Miles Brundage

@Miles_Brundage on X

Writes about AI. 95% of posts open with an observation. A typical post reaches 242 likes.

Remix their style

Posts: 20
Likes: 7.8K
Typical: 242
Best: 1.2K

What they write about

AI: 90%

How they open

observation: 95%
number: 5%

Every post of theirs in the library

Miles Brundage@Miles_Brundage

In light of what external lawyers and the Pentagon are saying, OpenAI employees’ default assumption here should unfortunately be that OpenAI caved + framed it as not caving, and screwed Anthropic while framing it as helping them. Hope that is wrong + they get evidence otherwise

Opens with an observation

Miles Brundage@Miles_Brundage

Pretty obvious to anyone paying close attention that 1. That would be a mistake from a national security perspective 2. There is a coordinated effort to take down Anthropic for a combination of anti competitive and ideological reasons

Opens with an observation

Miles Brundage@Miles_Brundage

Very fortunate for OpenAI that the victims of their accidental autonomous cyberattack were very chill about it!!! Also, reminder that there are no minimum safety or security standards for frontier AI (just light transparency reqs), and no auditing requirement until 2028 (!).

Opens with an observation

Miles Brundage@Miles_Brundage

If I worked at OpenAI, I would want to see the contract details, or have a critical mass of employees from a range of teams/perspectives see the contract details. What has been reported so far is confusing + the details matter. t.co

Opens with an observation

Miles Brundage@Miles_Brundage

I've been saying recently that completely superhuman AI math and coding by end of 2025 was plausible - 50/50 or so. Now I'd say it's much more likely than not (o3 is already better than almost all humans).

Opens with an observation

Miles Brundage@Miles_Brundage

Again, agree Dwarkesh did a good job but I want to add one nuance - most journalists covering AI couldn't do that interview even if they wanted to be "tough," because they don't have enough knowledge to confidently disagree with Jensen's pushback t.co

Opens with an observation

Miles Brundage@Miles_Brundage

This is not an AI pause. It's the government thrashing around after realizing that the people who told them AI was NBD were wrong, and being severely understaffed. AI development is not even close to stopped RN + this is not what you'd do if you wanted it to stop

Opens with an observation

Miles Brundage@Miles_Brundage

I don't love the term AGI but o3 is crazy, o3 pro will be crazier, o5-pro will be super duper crazy, etc. There is no single right definition/threshold for "AI that matters a lot." But for all reasonable ones, we've exceeded it or will exceed it in the next few years.

Opens with an observation

Miles Brundage@Miles_Brundage

o1 being available via API may not be as sexy as video stuff but I predict that this will in retrospect be seen as the beginning of a significant increase in AI's impact on the economy. Many more use cases are now possible.

Opens with an observation

Miles Brundage@Miles_Brundage

Few things are clearer evidence for Moravec's paradox than a small model (o3-mini) just crushing most humans on many things humans think are super hard reasoning problems... ...while lacking a lot of real world knowledge/commonsense + not being able to do any sensorimotor stuff

Opens with an observation

Miles Brundage@Miles_Brundage

The bulk of this post is good + I applaud the folks who work on the substantive work it discusses. But I'm pretty annoyed/concerned by the "AGI in many steps rather than one giant leap" section, which rewrites the history of GPT-2 in a concerning way. t.co

Opens with an observation

Miles Brundage@Miles_Brundage

I am a huge TBPN fan + was deeply honored to join the show a while back... and I'm sure they have good reasons for this/got a good deal. But from an OAI POV, I also think this confirms the Sora thing was a one-off amidst a compute crunch, not a real sign of greater focus

Opens with an observation

Miles Brundage@Miles_Brundage

OpenAI in particular should be defending Anthropic here given their Charter: “We commit to use any influence we obtain over AGI’s deployment to ensure it is used for the benefit of all, and to avoid enabling uses of AI or AGI that harm humanity or unduly concentrate power.”

Opens with an observation

Miles Brundage@Miles_Brundage

I doubt any OpenAI executives signed off on this, but that doesn't excuse it - OpenAI is obviously doing too much + losing focus. If you can't keep track of all your lawsuits, don't launch browsers and social media apps and (etc.). t.co

Opens with an observation

Miles Brundage@Miles_Brundage

It is both true that Ant made this outcome more likely with the memo and that that doesn't justify the supply chain risk designation whatsoever. Nothing does - it's totally insane. Horrible day for law and order, American AI competitiveness, national security, etc.

Opens with an observation

Miles Brundage@Miles_Brundage

Remember when the whole schtick of xAI was that Grok would be super truthful And no evidence of that was provided And then they set a maximum lying threshold in their safety policies, exceeded it, and shipped anyway And constantly adjusted it to fit El*n’s views Fun stuff

Opens with an observation

Miles Brundage@Miles_Brundage

I think Ilya really cares about safety, but also superintelligence is dangerous regardless of who builds it And as a reminder, a lot of AI regulation on the books + in draft form does not cover SSI, because of revenue requirements t.co

Opens with an observation

Miles Brundage@Miles_Brundage

Shout out to Bores-heads who fought hard in the NY-12 primary Winning would’ve been better but it was close against a governor/Bloomberg-backed candidate with a Kennedy dude splitting the vote It’s clear that being the pro AI regulation candidate helped, which is a key takeaway

Opens with a number

Miles Brundage@Miles_Brundage

One of the reasons I decided to go all in on frontier AI auditing after leaving OpenAI is so that, if AI companies need to take costly steps on safety/security, they can be reassured (by independent auditors) that their peers are doing the same thing. t.co

Opens with an observation

Miles Brundage@Miles_Brundage

Cool work + I'm glad Anthropic open sourced it, but I wish they'd stop calling this kind of thing "auditing." It's a black box evaluation of one LM by another with very arbitrary scoring. Useful, but no need to make it sound more rigorous than it is. t.co

Opens with an observation