"we could find a nail for this hammer... or we could just hammer everything until we find the nail or make too much money hammering for a bit then sell in may and go away".
My "oh sh* moment" with GenAI is ongoing and is watching all the correlated financials unwind when TSMC said "we can only support so much"[1].
Very few things in life experience exponential growth and assembled systems don't often stay that way if they don't become sigmoidal. ie its exponential and end is nigh xor its exponential then sigmoidal xor linear.
Also a bit hilarious to believe that a single 3 month cycle at TSMC could determine a severe amount of propped valutions.
But it would still be very ignorant to just not follow it very very closely and take it serious if something throws a lot of signals like AI and Robotics does.
We are in 2026, supply chains are highly optimized which means we could make and deliver a lot of robots in a relative short period of time.
We also have solved all fundamental issues we had 20 years ago like communication thanks to the internet, translation and co.
We know have a system, which can be copy and pasted and run in parallel with a snap of a finger.
If (and this If is not that crazy) some breakthrough happens tomorrow, this can be used the day after tomorrow.
I currently say that it could become very very interesting in 5-15 years. I still follow AI very closely and i do not have the feeling anything is slowing down.
And independent of something happening, a lot of people did not find jobs due to this AI investment, a lot of jobs already disappeared too.
Long-term is not what ppl are arguing for and that's not what the stock multiples imply.
Companies like Nvidia, up 1000% since 2023, clearly cannot rely on "oh you'd get returns in 5-15 years".
They (Jensen and others) are arguing for AGI in a year (he said this at a talk at Stanford ~ 1yr ago).
Long-term, ie 5-15 years there will be many technologies that change the world. Some will come from transformative tech, others will come from other places.
Financial instruments, investors etc have timelines.
Blowing timelines can risk the future if not properly aligned, especially in correlated risk. See the internet bubble, where it was clearly important but took 15 years to recover after the 2000 boom/bust.
In general, longevity risk can include the financial instrument timelines, it can include other technical factors about the technology, and it can also be that the boom serves important people too soon with gains, real limits are imposed, and then the whole system resets until the real limits are overcome or resolved.
While Vivado/Vitis etc do amazing things, I challenge anyone to find a person who enjoys using them without TCL interfaces.
These tools do need attention, it's too bad there's not a better model than subscription bases like these.
Pretty sure, based on TCL base, that these tools were native Unix at some point, so the no-linux-free-beer vs windows-free-beer version are hilarious...
Ultimately one has, with so many vendor tools, a windows box somewhere so make it a remote compile machine.
Seems senators had questions about why CISA was scaling back efforts related to election security[1]. Tulsi's resignation today seems interestingly timed to when this became public.
I don't know why US senators are up in arms about this. Trump was extremely clear when he gave them his budget that he wanted CISA's budget drastically cut. He also specifically directed CISA to shut down their election security office.
This is the "who killed Hannibal" meme. If Padilla and Warner didn't know about this, then they're incompetent themselves. Especially because they reported on it last year:
It takes brains to run propaganda that successfully changes minds.
Propaganda that just confirms preexisting mass delusions is actually pretty easy to run if you have a lot of support from similar actors running adjacent campaigns.
reciprocal tariffs had put the non-tech and tech economy in stasis (except for hardware for AI). they are also are better than tax breaks and will supercharge bottom lines for large corporations once reclaimed and if prices remain high.
also if you want to test/force ai adoption you have to put pressure by firing some
now wars will put us into further stasis or decline via increased inflation pressure.
people have been talking about "models of models" for arbitration opportunity in inference for about 1.5 yrs.
Arbitration idea: if a user doesn't need high QOS of newest LLM, slip them a cheaper LLM, run their query at reduced quality. measure if they cost you fewer $s in the lower QOS. => profit.
For chatgpt the arbitration opportunity looks more like "we could allocate this amount of gpu to training or inference, we are losing money if we offer the highest quality infra"
In addition there's other interesting economics scaling that can be done outside of "models of models" that are far more profitable. I won't go over all of them (and some of them I feel are quite powerful) but the laziest one is that subscription models count on some zombie users as a counterweight to highly expensive single users, and as a source of stable cashflow.
Zombie users are ones that are paying for sub but not actively or barely using the service
They made a big point of explicitly advertising this as a feature with the GPT-5 rollout, no? Routing to cheaper models/less reasoning depending on the input prompt.
"we could find a nail for this hammer... or we could just hammer everything until we find the nail or make too much money hammering for a bit then sell in may and go away".
reply