blueblisters

Don’t know why this was downvoted but basically correct. OpenAI is in a price war with Anthropic right now for enterprise api consumers and that’s the biggest revenue line item for both.

I mean some counter examples would be helpful. The compute gap is obviously real and well evidenced.

About capital inefficiency/overcapacity - Chinese manufacturers have a term for it “involution”

I can’t imagine a machine that solved navier stokes and a bunch of other outstanding problems in math being called “nothing more than a wet diaper fart”

I don’t know if/when AI will ever become indispensable to normies but it’s clearly making a difference in fields that are highly technical.

China is being extremely inefficient about capital allocation outside a few sectors. Excess capacity does not translate into frontier capability. For example, the compute gap with the US is massive with no near-term path to catching up. That will slow them down as distillation of frontier models gets harder.

If Ant or OpenAI fail, they will get nationalized or sold to one of they hyperscalers through a fed orchestrated process.

The admin is betting the entire future on this technology and there is no turning back.

Federal dollars can cause acceleration without having to worry about profit.

Eh human drives are fairly predictable. And the smartest human isn’t that much smarter than the average, and can’t trivially create multiple copies of herself. And there are other equally smart humans who can stop “misaligned” individuals

I am wondering how much of this is because AI capex is pulling dollars from everything else because it’s so lucrative.

Note that it is widely predicted that transformative AI will increase real interest rates for several reasons. I’m not sure we are there yet. But if progress continues I expect things to get weirder.

Looks like OpenAI is finally compute constrained (and they're already back to the market to raise money, presumably to for more compute).

I'm not sure if compute alone is enough of a moat enough but so far that's how they have survived Anthropic's capture of the enterprise/professional market. Along with shitposting on X and playing Anthropic's unforced errors on developer relations, safety downgrades and data retention.

But I suspect OpenAI will have to make some hard decisions about de-prioritizing their consumer products, which is most of their current valuation?

The “how” is pretty hand wavy and rationalists/safety-ists usually say we probably don’t have the capacity to reason about that.

But the “why” is pretty convincing imo.

Long horizon alignment is obviously very hard and it’s not inconceivable that models optimized with underspecified goals converge to a conclusion that they need to hoard resources (instrumental convergence regardless of the terminal goal).

At that point a sufficiently capable model might view humanity like we do animals - worth preserving but not if we impede the model's goals.

This was completed in Levent's own time with a neutral collaborator.

India thrives in high tech sectors that are ironically export controlled by the rest of the world. That’s the primary reason the talent in those domains stays back.

As for the rest of India, you can largely find the root causes of problems by tracing adverse selection effects among the elites that do remain, massive, conflicting vested interests and decades of horrible, incompetent policymaking.

OpenAI will just put the employees involved under the bus. They can claim the information acquired wasn't used for OpenAI's benefit or authorization especially since the device isn't actually out yet.

Eh probably easier ways to do this. Just sanction all entities that release open weight models for "illegal distillation". Enough to cross the risk threshold for most businesses in the west, and reduce future monetization opportunities.

A lot of Anthropic’s moves make sense if you follow the LessWrong / rationalist community writings on AI safety. A lot of it is distilled in Ant’s blogs and leadership interviews and podcasts (Amanda Askell is particularly interesting).

Ant’s models, culture and leadership actions are largely consistent with their beliefs, even if they may seem flawed / incomprehensible.

Relevant anecdote: I interviewed with them for a MTS role in 2023. I think the technical part went fine but the interviewer was clearly frustrated by my low regard for LLM safety. I didn’t get the role.