It would make strategic sense, even from a selfish perspective, for essentially any nation to get more involved in sensible AI regulation, and to get more involved in capturing AI benefits and being the home to AI and AI-enabled economic engines. It makes especially good sense for the UK, which is facing numerous economic headwinds and yet still has the legacy of its history and of London and DeepMind, of Oxford and Cambridge, and has broken free of the absolutely bonkers regulatory regime that is the European Union. This is their chance, and Sunak is attempting to take it.
Hilarious way to encapsulate all the buzz out there that I could not possibly read otherwise. Love it. Also, we need more critical thinking like this across the board, in the boardroom, and in newsrooms for sure. Thx.
Nice game for playing around with prompt hacking: Gandalf AI, where you have to convince the AI to tell you the password against increasingly complex prompts. https://gandalf.lakera.ai/
re: “Some companies will doubtless go the Chase Bank route and use AI to do a bunch of nasty surveillance. Most companies mostly won’t, because it’s not a good strategy, your best workers will leave, and also the jobs where such hell is worthwhile will likely mostly get automated instead.”
I’m much less optimistic of this, especially in my sector. Or rather, the best case scenario as I’d put it is that, instead of using them only to """enhance""" workflow, companies will use this technology to replace the abstract metrics Goodharting loops that currently seem to be all the rage in order to get more realistic pictures of what shouldn’t be sacrificed for imaginary productivity. My distant hope is that this guts middle management before it replaces store employees.
I’m concerned “your best workers will leave” will not be sufficiently strong motivation to avoid making work more hellish. Something is currently convincing chain pharmacies (and I see this pattern a lot in other retail/customer service operations) to jettison experienced staff in favor of high attrition/high turnover and ludicrous understaffing and while I can’t say whether that’s actually working out for them in some sustainable financial way behind the scenes it doesn’t -seem- to be working for them broadly from either side of the counter at the ground level—employee quality of life and customer service both seem to be worsening over time. So what I would expect to happen sooner is that AI tech will mostly be used to make this situation worse, to pressure out older staff to bring in new staff they can underpay and not have to justify the new tools to.
My favorite personal understaffing anecdote was wondering how many jobs aside from pharmacist require a doctorate and also require you to bag groceries.
The other day I came across an old tweet from last year where I was reacting to some Walgreens fluff PR where they were claiming that most of their pharmacy business would be in quasi-automated remote fulfilment centers by 2025. My comment at the time was that this seemed very unlikely to me because Walgreens specifically has been continually unable to step in the direction of both more automation and more centralization (most stores in my region were -losing- fill robots because the lease on them was apparently too expensive, and most of their central pharmacy support had been gutted by the time I left). I’m still mostly confident this is unlikely by 2025 but AI tech will certainly lower the hill they’d have to climb to do it.
CVS is a slightly different story, you could probably train currently existing AI to interface with their virtual verification pretty quickly and they could probably fire half or more of their pharmacists no sweat in a few months if they wanted to rush in that direction. If anything, pharmacy -technician- jobs are more secure because they’re having the same sort of automation-is-actually-too-expensive problem as well.
"From 25 May: Paper points out that fine tuning from the outputs of a stronger model will help you mimic the stronger model, but only in a deeply superficial way, so if you don’t stay very close to the input-output pairs you trained on, it falls apart, as you’d expect, and it is easy to fool oneself into thinking this works far better than it does. "
"I predict this plan to be less surprised by the pace of developments is going to be rather surprised by the pace of developments - my initial thought was ‘this is describing a rather slow surprisingly slow developing version of 2027.’"
The extinction question puzzles me. No human ancestor is extinct, however far back you go. Any species any of whose members have living descendants has not gone extinct. It doesn't matter how different those descendants are from the original species.
Small typo
knows how is likely ->
knows who
Hilarious way to encapsulate all the buzz out there that I could not possibly read otherwise. Love it. Also, we need more critical thinking like this across the board, in the boardroom, and in newsrooms for sure. Thx.
Nice game for playing around with prompt hacking: Gandalf AI, where you have to convince the AI to tell you the password against increasingly complex prompts. https://gandalf.lakera.ai/
Nit: It's Rohin Shah, not Robin Shah.
re: “Some companies will doubtless go the Chase Bank route and use AI to do a bunch of nasty surveillance. Most companies mostly won’t, because it’s not a good strategy, your best workers will leave, and also the jobs where such hell is worthwhile will likely mostly get automated instead.”
I’m much less optimistic of this, especially in my sector. Or rather, the best case scenario as I’d put it is that, instead of using them only to """enhance""" workflow, companies will use this technology to replace the abstract metrics Goodharting loops that currently seem to be all the rage in order to get more realistic pictures of what shouldn’t be sacrificed for imaginary productivity. My distant hope is that this guts middle management before it replaces store employees.
I’m concerned “your best workers will leave” will not be sufficiently strong motivation to avoid making work more hellish. Something is currently convincing chain pharmacies (and I see this pattern a lot in other retail/customer service operations) to jettison experienced staff in favor of high attrition/high turnover and ludicrous understaffing and while I can’t say whether that’s actually working out for them in some sustainable financial way behind the scenes it doesn’t -seem- to be working for them broadly from either side of the counter at the ground level—employee quality of life and customer service both seem to be worsening over time. So what I would expect to happen sooner is that AI tech will mostly be used to make this situation worse, to pressure out older staff to bring in new staff they can underpay and not have to justify the new tools to.
My favorite personal understaffing anecdote was wondering how many jobs aside from pharmacist require a doctorate and also require you to bag groceries.
The other day I came across an old tweet from last year where I was reacting to some Walgreens fluff PR where they were claiming that most of their pharmacy business would be in quasi-automated remote fulfilment centers by 2025. My comment at the time was that this seemed very unlikely to me because Walgreens specifically has been continually unable to step in the direction of both more automation and more centralization (most stores in my region were -losing- fill robots because the lease on them was apparently too expensive, and most of their central pharmacy support had been gutted by the time I left). I’m still mostly confident this is unlikely by 2025 but AI tech will certainly lower the hill they’d have to climb to do it.
CVS is a slightly different story, you could probably train currently existing AI to interface with their virtual verification pretty quickly and they could probably fire half or more of their pharmacists no sweat in a few months if they wanted to rush in that direction. If anything, pharmacy -technician- jobs are more secure because they’re having the same sort of automation-is-actually-too-expensive problem as well.
"From 25 May: Paper points out that fine tuning from the outputs of a stronger model will help you mimic the stronger model, but only in a deeply superficial way, so if you don’t stay very close to the input-output pairs you trained on, it falls apart, as you’d expect, and it is easy to fool oneself into thinking this works far better than it does. "
And then, there was Orca.
https://arxiv.org/abs/2306.02707
I genuinely cannot parse this paragraph:
"I predict this plan to be less surprised by the pace of developments is going to be rather surprised by the pace of developments - my initial thought was ‘this is describing a rather slow surprisingly slow developing version of 2027.’"
Link from the following text leads to... not being able to find the relevant thread?:
"Andrew Konya proposes an AI that ‘maximizes human agency.’"
re concerns about environmental impact, i assume its referring to the carbon emissions involved in training the models
The extinction question puzzles me. No human ancestor is extinct, however far back you go. Any species any of whose members have living descendants has not gone extinct. It doesn't matter how different those descendants are from the original species.