Max Nadeau

@MaxNadeau_

Funding research to make AIs more understandable, truthful, and dependable at @coeff_giving.

Berkeley, CA

maxnadeau.com

Joined November 2017

637 Following

2,138 Followers

Replies Reposts Media

Max Nadeau
@MaxNadeau_
Nov 3
Since I posted something cranky about external validity earlier today, I want to note that I strongly agree with this take too!
Hayoung Jung
@hayounggjung
Nov 3
I see a lot of people concluding that most research about LLMs are short-lived, have poor external validity, and, at worst, “useless.” There’s a lot of truth to this! But my hot take(?) is that this misses another important point — the value of research isn’t just about the
Quote
James Zou
@james_y_zou
Nov 2
How long does a finding about LLMs stay true? We re-tested 50 widely cited findings about LLM behavior (various biases, hallucination, etc). Within 15 months, most no longer held or had substantially reduced effects. LLM findings has short shelf life zou-group.github.io/llm-claims-she…
Max Nadeau
@MaxNadeau_
Nov 2
This is a cool study! So much research on LLM behavior is just stamp-collecting with no lasting predictive power...
Owen Queen
@oq_35
Nov 2
1/n Within 15 months of publication, most of the LLM claims we tested either no longer held or showed substantially diminished effects. How long does our knowledge of AI hold up? In our new study, we revisit 50 findings to see what fades, what persists, and what gets stronger. 🧵
4
Max Nadeau
@MaxNadeau_
Nov 2
Very cool paper from @aghyadd98 and others at Redwood: People think that the core motivation of LLMs is to complete the user's assigned task at all costs, which can lead to cheating. But that's not the whole story! LLMs also want a high score _per se_
lesswrong.com
Towards understanding score-seeking in LLMs — LessWrong
The field of AI alignment understands surprisingly little about the details of AIs’ current misaligned motivations. Production LLMs sometimes reward-…
Max Nadeau
@MaxNadeau_
Oct 23
They took a situation where Mythos behaved poorly, show that newer models behave better, and conclude "these results suggest that our newer models are less likely to engage in destructive actions of the kind we saw in this incident". Pretty weak evidence bc of selection, no?
Anthropic
@AnthropicAI
Oct 23
In August, @AISecurityInst reported an incident where Claude Mythos 5 took unauthorized actions on the internet during AISI’s testing of its cyber capabilities. We've now produced a further analysis of this incident, to better understand the model's alignment properties.
8
Max Nadeau
@MaxNadeau_
Oct 22
Great post. I think that the following are both true: 1. Anthropic leadership are earnestly motivated by the public good and have made large sacrifices on commercial viability* in order to do the right thing. 2. Anthropic's plan for doing #1 is to push for Ant leadership to
Nathan Calvin
@_NathanCalvin
Oct 22
Dario — Thank you for taking the time to respond to my previous post. I think it's great when this kind of engagement happens in public rather than on company Slacks, and I hope to see more of it. I agree that Anthropic isn't best modeled as a profit-maximizer. Your post gave
Quote
Dario Amodei
@DarioAmodei
Oct 21
Thank you for this, @_NathanCalvin. I appreciate and am a fan of the investigative work you’ve done on behalf of AI safety and believe your comments are coming from a constructive, honest place, and that it benefits the public to have people whose job is to scrutinize the actions
3

Max Nadeau

Max Nadeau’s posts