It is a gamble for sure against innovation and a blind one too. I say this as it is clear right now that scaling up LLMs while very effective at substantially improving many AI metrics, it really did not have much impact on logic. I have been calling this the Cognitive Gap and it is really holding back AI.
Clearly the big LLM companies do not have a solution to this gap despite efforts like the reasoning models and that likely means we need an entirely different tech to front end LLMs or replace them.
This begs the question…who has a line of sight on how to scale up logic and the answer as near as I can tell is no one right now. Maybe there is something in a lab somewhere, or even with just a small team or individual, but it is not presently visible. It could come out any day now and make all those Data Center investments worthwhile or may take years before we see the Cognitive Gap close which will really make those same Data Centers completely out of alignment with the value they bring.
Shorting the AI industry is a roll of the dice, but less so than the blind investments still happening in Data Centres despite no clear path to improve logic and close the Cognitive Gap. In fact shorting seems like the safer bet.
Going to be interesting as if the Cognative Gap is not closed for years to come, those Data Center investments are never going to pay off as the value will just not be there. The entire USA economy is tied to AI it seems right now so the roll of the dice is perhaps the biggest risk / reward in history.
And even if they solve some problems with AI and make them smarter, they still have to solve the “actually making a profit” problem to justify these share prices. LLMs already have some use at their current level, but certainly not for the price they’d need to charge to break even, let alone actually making a profit. If they double the smarts but double the training and/or inference cost, they’ll still end up in the same place.
I read a stat that if the true costs of AI was past onto users it would 42 times more expensive or something along those lines. Who know the right number, but it is obviously out of alignment with the value AI is able to bring today. Imagine instead of $20 / month it was $840 / month to just break even.
Funny thing is even that number would be based on all the people who can pay 20 but not 800. If it all has to be balanced by the users it would probably be an order of magnitude… Or the market would play like it logically should and it would be like 3 companies paying a million a month. But what would they do with it? Sell it at a loss for like 20 per user and hope they figure out some way to make money :)
It is a gamble for sure against innovation and a blind one too. I say this as it is clear right now that scaling up LLMs while very effective at substantially improving many AI metrics, it really did not have much impact on logic. I have been calling this the Cognitive Gap and it is really holding back AI.
Clearly the big LLM companies do not have a solution to this gap despite efforts like the reasoning models and that likely means we need an entirely different tech to front end LLMs or replace them.
This begs the question…who has a line of sight on how to scale up logic and the answer as near as I can tell is no one right now. Maybe there is something in a lab somewhere, or even with just a small team or individual, but it is not presently visible. It could come out any day now and make all those Data Center investments worthwhile or may take years before we see the Cognitive Gap close which will really make those same Data Centers completely out of alignment with the value they bring.
Shorting the AI industry is a roll of the dice, but less so than the blind investments still happening in Data Centres despite no clear path to improve logic and close the Cognitive Gap. In fact shorting seems like the safer bet.
Going to be interesting as if the Cognative Gap is not closed for years to come, those Data Center investments are never going to pay off as the value will just not be there. The entire USA economy is tied to AI it seems right now so the roll of the dice is perhaps the biggest risk / reward in history.
And even if they solve some problems with AI and make them smarter, they still have to solve the “actually making a profit” problem to justify these share prices. LLMs already have some use at their current level, but certainly not for the price they’d need to charge to break even, let alone actually making a profit. If they double the smarts but double the training and/or inference cost, they’ll still end up in the same place.
I read a stat that if the true costs of AI was past onto users it would 42 times more expensive or something along those lines. Who know the right number, but it is obviously out of alignment with the value AI is able to bring today. Imagine instead of $20 / month it was $840 / month to just break even.
Funny thing is even that number would be based on all the people who can pay 20 but not 800. If it all has to be balanced by the users it would probably be an order of magnitude… Or the market would play like it logically should and it would be like 3 companies paying a million a month. But what would they do with it? Sell it at a loss for like 20 per user and hope they figure out some way to make money :)