jsd

@EpochAIResearch. My DMs are open.

Anonymous feedback:

[jsd @datagenproc]

Hi all, I'm interested in feedback!
You can leave anonymous comments here: https://t.co/9IS5haHbFh

[jsd @datagenproc]

reminder
https://t.co/EUKRcA0zeL

[jsd @datagenproc]

@AaronBergman18 I also like "FU community" from @Scholars_Stage

https://t.co/dn3vVsIs5L

[jsd @datagenproc]

@ayedtay @distributionat Definitely it's a very noisy signal.

[jsd @datagenproc]

@ayedtay @distributionat eg https://t.co/EdhRpiBmmO or https://t.co/B3NUd9dbJy

[jsd @datagenproc]

@ayedtay @distributionat Also I'd expect job listings to be more representative for junior roles compared to senior roles.

[jsd @datagenproc]

@ayedtay @distributionat I think lack of job listings doesn't mean lack of hiring for a role. But if you have job listings for a role it does mean you're hiring for it?

[jsd @datagenproc]

> It could probably also find a way to experiment on long-term effects of drugs with a quicker feedback loop, for example via high-fidelity simulations of human brains and bodily processes

It's not clear to me how you go from intelligence to high-fidelity simulations. Wouldn't you need more serial experimentation to make your simulations good?

[jsd @datagenproc]

@distributionat There’s a couple SWE listings that mention 4+ yoe

[jsd @datagenproc]

@distributionat Hmmm they have job listings for Senior, Staff+, Staff SWE.

But even Staff SWE asks for 7+ years of experience.

[jsd @datagenproc]

@sebkrier This is now the Epoch Capabilities Index! https://t.co/ESwXbh6uVK

[jsd @datagenproc]

@Citrit_ @AvitalBalwit @dwarkesh_sp @nanransohoff they are in favor of prop 12 which SOB would overturn https://t.co/SoHrh3QJTw

[jsd @datagenproc]

@zhengyaojiang @testingham icymi another data point

[jsd @datagenproc]

@norvid_studies @pleometric will you be at lessonline norvid

[jsd @datagenproc]

I think a big part of it is not as much access to high quality data. (Unclear how much access to usage data plays a role but my guess is, for now not a huge deal for raw capabilities but matters a lot for having a good product)

More speculatively, sometimes the incentive landscape seems a bit different for open models, you're maybe less driven to make a model that's really good and directly making money and more driven to make a model that appears 'close to the frontier'.

[jsd @datagenproc]

Enjoyed this podcast a lot.

Quick clarification on Rosetta stone (≈ ECI)

① @rohinmshah says the ECI is mostly linear over time. I disagree with this, cf @AlexBarry4 and my recent analysis in "Have AI Capabilities Accelerated?" https://t.co/YHnbOl39bR. (Note that this analysis is from April 2026, while the podcast was recorded in December 2025)

② I agree with something like "the ECI trend is overall very smooth compared to what you might expect from a statistical model that does not encode time information at all".

③ I think that this acceleration in ECI has come largely from increasing correlation between benchmarks and tasks that AIs are trained on directly. If we were looking at a broader set of tasks, including harder to benchmark tasks, we'd likely see less (or no) acceleration.

[Rob Wiblin @robertwiblin]

My best interview in some time.

Rohin Shah leads AGI alignment/safety at DeepMind.

And he has a lot of spicy personal takes:

We probably won’t get catastrophic misalignment (00:49)
Safety 'commitments' have severe limitations (10:38)
The intelligence explosion probably isn't imminent (1:52:44)
Why he's not working to pause AI advances (51:44)
Pre-deployment evals aren't the right focus (for catastrophic risks) (37:41)
Signalling concern for safety sometimes diverts resources from actually making AI safe (01:09:51)
Reading AI thoughts is v useful for safety – and we'll probably be able to for years to come (54:17)
Governance is somewhat more likely to be the bottleneck than alignment (43:55)
Rohin's team doesn't have a veto, and that's OK (27:36)
Central banks are a promising model for regulating AI (33:34)

Also:

Google DeepMind's actual plan for building AGI safely (1:40:29)
How external researchers can positively influence big AI companies (2:21:55)
The roles GDM most needs to hire for (2:37:03)

On the 80,000 Hours Podcast. Links below - enjoy! (@rohinmshah)

[jsd @datagenproc]

@ArthurConmy @AsaCoopStick Interesting, I'd heard that Claude mentions Redwood and Buck a lot!

[Rob Wiblin @robertwiblin]

My best interview in some time.

Rohin Shah leads AGI alignment/safety at DeepMind.

And he has a lot of spicy personal takes:

We probably won’t get catastrophic misalignment (00:49)
Safety 'commitments' have severe limitations (10:38)
The intelligence explosion probably isn't imminent (1:52:44)
Why he's not working to pause AI advances (51:44)
Pre-deployment evals aren't the right focus (for catastrophic risks) (37:41)
Signalling concern for safety sometimes diverts resources from actually making AI safe (01:09:51)
Reading AI thoughts is v useful for safety – and we'll probably be able to for years to come (54:17)
Governance is somewhat more likely to be the bottleneck than alignment (43:55)
Rohin's team doesn't have a veto, and that's OK (27:36)
Central banks are a promising model for regulating AI (33:34)

Also:

Google DeepMind's actual plan for building AGI safely (1:40:29)
How external researchers can positively influence big AI companies (2:21:55)
The roles GDM most needs to hire for (2:37:03)

On the 80,000 Hours Podcast. Links below - enjoy! (@rohinmshah)