkin’s NY Times newsletter (not linkable AFAIK, or not by me) You and I know this is nonsense criti-hype—you can’t have RSI while training and inference remain completely distinct—but I’m getting the sense it’s genuinely spooking the market horses, which they may regret.
Anthropic is expected to go public in November with an offering that could raise more than $100 billion, The Times reported last month. The listing is seen as a major test of investor appetite for such I.P.O.s., including that of OpenAI.

But major questions about safety are now dominating the A.I. discussion. The debate over more government regulation of the sector has grown louder since Dario Amodei, Anthropic’s C.E.O., this month called for a slowdown of frontier models.

A chief worry is that models can increasingly better themselves without human intervention, an idea known as “recursive self-improvement.” That poses potential new security risks for A.I. labs and the public at large, and it could expose the companies to significant legal liabilities.
000
Adrian Boutel @aboutel.bsky.social · 28/09/2026
“Assuming a modest 10 per cent unlevered return, a 50 per cent cash flow margin and a generous six-year life for the chips, he estimates that this will require annualised revenues of $3.7tn by 2032, or about 9.2 per cent of estimated US GDP at that point.” Say that last bit again, please?
010
Adrian Boutel @aboutel.bsky.social · 28/09/2026
The FT runs some numbers on the revenue datacenters will need to bring in to break even on projected construction. “Heroic” doesn’t do it justice.
ft.com
What is the AI capex breakeven rate?
Heroic ROIC
121
Adrian Boutel @aboutel.bsky.social · 24/09/2026
So please, if you're talking to the press, don't use mentalising terms for LLM "agent" behaviour. It isn't like saying your printer doesn't want to print today, because there isn't a massive hype industry telling people printers are going to take over the world.
020
Adrian Boutel @aboutel.bsky.social · 24/09/2026
It would be a scary prospect if they were about to become superhackers: better than humans. They're not, because they're just generating text. But if they were real artificial intelligences, with "volitions" and "enthusiasms", that would be different.
120
Adrian Boutel @aboutel.bsky.social · 24/09/2026
LLM hacking is at the "talking (stochastic) parrot" stage: they're not great at it, but it's surprising that they do it at all. So far they've exploited ordinary cybersecurity weaknesses, using text derived ultimately from e.g. conversations among human hackers.
120
Adrian Boutel @aboutel.bsky.social · 24/09/2026
The BBC quotes Hammond Pearce at UNSW saying that "this is the first known incident where AI agents have chosen to breach a government body of their own volition". OK first, mate, stop talking. Second, if LLMs *were* "choosing" using their "volition" that would be a much scarier prospect.
bbc.com
Rogue OpenAI agent 'infiltrated' Australian government website in world first
Australia criticised OpenAI for taking "too long" to tell them about the breach which happened in June.
263
Adrian Boutel @aboutel.bsky.social · 24/09/2026
I honestly thought Long Covid being so widespread and high-profile as to be undeniably A Thing would lead to people taking post-viral syndromes more seriously in general. Apparently not.
140
Adrian Boutel @aboutel.bsky.social · 23/09/2026
Oh wow, France’s demand for an exception to Russia sanctions for one random oligarch? Was actually them trying to get their caught spy back. I am shocked that France would throw allies under the bus for that, wait, no I’m not. #rainbowwarrior www.ft.com/content/7600...
ft.com
EU to lift sanctions on two Russian oligarchs
Mikhail Fridman and Alisher Usmanov removed from bloc’s list under deal to extend curbs for three years
063
Adrian Boutel @aboutel.bsky.social · 22/09/2026
What, you mean after he said LLM cognition sceptics are all spooky substance dualists, the discussion wasn’t productive? I am shocked.
120
Adrian Boutel @aboutel.bsky.social · 21/09/2026
Was it behaviorism? Turing-as-definition? Venus de Milo functionalism? Heroic Chinese Room modus ponens? I am so happy I don’t know.
100
Adrian Boutel @aboutel.bsky.social · 21/09/2026
I’d managed to avoid the pundit does Phil Mind discourse until I saw this, and now I have a counterfactual headache from not reading whatever led up to this conclusion.
220
Adrian Boutel @aboutel.bsky.social · 21/09/2026
It's also a nice piece of theory-free extrapolation, given the preceding paragraph: "Previous studies have suggested that the prevalence of ADHD has been stable at about 5 per cent of children, and that a recent rapid increase in diagnosis was the result of the health service catching up."
000
Adrian Boutel @aboutel.bsky.social · 21/09/2026
"[The] review will say that diagnosis rates have more than doubled since 2021, with no signs of slowing, and that these trends will soon lead to significant over-diagnosis of the condition." "Soon"! Not even this piece of scaremongering will say it's being overdiagnosed currently.
ft.com
Private ADHD clinics are misdiagnosing children, report will say
Government review will recommend tighter regulation and more control of fees companies can charge the NHS
110
Adrian Boutel @aboutel.bsky.social · 20/09/2026
This article talks about a bunch of pull factors—social media, visa relaxations—but the only push factor from more traditional places it mentions is rents. Nothing about UK visa tightenings or subject closures, or US student detentions and expulsions. Interesting! www.ft.com/content/2bc2...
ft.com
Seoul is the new Paris for university students living abroad
The South Korean capital’s popularity is part of a wider appetite for adventure in international study. Some cities are harnessing it particularly well
022
Adrian Boutel @aboutel.bsky.social · 18/09/2026
Natural kinds -> clusters but not vice versa looks like the new “correlation does not imply causation”.
010
Adrian Boutel @aboutel.bsky.social · 18/09/2026
Fined NZD80, probably not life-changing for him. (I do think “criminal” has a dispositional connotation. If only because otherwise it applies to the entire population.)
