The story’s different when it comes to specialist biodesign tools – going back to our DNA synthesis scenario at the top. But the regulatory approaches that may work for big frontier labs (e.g. compute thresholds) don’t translate so well for these. The scale/capability
@nathanbenaich
-
Governing AI Models: Frameworks for Responsible Development
By
–
We’re seeing work like this from @tessafyi
, @mslkmp
, @samuelmcurtis and co. setting out frameworks for responsibly governing BDTs. These are eminently sensible and industry should likely adopt them, but there are lingering questions about how to tackle de novo threats -
LLM Biorisk Research Lacks Central Question Demonstrations
By
–
There are certain central questions LLM-focused biorisk research needs to demonstrate to us, but we haven’t seen signs of it yet. That, of course, may change.
-
Biorisks LLMs Critical Analysis Open Source Security
By
–
Much of the biorisk work on text-based LLMs struck us as reasonably unconvincing: it often shoed the ability of LLMs to answer a handful of Qs about pathogens…and not much else. @1a3orn
’s takedown of this work, along with their wider writing about attacks on open source, is -
AI Accelerating Biomedical Research and Therapeutic Innovation
By
–
First thing’s first, we should be clear that we are optimists about the potential of AI to accelerate research in bio and have no sympathy with calls to ‘pause AI’. We think the kind of therapeutic work companies like @profluentbio are doing will change lives for the better.
-
Inference-Time Compute: Unlocking Advanced Model Capabilities
By
–
how can we use inference-time compute to unlock improved model capabilities?
— Nathan Benaich (@nathanbenaich) 5 décembre 2024
here's @eisokant of @poolsideai on using inference-time compute, synthetic data and verifiers.
ft on @airstreetpress pic.twitter.com/3QMYb6v8fGhow can we use inference-time compute to unlock improved model capabilities? here's @eisokant of @poolsideai on using inference-time compute, synthetic data and verifiers. ft on @airstreetpress
-
User Experience Beats Model Intelligence in AI Development
By
–
“Only weirdos care about how smart the models are. Most people just want new features and nicer UX.”
-
Obesity Drug Stock Outperforms S&P 500 by 4x in 2026
By
–
yeh sp500 ripped this year = 23% or so but obesity cis ripped even more = 93% or so
-
Scaling Laws and AI Performance Walls: Analysis and Implications
By
–
and analysis from @jeremyakahn on scaling laws / walls from yesterday in @FortuneMagazine https://t.co/it4B6KWII7 pic.twitter.com/whwa30g1tD
— Nathan Benaich (@nathanbenaich) 4 décembre 2024and analysis from @jeremyakahn on scaling laws / walls from yesterday in @FortuneMagazine
-
Synthetic Data: Breaking Through AI Scaling Walls?
By
–
is synthetic data the key to pushing past scaling walls or is the snake devoring its own tail?@poolsideai co-founder and cto @eisokant, explains why he’s an optimist in our recent discussion on @airstreetpress. pic.twitter.com/J5vtthzqF3
— Nathan Benaich (@nathanbenaich) 4 décembre 2024is synthetic data the key to pushing past scaling walls or is the snake devoring its own tail? @poolsideai co-founder and cto @eisokant
, explains why he’s an optimist in our recent discussion on @airstreetpress
.