All of this low purchasing power/tough times hitting newer generations leads to issues related to housing when the real objective is to have you have less children or simply none.
Things being tough means you won't have a family. By design. Housing is 1 of the means to that end.
We are thought sampling machines and if we're not honest, we will never be able to find and follow signal.
The following is the ultimate information theory/geometry gift:
A group beats its best member when three things hold:
-members carry independent information about a shared target;
-they express it in a shared convention;
-they are selected or weighted by quality.
The main labs give us free tokens (subsidies) and get access to your data.
They then use that data pool to make the model better and obviously to implement also good product ideas (subset of data that was proprietary alpha).
Finally, you use their better models to build AI infrastructure/projects that will represent 99.9% of token usage in the world using open sourced models.
In the limit, their job is to make sure they get well payed for the 0.1%.
The limit = 2028
There's a lot of important things that will come out from heuristic optimization methods we known for decades about.
Powered by the same speed and scale that brought the DL revolution - GPUs.
In its current form, AI has reached 90+% of its TAM.
It's TAM is 5-10% of people. Maybe some people live in delusion but at least 90+% gain absolutely nothing from paying for the cheapest AI tier provided by any company.
Now this is not to say that the current market wont need to scale, because the current users still need at least 100x more inference for their current and future needs.
Good labs are like good hedge funds, they stick to doing the basics at a high level while every other shop claims a new alpha appeared just to get smoked weeks after.