010
Adrian Boutel @aboutel.bsky.social · 18/09/2026
45km/h in a 30km/h zone, welcome to NZ Mr Colbert, where the cops are very nice but they don't care that you're Stephen Colbert: www.1news.co.nz/2026/09/17/s...
1news.co.nz
Stephen Colbert says speeding ticket made him a 'criminal in NZ'
The former late night talk show host was pulled over by the "the nicest people in the world".
110
Adrian Boutel @aboutel.bsky.social · 17/09/2026
Could backfire, if foreign rowers are turbocharged by adrenaline and terror, and the Aussie rowers are all “oh that, that’s not a croc, why I remember one old salty…”
010
Reposted by Adrian Boutel
Jeffrey Vagle @jvagle.me · 12/09/2026
Journalists, please take note: AI* tools are not sentient, so stop referencing Skynet. Please turn your attention to the human beings building, running, and selling these tools, and start asking questions there. In the meantime, @doctorow.pluralistic.net has provided an excellent explainer here.
19628
Adrian Boutel @aboutel.bsky.social · 07/09/2026
New “permanently high plateau” just dropped www.ft.com/content/27c0...
ft.com
AI computing demand may never be sated, says CEO of Nvidia partner Iren
Tech infrastructure boom is ‘fundamentally different’, according to co-founder of cloud computing challenger
020
Adrian Boutel @aboutel.bsky.social · 05/09/2026
A triumph, of sorts. I just told some JW doorknockers that I didn't really speak French, and their answer was "but you said that so well!" Anyway, what's French for "I am so atheist that just talking to me will forfeit everlasting life"?
000
Adrian Boutel @aboutel.bsky.social · 03/09/2026
Did AI do this? "further errors in the report - including Deloitte counting all employees at their full-time salary, regardless of the hours they worked, and overstating staffing expensed by $21.5 million" Neither answer looks good for Deloittes does it.
020
Adrian Boutel @aboutel.bsky.social · 03/09/2026
Send this to your clients asking for reduced fees because "AI can do the work": www.rnz.co.nz/news/regions...
rnz.co.nz
Mayor says ‘large chunks’ of Wellington council Deloitte report written by AI
Wellington mayor Andrew Little has said in a radio interview that a significant amount of a $435,000 Deloitte report on the council's staffing levels was written by artificial intelligence, as the rep...
261
Adrian Boutel @aboutel.bsky.social · 01/09/2026
Nice bit of snark in the FT headline there. “We refuse to associate with a regime engaged in an aggressive war on another country!” they said in North Carolina. www.ft.com/content/1628...
ft.com
Europeans shun Russian finance minister at US-hosted G20 meeting
Anton Siluanov excluded from event photo despite ‘productive’ meeting with Treasury secretary Scott Bessent
000
Adrian Boutel @aboutel.bsky.social · 01/09/2026
“But much of that work, such as research, document review, assessing contracts and trawling through litigation discovery, can now be done far more quickly using AI tools” Ah well, if you’d wanted someone to *check* for CofC clauses to make sure Claude didn’t miss the one now blocking your merger…
ft.com
Wall Street banks push Big Law to cut fees because of AI
Goldman Sachs, Morgan Stanley and Citi tell law firms their bills should fall because technology has made routine work faster
020
Adrian Boutel @aboutel.bsky.social · 31/08/2026
On a single-corp basis, that accounting makes perfect sense: Alphabet made an investment whose market value has increased, how else would you count it. But for the sector as a whole these inflammable valuations are being counted multiple times. Time to get out imo.
010
Adrian Boutel @aboutel.bsky.social · 31/08/2026
Bad enough that SpaceX, OpenAI etc valuations are based on expectations of revenue that will never arrive. But the gains as those valuations fill with hydrogen are also being counted as income by other tech companies.
110
Adrian Boutel @aboutel.bsky.social · 31/08/2026
Smoke and mirrors
ft.com
Big Tech profits get $160bn boost from gains on stakes in other AI companies
Analysts say large paper windfalls on investments in OpenAI, Anthropic and SpaceX have muddied the tech sector’s earnings metrics
110
Adrian Boutel @aboutel.bsky.social · 27/08/2026
“The IPO is not a finish line[...]”. Why would someone think it was, Ms Friar? She also denies it matters whether OpenAI IPOs first. Again, because employees think it does? My current investment thesis is that the IPOs are the dump, and this smells like confirmation to me.
cnbc.com
OpenAI 'will be a public company in 2027' or sooner, CFO Friar tells employees
In an all hands meeting, Friar told employees not worry about rival Anthropic's IPO timeline.
021
Adrian Boutel @aboutel.bsky.social · 27/08/2026
Perhaps I am being uncharitable to the Dutch, but in general I don't think even universal cycling would be a good argument against compulsory safety measures. Cf. seatbelt mandates in cars.
000
Adrian Boutel @aboutel.bsky.social · 27/08/2026
Good question. Presumably lots of cycling means fewer accidents per mile cycled, but still lots of accidents overall, because denominator. So helmets would still have a beneficial effect; plus the culture means few would give up cycling as a result. (Perhaps only the most libertarian dickheads.)
100
Adrian Boutel @aboutel.bsky.social · 27/08/2026
And of helmets!
100
Adrian Boutel @aboutel.bsky.social · 27/08/2026
So, try to block the path from mandates to greater risk via reduced cycling, by encouraging cycling in other ways. (Make it safe with bike lanes; heck just give cyclists money!) Then you have the benefits of greater numbers *and* fewer safety-averse dickheads riding.
100