THE COMING AI CONTRADICTION In today's Odd Lots newsletter, I wrote about the gulf between how traders are thinking about AI vs. how Silicon Valley is. At some point, this is going to have to break - ThreadSky

weisenthal.bsky.social • 23 days ago

THE COMING AI CONTRADICTION

In today's Odd Lots newsletter, I wrote about the gulf between how traders are thinking about AI vs. how Silicon Valley is. At some point, this is going to have to break

1 / 2

Comments

konspikuous.bsky.social•23 days ago

I'm not putting money in someone's compulsive greed for more.

They will never stop. Funding those individuals is a mistake. They do not have a vision, what they have is, a play. Remove the investors and see tech firms masks come off.

jamescham.bsky.social•21 days ago

So what is the Joe and Tracy production function? How are you able to be so productive?

jamkc.bsky.social•23 days ago

Does anyone else enjoy planning their own vacations and not want to completely hand that over to an artificial brain?

benbarclay.bsky.social•23 days ago

Exactly. There are real, important use cases, and I'm sure more will develop, but a lot of the ones being suggested are "let's automate something that people actually enjoy doing".

johnnymorton.bsky.social•23 days ago

The advantage I see in travel is that it can produce highly individualized itineraries that would take me way too long. "Get me the cheapest way to South Beach between these dates using cash, Delta points and three Marriott certificates. Oh, and my wife's gluten free so we need restaurants nearby."

benbarclay.bsky.social•23 days ago

Maybe, but there are tools for that kind of thing which are basically algorithms rather than genAI, and I’m sure they’ll become more sophisticated.

And the suggesting was that it would plan the fun bit of your holiday for you, rather than just the logistics. That’s what creates anticipation.

sambarcroft.bsky.social•22 days ago

What will humans do all day?

snoozebuttonben.bsky.social•23 days ago

Option 3: the real AGI is the friends we made along the way

mgorecki.net•23 days ago

Ben, this positive option works aligns perfectly with your photo :) World needs more of both

mariekeblom.bsky.social•23 days ago

To add to the confusion: don't they also say only one country can get through the portal? Still true, with DeepSeek? If (assumed) true, what happens when they get near the portal?

robertdean42.bsky.social•23 days ago

Can’t help but wonder if AGI solution is anti-capitalist; what will the investors do haha! What will the billionaires do if AGI identifies them as toxic to the economy and requiring debridement?

Also seems like there’s some magical thinking that AGI will create a post-scarcity society?

robertdean42.bsky.social•23 days ago

Also is this the plot of Person of Interest?

eliteprogresscheck.bsky.social•23 days ago

Deepseek turned a fair amount of VC firms into Zombies overnight, now we will watch them shamble along as the walking dead for a year or so and then collapse.

12tracks.bsky.social•23 days ago

Things like this do make me wonder about the larger societal implications what work is. Businesses exist now to provide services, however you need money and for most this means a job to function in todays world. The cost of living has priced a people with full employment in my town out of housing.

12tracks.bsky.social•23 days ago

I wonder the greater implications of more work being phased out on peoples livelihoods. It seems like growth in the recent past has been based on extracting more out of the consumers and not necessarily providing better or more service. i.e. RealPage. I don't seem to see much on this angle of AI...

maybeduck.bsky.social•22 days ago

If AI reaches a point where it can do everything, there won't be abundance. AI will keep slowly replacing workers. Production will keep dropping due to lowering demand. That will keep going on until workers are no longer needed, and AI just makes stuff for the rich who own it. Everyone else is out.

lexical-lushes.bsky.social•22 days ago

the investors will basically always have the right of it, because LLMs are fundamentally incapable of becoming AGI.

it's always been a scam to make money. the AGI proponents are selling bullshit, they know that throwing more and more compute at an autocorrect wont create thinking minds.

lexical-lushes.bsky.social•22 days ago

the cult leaders know exactly what they're doing - Altman talked about "rewriting the social contract" as an excuse to try and manufacture consent for abolishing privacy and copyright laws to feed the AI training data maw.

