Skip to main content


I was just talking to a colleague about the AI bubble. These companies are in so deep they can't tell the truth. They are all lying about the efficacy, costs to consumers and most importantly how & when this tech works or doesn't.

Is there enough money on the line to kill over?

There's likely a trillion bucks of valuations across the industry. Billions in sunk costs, billions in c suite remuneration, billions in VC mgmt costs.

RIP Suchir

mercurynews.com/2024/12/13/ope…

#OpenAI #AI #VC #SuchirBalaji

This entry was edited (1 week ago)
in reply to nullagent

This whole situation is about as believable as a Russian dissident or oligarch falling out of a midrise window in Moscow. Like the Boeing whistle blower I don't think it much matters if "foul play" was involved because this whole scenario is foul af.

This young man was very brave and righteous to blow the whistle and was undoubtedly under immense social and professional pressure not to. He's fucking up A LOT of people's grifty gravy train.

We should read his words

suchir.net/fair_use.html

#AI

#AI

reshared this

in reply to nullagent

I was explaining to a non-tech friend how "AI" works - ie what an LLM is, and does, including hallucinations and the fact that it's basically spicy autocomplete.

Once he understood, he was open-mouthed, aghast. "But all the media articles saying its going to replace 80,000 jobs, and make our lives easier? Is that all BS?"

It was depressing watching it sink in how deep the lie was that's been sold by all these huge corps.

@cstross

in reply to Mark Otway

@markotway @cstross

Oh it might well replace those jobs. And it'll be absolute dogshit at doing it.

The management class are dumb as fuck.

in reply to [intentionally left blank]

@drunkenmadman
This is a common refrain over on @pluralistic’s blog: AI might not be able to do your job, but that doesn’t mean they can’t convince your boss of that.

@markotway @nullagent @cstross

in reply to 0xC0DEC0DE07E8

A small variation on this reasoning is that AI may not be able to do your job, and management may know that all too well, but if a critical mass of companies replace people with AI, everyone will just lower their expectations and management will be fine.

See also: Self-checkout in retail.

This entry was edited (5 days ago)
in reply to 0xC0DEC0DE07E8

@c0dec0dec0de And I think one of the major lessons in the recent Hollywood writers' strike is that in many instances AI won't really replace humans but will be an excuse to pay them less. (You're not the screenwriter anymore; you're revising the AI's script.) @drunkenmadman @pluralistic @markotway @nullagent @cstross

reshared this

in reply to Thad

@Thad @c0dec0dec0de @drunkenmadman @pluralistic @markotway @cstross

Women have experienced this process over and over.

Male manager retires or quits, woman gets the job but it's now called a "coordinator" job at 40% less, because a typewriter & secretary was replaced by a laptop & word processor.

Same job responsibilities, different title, less money.

Wage suppression tactics using titles and tech is an old old technique.
nytimes.com/2016/03/20/upshot/…

msnbc.com/know-your-value/care…

in reply to Mark Otway

@markotway @Rob_T_Firefly @cstross I dunno. I definitely see low/early skill knowledge workers being tossed. Which means there will be no senior people sometime in my lifetime. Which means more AI to counter that. Which means less jobs. I think you did your friend a real disservice by dismissing AI as “spicy auto complete”. It’s taking jobs today, even in its infancy.

I don’t have a position on whether that’s good or bad, though. Throughout history jobs have fallen by the wayside due to progess and new jobs show up to support that progress.

This entry was edited (6 days ago)
in reply to ᒍᗪᙎ 🇨🇦

@jdw @markotway @Rob_T_Firefly You need to distinguish between generative AI (spicy autocomplete) and analytical/pattern recognition AI (which works surprisingly well). And also remember it's not "intelligence" in any human sense, it's just statistical modelling on a huge scale.

The term "AI" is pure sales hype, to extract money from private equity and sell the tech to oligarchs who can use it to attack their work force.

in reply to Charlie Stross

@cstross @jdw @markotway @Rob_T_Firefly @nullagent Here's the thing about pattern matching AI: training is a bitch, and it takes a lot of data/effort (in most cases) to keep it up to date. So even if it does work, it's usually not sustainable.
in reply to madopal

@madopal There are areas where obtaining it the data to train it is unobjectionable—mostly the observational sciences and some areas of medical diagnostics. Alas, these are not the applications of AI that attract the snake-oil investor dollars for some reason.
in reply to Charlie Stross

@cstross It's not so much about the objectionability as much as the work to assemble/verify/train. And if the set can change, that process is usually arduous and horribly manual.

But yes, the hidden surface is what it's trained on & how, and I hate that even when it works, that whole aspect is often very obscured.

As usual, Mike Judge/Silicon Valley nailed it.
youtu.be/vIci3C4JkL0?t=51

in reply to Charlie Stross

@cstross @jdw @markotway @Rob_T_Firefly
“You need to distinguish between generative AI […] and analytical/pattern recognition AI”

Thank you.

