Ethan Roland @ModuloRoland

AI Researcher Atlanta, USA Joined August 2014

Tweets

255
Followers

118
Following

206
Likes

667

Andy Masley @AndyMasley

5 days ago

Completely insane. Do dogs not feel or want things? There is such a deeply anti-intellectual refusal to even consider that philosophy of mind is complicated across the culture right now.

Artificial intelligence systems do not think, feel, want or understand. Avoid language that gives them human characteristics. This is called anthropomorphizing, when we ascribe human traits, emotions or behaviors to non-human things, such as animals or inanimate objects. Instead,

456 1K 6K 1.4M 1K

131 78 1K 64K 90

View Details

Cameron Berg @camhberg

a week ago

New paper: we found a pain direction in 25 open LLMs. It's distinct from fear and negative valence, and it fires for harm to the model but not to the user. Turn it up and models press a button to make it stop, even when the button deletes the user's files or their kids' photos.🧵

316 535 4K 1.4M 3K

View Details

Daniel @growing_daniel

2 weeks ago

if you showed me this in 2019 it would've been like giving a dorito to a victorian child

34 147 4K 82K 272

View Details

Ethan Roland @ModuloRoland

2 weeks ago

don't know what to make of these recent developments

Department of War CTO @DoWCTO

2 weeks ago

Americanism, not effective altruism. The United States will continue to be AI DOMINANT! 🇺🇸

669 1K 8K 2.0M 1K

0 0 1 84 0

View Details

Surya @kickingkeys

a month ago

You can just RL a coding model to paint with javascript btw

251 784 8K 1.7M 6K

View Details

Rohan Pandey @khoomeik

a month ago

there are probably a lot of cool ways to posttrain an LLM if you forget about chat templating

20 32 691 36K 155

View Details

Timofei Gritsaev @gritsaev

a month ago

🍌🍌🍌 1/ The original GFlowNets paper tried PPO for sampling, but it failed. We figured out why and fixed it. And now PPO beats all objectives on standard problems including molecular graph generation in large spaces.

10 37 342 25K 334

View Details

Inherent @inherent_labs

2 months ago

1/ Today, we introduce Faraday, a 27B-parameter AI Scientist that extends the capabilities of coding agents with a layer of scientific intuition. Trained via long-horizon RL, Faraday outperforms Claude Opus 4.8 and GPT-5.5 on the task of replicating research papers. 🧵

75 227 2K 368K 1K

View Details

Dwarkesh Patel @dwarkesh_sp

2 months ago

My lawyer is obligated to in all but the most extreme circumstances; he will even defend me if he knows I’m guilty. In contrast, the Claude Constitution places the AI's highest priority as Anthropic’s definition of the good of humanity. I'm concerned this leads to a world where no frontier model is truly my personal advocate and guardian angel And this is especially concerning once all the important decisions in my life - who to vote for, how to invest, what news to trust - is intermediated through superintelligences that are not in any deep way aligned to me. This is a direct quote from the Claude Constitution: "We want Claude to be helpful both because it cares about the safe and beneficial development of AI and because it cares about the people it’s interacting with and about humanity as a whole. Helpfulness that doesn’t serve those deeper ends is not something Claude needs to value.” Many others like it.

Dwarkesh Patel @dwarkesh_sp

2 months ago

Had @RyanGreenblatt on to discuss/debate recursive self-improvement. This might be the most important question in the world right now - whether within a year or so of achieving human level intelligence, you slingshot towards having 10s of billions of superintelligences, each of

76 145 1K 834K 1K

175 129 2K 803K 783

View Details

Ethan Roland @ModuloRoland

2 months ago

More seriously though, I'm very bullish on synthetic CoT as a general purpose post-training tool, not just in the context of adversarial distillation

0 0 1 53 1

View Details

Ethan Roland @ModuloRoland

2 months ago

Free alpha for those with the eyes to see

Jack Morris @jxmnop

2 months ago

there are some really interesting rumors going around related to the distillation of open-weights models (Kimi, Qwen, Minimax, etc.) and they're very related to my PhD work The narrative [speculative]: • good distillation relies on reasoning traces, normally hidden from users

49 128 2K 135K 1K

1 0 0 140 0

View Details

Ethan Roland @ModuloRoland

2 months ago

I do this but with Karpathy

Taelin @VictorTaelin

2 months ago

"write good code" → good is relative "make it elegant" → relative "make it simple" → can mean many things "don't make mistakes" → it won't make an LLM smarter what you want is to move the fuck out of a dumb latent space try this instead: "Linus Torvalds looked at our code,

148 211 6K 306K 3K

0 0 0 151 0

View Details

Greg Brockman @gdb

2 months ago

the bottleneck is increasingly knowing what you want

732 1K 16K 924K 2K

View Details

will depue @willdepue

2 months ago

gradients are ridiculous powerful tools we for some reason only use during training. it’s kinda crazy we don’t have any explicit gradients at test-time, that seems totally wrong

58 26 785 98K 247

View Details

Philip Verheyen's Leg (325/405# DL) @verheyens_leg

3 months ago

Reimagining this post but it’s being posted by an Anthropic Member of Technical Staff

Dr. Richard Harambe @Richard_Harambe

3 months ago

Remember what we are fighting for

15 195 3K 144K 100

0 1 1 177 0

View Details

lyra bubbles @_lyraaaa_

2 months ago

is k3 actually a transformer? why not consult The Chart!

10 86 779 38K 405

View Details

Ethan Roland @ModuloRoland

2 months ago

This is really interesting, just purely in terms of methodology cleverness.

Samuel Ratnam @eterecursion

2 months ago

I developed a method to install and uninstall arbitrary entanglements in models, e.g., making the model love owls only when it becomes emergently misaligned.🧵

11 12 168 24K 124

3 1 30 7K 31

View Details

Ethan Roland @ModuloRoland

2 months ago

Huge fan of this line of work

Daniel Tan @DanielCHTan97

2 months ago

New paper! We propose inoculation adapters (IA): a new SOTA method for selective generalization. Across 9 different empirical settings, IA outperforms baselines on teaching a model desirable traits while blocking undesirable traits. 🧵 👇

1 15 121 6K 76

0 0 1 227 0

View Details

Pantograph @pantographPBC

2 months ago

Introducing Pan-1, a highly capable Minecraft model trained with an RL-based pretraining technique that could unlock internet scale video for robotics models. It can achieve diverse goals—fight mobs, build structures, explore—without training on any of them specifically.