lexical-lushes.bsky.social•22 days ago

AI development is fundamentally disinterested in how minds actually function. if they were taking it seriously they'd be studying and trying to replicate real brains.

closed-infinity.bsky.social•23 days ago

The only question that matters, already 3 years into this: where are the massive layoffs?

benbarclay.bsky.social•23 days ago

Particularly at the companies pushing this, who presumably know how to make best use of it. For the moment, OpenAI's headcount continues to rise rapidly. That may change, but it hasn't yet.

rehnquist.bsky.social•23 days ago

Who will be laid off almost certainly: Low skill admin jobs of which there are millions
Who will possibly be laid off: creative jobs (writers, computer artists, designers), skilled jobs of 'closed end' processing nature - think accountants

Programmers working at AI companies will never be laid off

bzitzow.bsky.social•23 days ago

Are you kidding?

Look at the software industry.

closed-infinity.bsky.social•23 days ago

Unimpressed. This is supposed to be no less than a revolution in productivity. The “last invention” bla bla bla. So far looking more like a glorified clippy

bzitzow.bsky.social•23 days ago

There will be pie in the sky when you die, too.

michaelsmit.bsky.social•23 days ago

"Never the twain shall meet?" Or perhaps they will, as investment and capital deployment decisions of both the financial market traders and Silicon Valley become more and more influenced by AI. Whose goals will those AIs be in alignment with? At what point are we watching the wrong race?

dneto.dev•23 days ago

I think Tom here put his finger on something really important, not covered by the AGI vs. no-AGI framing. That is, if DeepSeek is the start of a massive efficiency breakthrough then the real transformation might be from personalizing AI: the age of the hobbyist.

https://bsky.app/profile/tomforsyth.bsky.social/post/3lgx73aytms2b

dneto.dev•23 days ago

We've seen this before with computers. Arguing over AGI vs. not is like watching the fight between IBM mainframes and DEC, but putting the PC (and the Commodores) into kids hands are where the real transformation/innovation/payoff come from.

Maybe?

juanmirod.bsky.social•22 days ago

The same tech people create the contradictions, in the same hour they can tell you that it will reshape society and that it will fill those boring school forms for you...

captainahab.bsky.social•23 days ago

AGI is a fairytale and we're wasting massive amounts of capital chasing it

various-threnodies.bsky.social•23 days ago

It's almost certainly never going to happen, and certainly not in our lifetimes. This is just the current version of the "singularity". I remember reading articles in like 2000 saying that we'd all have immortal robot bodies by like 2025 thanks to the "singularity". Well, here we are!

captainahab.bsky.social•23 days ago

I would find it very funny that the singularity cultists tricked Wall Street into lighting billions of dollars on fire if I wasn't worried about what all that wasted capex might do to the economy

various-threnodies.bsky.social•23 days ago

Decades of marinating in science fiction media that almost universally features AGI has created massive unconscious biases towards the belief that AGI is an inevitability when in reality it's anything but.

lexical-lushes.bsky.social•22 days ago

there are good reasons to believe AGI (at least in the proper sense of "a computer that can generalize the same way a human can") is possible, because our brains exist.

lexical-lushes.bsky.social•22 days ago

but AI researchers are chasing shortcuts like fancy autocorrect instead of putting the work in to understand and copy the only existing AGI - our own brains.

even a computer with the cognitive abilities of a dog - of a bee! - would be revolutionary in what it could do.

lexical-lushes.bsky.social•22 days ago

but that would require a very different approach than LLMs and these mathematically averaged models.

we bias "sounding intelligible" over actual intelligence, unfortunately.

unclequagsire.bsky.social•23 days ago

Did you not get yours yet?

frankenmizer.bsky.social•23 days ago

I enjoyed the piece - good insights!
Thank you. 🙇‍♂️

oldcoleminer.bsky.social•23 days ago

Read a little lit theory and you quickly realize that AGI via LLM is rubbish at best, a menacing lie at worst. These guys know nothing about language, let alone consciousness, and they’re doing horrible things to make you believe they do.

