humans built machines that talk to us like people do and everyone acts like this is normal now. it's pretty nuts

the reason that science turned into slop in the second half of the 20th century is that it became a model trained mostly on its own output

I agree with this in principle and overall. On the ground, at the moment, I see the following downsides, and the need to substantiate the high level principles. - It seems to me that there is a significant lack of agreement on what exactly a critical dangerous capability is

Someone should publish a safety paper that flips all the experimental results exactly around, watch to see who details how it confirms their previous views, then do the reveal.

Daron Acemoglu "Whether near-term AGI is an achievable goal remains an open question" Are your estimates on jobs (5% heavily impacted) and productivity (0.06% yearly TFP increase) this decade conditional on AGI not being achieved, or are they taking the possibility into account?

in case you're wondering how it's going

Benchmarks saturate quickly, but don’t translate well to real-world impact. *Something* is going up very fast, but not clear what it means. Thus the wide range of expert opinion, from “superintelligence in a few years”, to “we’ve already hit a wall”. Our results shed some light:

Frontier AI safety frameworks have emerged as a critical tool for managing potential risks to public safety and security. In a series of technical reports over the coming months, the Frontier Model Forum will examine how these frameworks can be implemented effectively.

AI governance is often more like improv comedy than I expected: don't try to come up with a clever idea, instead just focus on the present moment and execute well on the just-next open thing: build taxonomies / frameworks, substantiate each aspect, repeat 🔁

pretraining is an elegant science, done by mathematicians who sit in cold rooms writing optimization theory on blackboards, engineers with total absorb of distributed systems of titanic scale posttraining is hair raising cowboy research where people drinking a lot of diet coke

Is the one law firm undereliciting or the other one not doing sufficient quality assurance? 🙇‍♀️🤷‍♀️

I’m excited to share the first part of an absolutely stunning analysis from the GPT-5 thinking model! I uploaded a huge spreadsheet, nearly 1,300 metabolites (lipids, carbohydrates, microbiome-derived compounds, and much more) measured in 150 ME/CFS patients and 100 healthy

BRILLIANT Google DeepMind research. Even the best embeddings cannot represent all possible query-document combinations, which means some answers are mathematically impossible to recover. Reveals a sharp truth, embedding models can only capture so many pairings, and beyond that,

The "you can just do things" ethos creates an adverse selection dynamic: while ostensibly democratizing agency, it primarily empowers those with minimal internal friction around action - precisely those least equipped to consider externalities or second-order effects. The

Should AI regulations be based on training compute? As training pipelines become more complex, they could undermine compute-based AI policies. In a new piece with Google DeepMind’s AI Policy Perspectives team, we explain why. 🧵

it's called encoder-decoder for a reason