Made of Triangles

The defense team presses Lairmore on whether the sandwich really 'exploded.' They return to the photo of the sandwich and wrapper on the ground.

"That sandwich hasn't exploded at all, has it?" defense asks.

"It looks like a little bit is coming out towards the bottom," Lairmore replies.

New York Post headline: Nearly a million New Yorkers ready to flee NYC if Mamdani becomes mayor - possibly igniting largest exododus in history: poll

You laugh, but this is how the exododus went extinct

It can help, although might be difficult to distinguish between posts documenting something bad vs encouraging something bad.

I just worry that straight content rules are too likely to run into intentional or effective malicious compliance because in the end it's not at all an easy problem.

Users numbers don't scale cleanly with profitability. That's why so much effort goes into 'engagement'.

A system that focused on matching levels of moderation with levels of engagement might be useful (hence my preference for biz-model/ads focus) but video/images/memes would make AI less useful.

I can appreciate the logic of that, agree that they should exist, but I suspect that middling effort would look like almost no effort at all from the outside.

It's just insane numbers - over 5000 posts per second on X. Assume a 1 in 1000 troublesome post and you're looking at 450k a day

What matters for my vote under FPTP is the voting landscape locally and how that interacts with likely outcome of 649 other local votes.

Happy to see a coalition with Green Party in it wouldn't want to see them with a majority.

My choice in the next election is:

1) Vote for someone who loses
2) Vote for someone who I don't want to win

So my very trivial contribution in 2029 will be 1 with the
vague and almost certainly vain hope of providing the merest featherlike nudge to policy positions in the 2034 election.

The more support the greens get the less likely I am to vote Green. Some sensible impulses and deserve to be listened to but a little too stuck in wishful thinking to be government material.

A criticism I'd level at Labour as well at the moment, although for different policy positions.

By what mechanism would you do that?

I think you might be able to mitigate and moderate it a bit via advertising taxation, particularly if you could structure it so that social media companies were encouraged to charge organisations for accounts, but solid guardrails looks like a hard problem.

Not an outcome I like, but so much an outcome I expect.

I think it's what happens when libertarianism collides with nationalism. Trying to express ideas like "benefit society" but where only individuals matter, but only about individuals of one nation without caring about the nation as a whole (because individuals). Ends you up in a bit of a paradox.

I am equally opposed to Monarchy or Presidency - unless we are rebuilding the whole constitution from ground zero we could and should replace the Head of State with a committee of ex-judges for nodding things through, a glorified receptionist for shaking hands and an old tree for symbolism.

I think we need to get away from the trap of thinking that algorithms 'make decisions' other than in a very vague colloquial way.

Algorithms are the encoding of pre-made human choices into a repeatable form.

ladies and gentlemen...we got him

Plus she had a degree that should have covered the act at least briefly as part of the course. I'd be willing to grant the benefit of the doubt to a 17 year old who claimed they hadn't heard of it.

28 and with a computer related degree? Not so much.

Yes, because Badenoch’s was a criminal offence carrying a custodial sentence, whereas Reeves’ was not.

If there is a skill issue to getting a benefit from AI then it seems it's more likely to be a bubble in terms of investment. It will effectively gate benefits based on the availability of (presumably expensive) human resources and cap the utility of infrastructure investment.

To a man with a generative AI, all problems begin to look like finding the most plausible sequence of tokens.

Worth noting there that certain types of software project models cannot be applied to certain types of software products.

Continuous deployment does not work well on products that are sealed in the factory before being shipped to customers never to be unsealed again.

Ah, maybe a term of art thing. If data can't be used to correctly infer the answers I need it isn't reliable (ie cannot be relied upon) - even if it is accurate.

"Extremely predictable error discovered far later than it should have been"

If someone didn't point out the likely error before the scheme started it seems the planning was a bit negligent.

They should have been aware that the data they were using wasn't reliable.

All out attack on Farage?

Given a majority of the public really don't like him I wonder if it's the only approach that:
1) Gets media attention
2) Unifies much of the country

Possible that if you combine every event with an attack on Farage you get sympathetic ear from 50-60% of the public.

I think VAT on everything vastly simplifies the case for lowering the VAT threshold as well - the administration overhead (and associated risk with getting the admin wrong) is as much a disincentive from going past the limit as the Tax, if not more.

The ring was only destroyed (and note the destruction of a source of power and control not violence against an individual or group) because of loyalty, stoic stubbornness, patience and a pattern of repeated mercy towards Gollum.

Who was I suppose a stateless refugee with a criminal past?

Eternal curse of domain knowledge - sprains from wincing so hard when domain usage of terms and colloquial usage of terms clash when 'experts' are marketing their ideas to the public.

Tech companies were discussing this internally (at the level of "come up with ideas that solve this that we can make into products!") eight, nine years ago. Not specifically around gen-ai but in practice the technology used to generate disinformation doesn't matter.

Limited success though.

Gelsinger might not be the best example of silicon valley power.

He did run a very large and complex manufacturing business, but he didn't run it very well. Not least because he was powerless against a number of external trends and internal choices that pre-dated his appointment.

Also, anyone who believes ad demographics should be expected to resemble national demographics is utterly stupid and can be safely ignored on all topics involving statistics, mathematics or processes more complicated than making a cup of instant coffee.

It's hard to put a precise number on it, but I think for about 80% of applications of AI I've seen so far we'd be better off without it.

Simply because it does nothing that needed to be done while costing resources that could be better used elsewhere

Not sure it won't be a net drag on growth.

Adverts do not reflect life is the first rule of advertising and the first rule of media literacy.