beckandcall.bsky.social•23 days ago

I read a lot of AI papers bc I build these models as a profession, and the AGI stuff has always seemed absurd to me. Like reading AIAYN you realize that the quadratic nature of self attention means there's a hard cap on the context these systems can have, which probably totally precludes AGI

oldcoleminer.bsky.social•23 days ago

Yeah a bit above my head there but sounds like no attention feedback loop is fixing that.

beckandcall.bsky.social•23 days ago

Basically the attention mechanism at the core of these LLMs is computationally inefficient, in a way that limits the overall system. But all the attempts at making it more efficient result in worse answers and more hallucinations

oldcoleminer.bsky.social•23 days ago

Oh yeah is that cause you’ve got to map every token to every other token?

beckandcall.bsky.social•23 days ago

Essentially, yeah. The attention mechanism needs to be able to conceptually link any two tokens in the input, and that work of token comparison has to be done somehow

noisyconfusion.bsky.social•23 days ago

I haven't heard about AIAYN, thanks for mentioning it

richardblogger.bsky.social•23 days ago

And coding. I've asked some of these models to write code, and it's given back what looks like good code, but with subtle bugs. I reckon (as a tech book writer) it's just taken verbatim from a book

oldcoleminer.bsky.social•23 days ago

For sure. I’m a real-time systems guy and it’ll never be able to write code worthy of high risk situations

richardblogger.bsky.social•23 days ago

I've asked around and it is what coders say. BUT. Those in management think coders will be replaced. My opinion of these models so far (across the board) is they are a reversion to mediocrity

oldcoleminer.bsky.social•23 days ago

I’m actually getting my MBA rn too. The bet for any sane businessperson is that they will make more money deflating or holding steady worker salaries using AI as a bargaining chip than what they invest in AI. It’s all salary suppression.

daniwriting.bsky.social•22 days ago

It’s this. Salary suppression and paying fewer employees to do more jobs with the promise that AI will make it more efficient somehow.

bzitzow.bsky.social•23 days ago

Ghandi famously, and rightly, identifies that these are the sources of violence in society.

bzitzow.bsky.social•23 days ago

Humans are governing and committing all 7 of the Social Sins:

* Politics without principles.
* Wealth without work.
* Pleasure without conscience.
* Knowledge without character.
* Commerce without morality.
* Science without humanity.
* Worship without sacrifice

amimigdalia.bsky.social•22 days ago

Nice!

bzitzow.bsky.social•23 days ago

It's pretty clear to me how this will go down. Here's why:

joycetoadway.bsky.social•23 days ago

"AGI, research what's neccesarry to make a matter replicator that uses any trash matter as an intake and give me the schematics."

"AGI, please create an API for my matter replicator in the previous question that you could use."

"AGI, please print 3 pounds of gold."

Repeat in every other home

lexical-lushes.bsky.social•22 days ago

bwomp, dead from the radiation generated.

we have gold transmuters now! they're called particle accelerators.

hardklaw.bsky.social•23 days ago

This all assumes that AGI ever exists or gets close to what is promised, which anyone can see is highly speculative

enkindler.bsky.social•23 days ago

More than speculative, the type of AGI OpenAI is shooting for is expected to be impossible. This is just one relatively accessible way to view it.

https://www.mdpi.com/1999-4893/13/7/175

markr51.bsky.social•22 days ago

The AGI scenario seems to involve a genocide of the redundant peasants. Nothing else makes sense.

rehnquist.bsky.social•23 days ago

One would need some additional fundamental breakthrough (such as the 'Attention is all you need' paper) for true AGI to happen.
The cutting edge LLMs are nothing short of amazing, but its not quite there yet and you can't (really) get there just by 'fiddling with the dials'.

jonbadly.bsky.social•23 days ago

How far away are we from walking through a portal to a reality where an AI can accurately identify how many “R’s” are in strawberry?

penis.house•22 days ago

i love how you talk at length about AGI but don't explain what those letters stand for

dynamicd.bsky.social•23 days ago

Maybe I'm cynical, but to me Silicon Valley is just traders with a tech focus.

All they really care about is drumming up buzzwords to push stocks up. That's it. In that they are perfectly aligned.

Comments

Posting Rules

Reply