Lightnews — Scholar-powered news

Nathan Lambert

@natolambert.bsky.social

13K followers 270 following 1.6K posts

A LLN - large language Nathan - (RL, RLHF, society, robotics), athlete, yogi, chef Writes http://interconnects.ai At Ai2 via HuggingFace, Berkeley, and normal places

interconnects.ai

Posts Media Videos Starter Packs

Pinned

Nathan Lambert @natolambert.bsky.social · Apr 16

First draft online version of The RLHF Book is DONE. Recently I've been creating the advanced discussion chapters on everything from Constitutional AI to evaluation and character training, but I also sneak in consistent improvements to the RL specific chapter.

rlhfbook.com

3 17 110

Nathan Lambert @natolambert.bsky.social · 1d

no its how AI works these days

Nathan Lambert @natolambert.bsky.social · 1d

For folks at COLM, my talk is in 524C @ 12:00PM to share the various things that go into building a reasoning model from scratch. See you soon!

Will not be recorded and slides will only be released when we can get models out that we're happy with.

Nathan Lambert @natolambert.bsky.social · 1d

When talks start with "we can all likely agree that reasoning is the path to AGI" you know its going to be a doozy

Nathan Lambert @natolambert.bsky.social · 1d

Talk from Wenting Zhao of Qwen on their plans during COLM. Seems like 1 word is the plan still: scaling training up! Let’s go.

2 2 11

Nathan Lambert @natolambert.bsky.social · 2d

TODAY, so in like 80minutes

Nathan Lambert @natolambert.bsky.social · 2d

Open Models Talk and COLM 2025 is happening at 524C (end of the conference center) at 2pm.

1 6

Nathan Lambert @natolambert.bsky.social · 4d

Perfect timing for COLM2025 here in Montreal.

Nathan Lambert @natolambert.bsky.social · 4d

Augmentation of humans and a restructuring of the research org is far more likely.

Among other thoughts on a great conference!

buff.ly/mzrRumA

Thoughts on The Curve

The conference and the trajectory.

buff.ly

Nathan Lambert @natolambert.bsky.social · 4d

The Curve is a new style of mini AI conference to debate AI progress.

Here I reflect on it and explain why the argument that AI will fully replace human research engineers, and then scientists, is far fetched in the years of compute scarcity.

1 14

Nathan Lambert @natolambert.bsky.social · 5d

I can't believe we're actually as a general public going to need to be using quadrillion all the time in every day discourse, starting with tokens processed.

We'll blink and measuring in trillions won't cut it anymore.

1 13

Nathan Lambert @natolambert.bsky.social · 5d

Oh yeah I do this but wear many hats and things change fast these days…

1 3

Nathan Lambert @natolambert.bsky.social · 5d

You should be spending 10+minutes on slides per minute of your talk.
Doing too many talks makes it so you don't have time for top quality.

2 1 9

Nathan Lambert @natolambert.bsky.social · 5d

One of the mindviruses we may never fully squash is that people think you should and CAN actually ban open weights AI models in the U.S.

Good luck with that buckaroo. We have plenty of time to prepare for any potentially dangerous open models of the future.

2 24

Nathan Lambert @natolambert.bsky.social · 5d

I gave a talk today at The Curve on the state of open models.
Here are the slides, recording soon.

Topics include: Chinese ecosystem, reflections on DeepSeek, the demise of Llama, who will fill the U.S. market, what local models do, ATOM project & ai2, and more topics
buff.ly/8BiC67C

3 6 59

Nathan Lambert @natolambert.bsky.social · 7d

A ton of attention over the years goes to plots comparing open to closed models.
The real trend that matters for AI impacts on society is the gap between closed frontier models and local consumer models.
Local models passing major milestones will have major repercussions.
buff.ly/ccMJydQ

1 8 56

Nathan Lambert @natolambert.bsky.social · 7d

What changed? Despite many wonderful models, Anthropic never really remotely translated to LMArena.

The core question -- has LMArena's users or Anthropic's models shifted? Or both?

2 1 20

Nathan Lambert @natolambert.bsky.social · 8d

Mostly releasing base models, but it’s not a substantive analysis just a tweet

1 2

Nathan Lambert @natolambert.bsky.social · 8d

Seeing Qwen as trying to build Android for AI models: Cheap, everywhere, powerful, modifiable, should give you a good sense of their strategy.

1 5 17

Nathan Lambert @natolambert.bsky.social · 9d

Recent dean: buff.ly/n2ygaU5
My original post: buff.ly/rUgJUxQ
More on character training with a model spec update: buff.ly/YwVvjiM
On sychophancy in GPT4o, and how it related to the model spec: buff.ly/p1b4M1H

"Be It Enacted"

A Proposal for Federal AI Preemption

buff.ly

1 2

Nathan Lambert @natolambert.bsky.social · 9d

As Olmo gets better, this has been on my list to craft one and share the process, difficulties of following it, and so on. I welcome pressure on needing to deliver this in order to set a better example.

1 1

Nathan Lambert @natolambert.bsky.social · 9d

The behavior of these models is actually remarkably steerable (sharing more research I'm involved with on this soon!) and the lack of model specs is pretty awful as a community standard.

Links to Dean's piece, and my older pieces on model specs are all below.

1 2

Nathan Lambert @natolambert.bsky.social · 9d

Largely these seem to be blocked on politics -- both internal where teams actually cant agree what the model should do -- and external, where labs fear pushback.

Nathan Lambert @natolambert.bsky.social · 9d

The model spec sets the intentions of how the model should behave, irregardless as to if it succeeds in it.

Again, I'm happy to discuss this with labs as a free consult as I think its great for the world.

1 1

Nathan Lambert @natolambert.bsky.social · 9d

I haven't posted about Model Spec's in a while, but Dean gave me a shoutout on my earlier writing on them, so its time to say definitively again that every frontier lab should have a model spec. It builds long term trust with users, developers and regulators.

1 2 9