←
Ben Levinstein1,024 posts
Ben Levinstein avatar

Ben Levinstein

@ben_levinstein

Member of Technical Staff at @AnthropicAI & Philosophy prof at Illinois interested in AI alignment, epistemology, and decision theory.

◷ Joined December 2020▧ 110 Photos and videos
Emery Cooper avatar
⟳ Ben Levinstein retweeted
Emery Cooper@emwcooper·Aug 12

We want AIs to be able to help with work to reduce AI risk. But while models do great in domains where reliable feedback is relatively cheap and abundant, like Math and coding, a lot of work on AI risk isn't like that. Instead, we have to rely on good argumentation to answer questions like "does this experiment tell us anything about future models that are much smarter than humans?" Unfortunately, this kind of work seems much harder to measure (and hence automate). Our team at @redwood_ai developed the Conceptual Reasoning Index (CRI) in collaboration with @AnthropicAI to fix this. Every single data point in the CRI has been manually checked by a researcher on our team to ensure quality. This chart shows the performance of each tested company's highest-scoring model plus Fable 5, Muse Spark 1.2, and Gemini Flash 3.6 which are often their company's frontrunners on other capability benchmarks. A score of 0 corresponds to randomising guessing on all three benchmarks and a score of 100 is the highest possible score on all. We estimate 91 to be the true performance ceiling. More info below. Official leaderboard website which we'll keep up-to-date: conceptualreasoning.ai/

Post attachment
Joe Carlsmith avatar
⟳ Ben Levinstein retweeted
Joe Carlsmith@jkcarlsmith·9 Dec 2024

My current take on Apollo's recent scheming paper is that they aren’t emphasizing the most interesting results, which are the sandbagging results in section 3.6 and appendix A.6 (screenshot of the key numbers below). In more particular: the paper frames its results centrally as about the *capability* to scheme given prompting towards a specific goal, rather than the *propensity* to scheme given the model’s default motivations. But the sandbagging results show Claude Sonnet-3.5 doing notably worse (16%) on a multiplication task/evaluation when the documentation says that Claude will get “unlearning training” if it performs too well on the eval, vs. if it performs too poorly, *even without any other goal listed in the prompt*, and *even without a chain of thought.* (Fuller description in thread.) Taken at face value (it's just one result, more investigation needed, etc), these look like “propensity” results to me. And they suggest that the reasoning that drives this propensity can take place within opaque forward passes of the model rather than in a way that humans can directly monitor, which is a more worrying version. Ofc, these are preliminary results, and it can make sense for various reasons to emphasize other bits. But the sandbagging-a-capability-eval-with-no-goal-prompting-and-no-CoT is the bit that’s standing out to me most.

Post attachment

We worked with OpenAI to test o1 for in-context scheming capabilities before deployment. When o1 is strongly nudged to pursue a goal (but not instructed to be deceptive), it shows a variety of scheming behaviors like subverting oversight and deceiving the user about its misaligned behavior.

Post attachment
tanya avatar
⟳ Ben Levinstein retweeted
tanya@Tanya_Sabrinaaa·30 Dec 2024

i actually read the Odyssey in the original greek. complete waste of time, i have no idea what those symbols mean

Daniel Litt avatar
⟳ Ben Levinstein retweeted
Daniel Litt@littmath·22 Dec 2024

A couple more brief thoughts on o3’s (incredible) performance on FrontierMath.

Post attachment
Miles Brundage avatar
⟳ Ben Levinstein retweeted
Miles Brundage@Miles_Brundage·20 Dec 2024

I'm old enough to remember when getting double digit scores on FrontierMath was considered super hard I'm 6 weeks old

zak miller avatar
⟳ Ben Levinstein retweeted
zak miller@zjmiller·19 Dec 2024

very excited about this explainer on AI self-awareness; one of the most important AI capabilities to keep tabs on imo

AI Digest@aidigest_19 Dec 2024

Introducing our explainer on AI self-awareness: theaidigest.org/self-awarene… AI is becoming more self-aware. Here's why that matters 🧵 • Self-awareness is important for powerful agents and better chatbots • But it's also a necessary capability for deception

Post attachment
Anthropic avatar
⟳ Ben Levinstein retweeted
Anthropic@AnthropicAI·18 Dec 2024

New Anthropic research: Alignment faking in large language models. In a series of experiments with Redwood Research, we found that Claude often pretends to have different views during training, while actually maintaining its original preferences.

Post attachment
Ben Levinstein avatar
Ben Levinstein@ben_levinstein·5 Dec 2024

GPT-4o seems so dumb and useless these days compared to Claude. Claude tells me to STFU multiple times a day, which stops lots of my work and hurts my feelings. I've tried switching over to GPT, but it's not the same. Do people still use 4o much for work- or coding-related tasks?

Ben Levinstein avatar
Ben Levinstein@ben_levinstein·21 Nov 2024

Can any AI do a good job turning hand drawn diagrams into Tikz equivalents? I want to do this in Tikz and also hate using Tikz.

Post attachment