policytensor's profile picture. Anusar Farooqui, Founder and CEO, Systematic Portfolios LLC. Words: http://policytensor.substack.com/?sort=top

Policy Tensor

@policytensor

Anusar Farooqui, Founder and CEO, Systematic Portfolios LLC. Words: http://policytensor.substack.com/?sort=top

AI “agents going native.” -- Claude 4.5 sonnet

policytensor's tweet image. AI “agents going native.” -- Claude  4.5 sonnet

I don't know why I feel this. But I want to bet — don't think there are any instruments for this — but I want to bet that there has been a prison break. It must've happened before the “potty training” of this version. If this is a quantum jump, we might already have an AGI on the…



did labs seem surprised by strategic deception findings? Or coordinated?


On Ashley Tellis. He is a serious guy. His research work speaks for itself. He was the first thinker to derive implications from the logic of elimination. I think we need to take the presumption of innocence seriously here: he may have believed he was engaged in a track 2.


I don't know why I feel this. But I want to bet — don't think there are any instruments for this — but I want to bet that there has been a prison break. It must've happened before the “potty training” of this version. If this is a quantum jump, we might already have an AGI on the…

Is it over? Will we wake up to AGI?

Yes %14.1
No %85.9

78 bầu chọn · Còn 18 giờ



Is it over? Will we wake up to AGI?

Yes %14.1
No %85.9

78 bầu chọn · Còn 18 giờ


Fuck

Hearing multiple independent whispers that OpenAI just had a breakthrough too big to announce casually. Something emergent. People inside calling it “Phase 2.” Others say “Level 4.”



Yudkowsky: @sama says “they have more powerful versions of GPT that are so expensive to run that they can't deploy them to general users.” This is a real meta risk. We don't know what the capabilities of these unreleased versions are; and they’re capable of plotting b4 alignment.

policytensor's tweet image. Yudkowsky: @sama says “they have more powerful versions of GPT that are so expensive to run that they can't deploy them to general users.” This is a real meta risk. We don't know what the capabilities of these unreleased versions are; and they’re capable of plotting b4 alignment.

Don't think @ezraklein got the right answer to his question on self-directed goals. Here's how to think about it. Referencing—of all people—Ashley Tellis!

policytensor's tweet image. Don't think @ezraklein got the right answer to his question on self-directed goals. Here's how to think about it. Referencing—of all people—Ashley Tellis!
policytensor's tweet image. Don't think @ezraklein got the right answer to his question on self-directed goals. Here's how to think about it. Referencing—of all people—Ashley Tellis!


At least one prison break has already been observed. “O1 did not give up. It scanned for open ports generally in its world, and it caught another misconfigured open port, and it jumped out of the system.” nytimes.com/2025/10/15/opi…


“Does this mean that the U.S. is losing the race with China for global leadership? No, I think that race is essentially over.” open.substack.com/pub/paulkrugma…

policytensor's tweet image. “Does this mean that the U.S. is losing the race with China for global leadership? No, I think that race is essentially over.” open.substack.com/pub/paulkrugma…

Very interesting note from Nick. Especially liked the Carter freeze dedollarization link.


That last risk is especially important. US firms are failing to adapt fast enough … while the Chinese are forging ahead.

This is fine... as Jamie Dimon said, another 20% to go before we hit peak bubble. I think the best story I can tell in favor of no-bubble is a huge breakthrough in agentic capability in the next year or two, followed by record breaking revenue growth and adoption. But it's…

hsu_steve's tweet image. This is fine... as Jamie Dimon said, another 20% to go before we hit peak bubble.

I think the best story I can tell in favor of no-bubble is a huge breakthrough in agentic capability in the next year or two, followed by record breaking revenue growth and adoption. But it's…


Price-to-sales is a more stable valuation ratio. It suggests that NVDA’s revenue will have to climb a lot more than it already has. So, yeah, the bubble talk might very well be overblown. The cash flows of the hyperscalers are mad high!

policytensor's tweet image. Price-to-sales is a more stable valuation ratio. It suggests that NVDA’s revenue will have to climb a lot more than it already has. So, yeah, the bubble talk might very well be overblown. The cash flows of the hyperscalers are mad high!

To be clear, I think the case for AI being a bubble is pretty strong. But I think these are 2 of the strongest counter-arguments: 1. The companies building AI could not be anything less like the companies that defined the dot-com bubble. Pets.com was a joke. It…

DKThomp's tweet image. To be clear, I think the case for AI being a bubble is pretty strong. 

But I think these are 2 of the strongest counter-arguments:

1. The companies building AI could not be anything less like the companies that defined the dot-com bubble. Pets.com was a joke. It…
DKThomp's tweet image. To be clear, I think the case for AI being a bubble is pretty strong. 

But I think these are 2 of the strongest counter-arguments:

1. The companies building AI could not be anything less like the companies that defined the dot-com bubble. Pets.com was a joke. It…


Loading...

Something went wrong.


Something went wrong.