It's simply mind-boggling how many people just inanely repeat the anti-"AI" mantra without any understanding of the scope of the field.

That said:

'The term "AI" is pure sales hype’

Well, no; it's a term that's been used — however “unfortunately" — for well over half a century also with rather broad scope.

in reply to mmalc

@mmalc @jdw @markotway @Rob_T_Firefly I'm aware of the origins of the term at the Stanford Conference: problem is, like "crypto", "hacker", and "cyber" before it, it has now been mutilated in public parlance to signify something other than its original meaning.
in reply to Charlie Stross

@cstross @jdw @markotway @Rob_T_Firefly There's the old saying/joke that if it works then it's not AI. AI research started with playing chequers, but Turing's min/max algorithm solved that. Then text recognition, optimising compilers, machine translation... The list of things that were AI before they became solved problems is long. LLMs and diffusion models have simply joined that list.
in reply to Charlie Stross

@cstross the problem with making that distinction is that it’s less and less technically accurate as folks notice that actually, large language models often perform better at their analytical workloads than their custom-trained ML models ever did.

Turns out it’s predicting the next token all the way down.

in reply to Charlie Stross

@cstross @jdw @markotway @Rob_T_Firefly The more useful, pattern-matching AI still needs to have its results looked over by an actual human to determine if what it discovered is actually real. It's a "well, that kinda looks something" generator.

I think pattern-matching AI is actually a fair amount worse than a human doing the same thing, it's just a whole lot faster, so combining it with someone checking its results ends up being just as good but faster.

in reply to Troed Sångberg

@troed @markotway @cstross

Its an interesting point. Imo this is the crux of the issue, what the AI craze has proven for me is how astonishingly efficient the human minds mix of physics and chemistry truly is.

Digital computing, tho impressive that spicy autocomplete can do so much, its astounding that people believe the energy expense is worth the squeeze. For sure, LLMs are worthy of researching but this is all clearly not economically viable presently and is an artifact of VC speculation.

This entry was edited (5 days ago)
in reply to nullagent

Brains are indeed extremely energy efficient at a level we don't understand, but I was more alluding to that according to all known science we're also just deterministic autocomplete machines.

This is such an inconvenient fact though so we tend to just claim - without any scientific backing - that "there must be more to it".

@markotway @cstross

in reply to Troed Sångberg

@troed @markotway @cstross >> according to all known science we're also just deterministic autocomplete machines.

No. The scientific consensus on human cognition says the opposite of this.

in reply to Misuse Case

@MisuseCase

Eh no. We know of no physics or chemistry that can result in anything except deterministic autocomplete.

(Thus you have some who talk about "quantum tubes" or pure dualism with "a soul outside of physics" etc)

Sabine Hossenfelder has a good video on why she doesn't think free will exists.

@nullagent @markotway @cstross

in reply to Troed Sångberg

@troed @markotway @cstross >> We know of no physics or chemistry that can result in anything except deterministic autocomplete.

neurosciencenews.com/ai-human-…

in reply to Misuse Case

@MisuseCase @troed @markotway "Deterministic" in the strictly formal sense is not incompatible with "sensitive dependence on initial conditions" (in the chaos theory) sense: it's possible for something to be deterministic and yet remain intractably unpredictable in practice.
in reply to Charlie Stross

@cstross @troed @markotway Now I may be a simple country rooster but IIRC generative AI is *non*-deterministic or at least that’s how it seems to be (even the people who build and maintain these things don’t really understand how/why they work).
in reply to Charlie Stross

@cstross

Exactly. The air brushing against the hair on our arms is also input into the brain so we'll never be predictable.

@MisuseCase @nullagent @markotway

in reply to Troed Sångberg

@troed @MisuseCase @markotway @cstross
There is a large debate about whether free will and determinism are compatible.
In any case, the question whether human cognition is deterministic is different from the question whether human cognition is computationally similar to autocomplete.
in reply to Rob Hughes

@robhughes @troed @MisuseCase @markotway

"Free will" as I understand it is a theological requirement (without it, original sin and redemption/damnation narratives make no sense), not anything we have any evidence for.

As an unbeliever I tend to roll my eyes when people start talking about free will, souls, or heaven and hell.

in reply to Charlie Stross

@cstross @robhughes @troed @MisuseCase There's this fascinating thing where when you measure brain activity it's clear we take action *before* we make the decision to do so.

nature.com/articles/nn.2112

That's of course not entirely correct. What is more precise is that the brain can be shown to exhibit patterns that predict an action before we become aware of our decision.

Which means what we *experience* as freely making choices is an illusion, a story our brain tells itself.

Note...

in reply to nullagent

there wont be a manhunt. There wont be national live news updates. There wont be loads of linkedin posts mourning the loss of a great person.

And everyone will continue to invest in openai.

Had the ceo of openai been found dead thered be a completely different movement in the corporate news media and tech industry

in reply to nullagent

Boost openai suicide mention mh-

Sensitive content