AI personal assistants/agents will become the dominant form factor of AI, surpassing traditional research-oriented chatbots in popularity and usage.
“I think the takeaways for me are number one that this will be the year of the rise of the personal AI assistant... I think this is going to be such a big product that it may become the dominant form factor of AI, meaning more popular than the sort of research-oriented chatbot.”
View on YouTube
Explanation
AI personal assistants and agentic products have grown significantly in popularity, but it is not clearly established that they have overtaken traditional chatbot interfaces as the dominant AI form factor.
Combining next-generation decode-optimized silicon with open-source AI models will cut the overall cost of running AI by about 90%, driving individual usage costs down to roughly $10 a day.
“post Grock what I think will happen is you're going to see an explosion of decode silicon. If you take these next generation systems and you marry them to open source you're going to cut the cost of AI by 90%. And when you do that you know Jason your bill is going to be 10 bucks a day.”
View on YouTube
Explanation
AI costs have fallen substantially with next-generation silicon and open-source models, broadly directionally consistent with the prediction, though the specific 90% cost cut and $10/day figure are not precisely verifiable.
90-95% of AI compute workloads will shift to run locally on stacked commodity hardware (like Mac Studios) rather than the cloud, giving users full control of their own data.
“the M5 chips coming to these Mac studios and if you want more power you stack it up and then the open source models are getting easier. So what's going to happen is 90% or 95% of our jobs are going to go to this local hardware will control it.”
View on YouTube
Explanation
The large majority of AI compute workloads continue to run in the cloud rather than shifting 90-95% to local commodity hardware.
Major AI model makers will launch a revenue-sharing program that pays users to license back their prompt-and-response data, potentially covering the full cost of hosting and running AI models.
“It's a matter of time until the big model makers create an incremental revenue stream for guys like you, Jason, to license back all your prompt and response data. And you'll probably make enough to pay for the all the costs of hosting and running these models.”
View on YouTube
Explanation
No major AI model maker has launched a broad program paying users to license back their prompt-and-response data at the scale predicted.
Architectural advances will cut AI model compute requirements by 70-100x, eventually allowing frontier-class AI models to run locally on iPhones without needing a cloud connection.
“There papers that indicate we could probably go down by 70 to 100x in terms of compute need in how the the model actually runs. So yeah, ultimately these things end up on the on the iPhone running locally and you don't need to go to the cloud.”
View on YouTube
Explanation
Significant efficiency gains in AI model architecture have occurred, but frontier-class AI models running fully locally on iPhones without any cloud connection has not become the norm.
David Sacks
Right
attribution: medium
01:04:59
aipoliticsgovernment
Within a year, the predominant real-world use case of AI will have shifted well beyond niche 'fantasy chatbot' interactions, undercutting AI policy debates framed around that narrow use case.
“that is just not the predominant use of AI. And it's certainly not going to be in a year.”
View on YouTube
Explanation
Within about a year, the predominant real-world uses of AI (coding, productivity, search, agents) had clearly moved well beyond narrow companion-chatbot use cases.
Unknown
Unvalidated
attribution: low
01:06:31
aitech
The share of code written by AI will rise from roughly 50% today toward 99%, reaching a point where essentially no human reviews AI-generated code, significantly raising prompt-injection security risk.
“So as AI coding becomes a bigger and bigger percentage. Right now it's probably what like 50%. As it goes to 99% no one's even checking it. You have to really worry about prompt injections.”
View on YouTube
Despite being viewed as a strong candidate with tech-world backing, Matt Mahan will lose the California governor's race, and Eric Swalwell will end up as the nominee/winner instead.
“I feel like the fact that you're telling me that Ma[han] is this great candidate and, you know, the whole tech world's going to get behind him. I'm sure that's true. And then somehow I think it's the reason he's going to lose and I think, you know, we'll end up with Swalwell”
View on YouTube
Explanation
Matt Mahan did lose the 2026 California governor's race as predicted, but Eric Swalwell also did not win, having resigned from the race after sexual assault allegations before the primary concluded.
California will face a roughly $1 trillion fiscal cliff from unfunded, legally locked-in public pension obligations that can only be resolved via a state constitutional amendment or a new state bankruptcy mechanism.
“California has a trillion dollar fiscal cliff coming up because of the pension obligations... there are two ways that California can be saved. Number one, you pass an amendment to the Constitution to fix this pension liability problem. And number two is the state has the ability to declare bankruptcy.”
View on YouTube
Explanation
California's pension-driven fiscal cliff has not yet triggered a resolving constitutional amendment or bankruptcy mechanism as of 2026.