None of this makes any sense unless there is some super secret versions of these AI that they’re holding back from the public. Nothing about the current models is what they claim it to be. Its a useful tool, sure, but its really only as useful as the skill of the operators, and I don’t mean “prompt engineer” or whatever other bullshit you call that. I mean actually developers and engineers who have had the time in the trenches to know what is going to work or not work.
Use AI to generate a codebase for an app. Get to an MVP. Then, try and make a change. I dare you to. Its a fucking disaster. These models are not up to the task. Its a structural issue. They have no conception of how things fit together. They’re constantly just making shit up whole-cloth, and beyond that, they’re just gaslighting you to sound smarter than they are.
They can be used to make demos. They can be used to solve already solved, well trodden problems. But thats really about it. You’ll spend 4 hours trying to get it to change one basic component. They’re exponentiating the amount of calories burned, but not the amount of work done.
Thats what I think. If they had the super secret “next best thing”, they’d be pumping it for all its worth. The models themselves have felt asymptoted since 4.0. The engineering around those models has improved, but the underlying models haven’t.
There is also a hanging chad around alignment. There was much-ado when I think it was 4.0 or 3.5, around some jailbreaks. Everytime a jailbreak got fixed, the models got worse.
What are the chances that this is orchestrated to bring more computing control to the larger companies who can weather the storm? If end users don’t have ram or storage that will make it even easier to push them into walled garden cloud platforms.
The only hope for consumers is that Chinese manufacturers pick up the slack, but guess which country western governments ABSOLUTELY don’t want you buying products from?
How much would it matter?
Even if they are Chinese manufacturers, they would want to go with the high-price buyer, which is larger companies.
And if they do keep some reserve for people, out of some long-term-thought or maybe just some morals, wouldn’t they just prefer selling inside China?
Edit: and they probably like the AI is flooding the open source market with shitty code and PR that can make it hard to know which projects/PR actually is actually done through human efforts for the betterment of the society
None of this makes any sense unless there is some super secret versions of these AI that they’re holding back from the public. Nothing about the current models is what they claim it to be. Its a useful tool, sure, but its really only as useful as the skill of the operators, and I don’t mean “prompt engineer” or whatever other bullshit you call that. I mean actually developers and engineers who have had the time in the trenches to know what is going to work or not work.
Use AI to generate a codebase for an app. Get to an MVP. Then, try and make a change. I dare you to. Its a fucking disaster. These models are not up to the task. Its a structural issue. They have no conception of how things fit together. They’re constantly just making shit up whole-cloth, and beyond that, they’re just gaslighting you to sound smarter than they are.
They can be used to make demos. They can be used to solve already solved, well trodden problems. But thats really about it. You’ll spend 4 hours trying to get it to change one basic component. They’re exponentiating the amount of calories burned, but not the amount of work done.
The hardware bought by a company can’t be bought and used by the competitor.
And “computing power” is… power.
It’s a grift, nothing more. The execs know this, but they want new yachts.
Thats what I think. If they had the super secret “next best thing”, they’d be pumping it for all its worth. The models themselves have felt asymptoted since 4.0. The engineering around those models has improved, but the underlying models haven’t.
There is also a hanging chad around alignment. There was much-ado when I think it was 4.0 or 3.5, around some jailbreaks. Everytime a jailbreak got fixed, the models got worse.
What are the chances that this is orchestrated to bring more computing control to the larger companies who can weather the storm? If end users don’t have ram or storage that will make it even easier to push them into walled garden cloud platforms.
The only hope for consumers is that Chinese manufacturers pick up the slack, but guess which country western governments ABSOLUTELY don’t want you buying products from?
How much would it matter?
Even if they are Chinese manufacturers, they would want to go with the high-price buyer, which is larger companies.
And if they do keep some reserve for people, out of some long-term-thought or maybe just some morals, wouldn’t they just prefer selling inside China?
Starting to look like that honestly
Edit: and they probably like the AI is flooding the open source market with shitty code and PR that can make it hard to know which projects/PR actually is actually done through human efforts for the betterment of the society
Yeah, they went all in on sabotaging the competition. Using their freely given product to destroy them.
You are correct, and now the emerging strategy is to just “re-roll” the entire codebase any time a change is needed. It’s insane.