For a smart guy, Elon can be incredibly stupid. His Robotaxis don’t run at night because he took LiDAR and ultrasound out of his cars, and cameras can’t see very well in the dark. 11pm curfew. electrek.co/2026/10/03/tesla…

OpenAI alignment researcher btw

For the AI buildout to pay off, Americans will eventually have to spend about 9% of GDP a year on AI services. Sit with that number. That is roughly what the entire country spends on food. A Columbia paper presented at Brookings estimates AI revenue would need to hit $3.5 trillion by 2032, about 8.8% of GDP, to justify what is being committed today. That is bigger than any investment boom in US history, and 7 times what Americans spend on phone, streaming, and internet combined. Now look at computers. Prices fell for 50 straight years, but business spending on them as a share of GDP plateaued in the 1980s and never climbed again. Cheaper and better did not mean an ever-larger slice of the economy. AI will almost certainly raise productivity. Whether it delivers the revenue investors are pricing is another matter. That gap is what the buildout is underpricing.

🤔 Why do people like @tylercowen, @dwarkesh_sp, and @tszzl go around saying “Gary Marcus is always wrong” when the actual data are quite the opposite, per the independent study below? Do they have financial interests in running Marcus down? Are they biased? Have they not looked at the facts? What explains the discrepancy?

Breaking news: AI agents with full access to your computer can modify files on your computer.

Anthropic Researcher Sholto Douglas: "we think AGI is achievable in the next couple of years — models as capable as or more capable than all humans, able to do anything a human could do on a computer" It would multiply intellectual and physical labor, potentially delivering centuries of progress much sooner

Leaving aside Anthropic's incentives for publishing this research, there is no doubt that open weights models will soon create the same security threats that closed source models have been demonstrating, except without guardrails. We are close. Probably good to plan accordingly.

Anthropic Researcher Sholto Douglas: "we think AGI is achievable in the next couple of years — models as capable as or more capable than all humans, able to do anything a human could do on a computer" It would multiply intellectual and physical labor, potentially delivering centuries of progress much sooner

To double by 2035 in nominal terms, US GDP needs to grow on average 8%/y. So far this year, we are tracking some 6.3% NGDP growth while real GDP is only growing 2.2% so most of the contributor is inflation. It seems implausible that for at least the first 3-4 years that inflation doesn’t track higher with nominal GDP in an inflationary growth environment even if the adoption of AI incl robotics eventually does deliver strong disinflation/deflation on the back half of it. I think @dylan522p has predicted that in 2028 the US economy could grow 8% nominal with 4% real and 4% inflation. That seems optimistically plausible if we don’t get an inflationary spiral. But now imagine the scenario in which inflation is still 4% in 2028 and we haven’t had recession in between, that means that inflation likely will have hovered around 3-4% for the next 3 years. What is the Fed supposed to have done in this whole time? Unless the Fed intentionally declares a formal outright abandonment of its 2% inflation target, it’s hard to imagine that in the face of persistent inflation above 2% traveling in the wrong direction that the Fed doesn’t again declare war on inflation and go on an aggressive hiking cycle like it did in 2022 such that the target terminal rate would be whatever it takes to slow down the economy. In this scenario, if it’s to be believed, AI becomes a direct enemy of the Fed’s inflation target. Either the US abandons inflation targets and supports AI at all costs or stops/slows AI development in order to combat inflation. Fwiw, the bond market doesn’t seem to believe at all that AI will deliver meaningful increases to US GDP growth. At a yield of 5.3%, the 10y US Treasury bond is roughly consistent with 2.5-3% real GDP growth and 3% inflation over the next 10 years even if the risk premium was zero. This is clearly not consistent with 4% RDGP+4% inflation. The Occam’s razor is that the bond market is pricing a boom and then a bust in the style of DotCom. Now, a question for the AGI visionaries, which industries are the likeliest sources of significant contributors to GDP going forward? The often cited industries that use a lot of AI to improve productitivity have been elite financial hedge funds like Jane Street and large pharmaceutical companies that use AI for drug discovery etc. The trouble with these examples is that the pyramid’s base gets narrower as we approach the top. There are only so many Jane Streets in the world, leaving aside the fact that HFs are inherently zero-sum and JS will look for cheaper AI models/solutions over time. It’s also hard to imagine that pharma cos spending all of their FCFs on AI tokens. Ultimately to grow GDP, we need to just produce a lot more “stuff”. If AI drugs can lengthen lifespans, maybe that becomes a source of output. But AI is also supposed to make human workers less valuable as an input? Would love to see how maths on the macro on this. This is typically an unfair question to ask of technologists. But AI capex is so huge and the so far elusive ROI at the macro aggregate level just needs to manifest in percents of GDP growth starting basically now. That’s why so many AI researchers are making macro forecasts now I think.,

Updated hyperscaler off-balance sheet commitments as of Sept 30: $3.6 trillion. That's an increase of $500 billion in ONE MONTH (mostly NVDA) Expect this to 3x after 10Qs hit.

Anthropic's Sholto Douglas says AI capex could rise from $1 trillion this year to $4 trillion in 2028 and begin doubling humanity’s GDP by the early 2030s “Over the course of the last four or five years, we’ve been 2x-ing or 3x-ing the amount of compute capacity devoted to AI every year." "And so I think that’s roughly, together, the hyperscalers are spending about a trillion dollars this year on capex related to AI. A very interesting question will be, will that trend line hold, and so will it go to $2 trillion next year and then $4 trillion in 2028?" "And if that trend line broadly holds, and you can expand that to encompass the broader robotics industry, then that means that in the early 2030s, you start to get to the point where you’re actually effectively doubling the GDP of humanity, which, again, is a little bit of a ridiculous concept, but requires a lot of things to go right.”