Nathan Lambert

@natolambert

Open model research, founder, director @trillium_labs. Prev. co-led Olmo at Ai2. Writes @interconnectsai, wrote posttrainingbook.com

Cambridge, MA

natolambert.com

Joined December 2014

964 Following

108.9K Followers

Replies Reposts Media Articles

Pinned
Nathan Lambert
@natolambert
16h
I regularly hear from people that my post-training book/writing has helped them get a job at labs by focusing on material that develops intuitions for how training methods work. It's very validating, because this type of "idea/math heavy" writing doesn't really sell many books.
49
Nathan Lambert
@natolambert
7h
Looks like the smartest western open weight model by a margin. While much smaller than muse spark open or mistal large. While with a better license (apache). Love it!
Thinking Machines
@thinkymachines
7h
Introducing Inkling 2 Small. Scoring 31 on the AA Intelligence Index, it's a big jump over its predecessor. Inkling 2 Small is available for fine-tuning and inference on Tinker today, with full weight release planned soon. thinkingmachines.ai/news/inkling-2…
17
Nathan Lambert
@natolambert
12h
imo a lack of good enough benchmarks is a bottleneck. At least for the type of RSI describe as an intelligence explosion. You can use existing benchmarks to make models way cheaper, though.
METR
@METR_Evals
13h
Since August, a team within METR has worked with internal Anthropic information to assess whether AI itself is accelerating AI development. We are now publishing a report summarizing the results of this risk assessment.
22
Nathan Lambert
@natolambert
16h
Have heard from many folks that score-centering is popular in industry (e.g. Tinker's StableReinforce using it), so it's the latest addition to the rlhf book. It's a nice derivation to understand training-inference-mismatch. Video version explaining it soon.
Martin Marek
@mrtnm
Sep 18
Why is RL so sensitive to training-inference mismatch (TIM)? With each training step, the trainer is pulled toward the sampler, resulting in accumulating drift. We exploit this intuition to devise a novel correction method to stabilize RL under TIM, called Score Centering.
GIF
14
Nathan Lambert
@natolambert
Nov 5
Post-training and post-post-training are really different arts
33

Nathan Lambert

Nathan Lambert’s posts