Anna Tsvetkov
@annatsv.bsky.social
Postdoc @ Princeton AI Lab
Natural and Artificial Minds
Prev: PhD @ Brown, MIT FutureTech
Website: https://annatsv.github.io/
Natural and Artificial Minds
Prev: PhD @ Brown, MIT FutureTech
Website: https://annatsv.github.io/
🔍 What are the limits of interpretability in ML?
Mech interp often stays at Marr’s algorithmic level but without the computational level (what the task is, what counts as the right solution) the mechanisms we find can look arbitrary. Why does a model learn one algorithm rather than another?
🧵 (1/2)
Mech interp often stays at Marr’s algorithmic level but without the computational level (what the task is, what counts as the right solution) the mechanisms we find can look arbitrary. Why does a model learn one algorithm rather than another?
🧵 (1/2)
November 25, 2025 at 11:53 PM
🔍 What are the limits of interpretability in ML?
Mech interp often stays at Marr’s algorithmic level but without the computational level (what the task is, what counts as the right solution) the mechanisms we find can look arbitrary. Why does a model learn one algorithm rather than another?
🧵 (1/2)
Mech interp often stays at Marr’s algorithmic level but without the computational level (what the task is, what counts as the right solution) the mechanisms we find can look arbitrary. Why does a model learn one algorithm rather than another?
🧵 (1/2)
Anthropic has a great new piece on “Signs of introspection in large language models” 👉 www.anthropic.com/research/int...
🤔 Neat evidence that LLMs can report on manipulated activations, with big caveats!
🧠 But leaves open: what are the “internal states” an LLM can introspect in the first place?
🤔 Neat evidence that LLMs can report on manipulated activations, with big caveats!
🧠 But leaves open: what are the “internal states” an LLM can introspect in the first place?
Emergent introspective awareness in large language models
Research from Anthropic on the ability of large language models to introspect
www.anthropic.com
November 1, 2025 at 4:49 PM
Anthropic has a great new piece on “Signs of introspection in large language models” 👉 www.anthropic.com/research/int...
🤔 Neat evidence that LLMs can report on manipulated activations, with big caveats!
🧠 But leaves open: what are the “internal states” an LLM can introspect in the first place?
🤔 Neat evidence that LLMs can report on manipulated activations, with big caveats!
🧠 But leaves open: what are the “internal states” an LLM can introspect in the first place?
Reposted by Anna Tsvetkov
This is a beautiful paper! The first third helpfully labels a stream of recent work in philosophy of AI as "propositional interpretability". The idea is to use propositional attitudes like belief, desire, and intention, to help explain AI in a way that we can understand. 1/n
a draft paper (for an invited talk at AAAI next month) with a philosophical analysis of work on mechanistic interpretability, with special attention to methods for propositional interpretability.
arxiv.org/abs/2501.15740
arxiv.org/abs/2501.15740
Propositional Interpretability in Artificial Intelligence
Mechanistic interpretability is the program of explaining what AI systems are doing in terms of their internal mechanisms. I analyze some aspects of the program, along with setting out some concrete c...
arxiv.org
January 29, 2025 at 1:24 PM
This is a beautiful paper! The first third helpfully labels a stream of recent work in philosophy of AI as "propositional interpretability". The idea is to use propositional attitudes like belief, desire, and intention, to help explain AI in a way that we can understand. 1/n
Reposted by Anna Tsvetkov
"The AI risk repository, which includes over 700 AI risks grouped by causal factors (e.g. intentionality), and domains (e.g. discrimination), was born out of a desire to understand the overlaps and disconnects in AI safety research"
#AIEthics
techcrunch.com/2024/08/14/m...
#AIEthics
techcrunch.com/2024/08/14/m...
MIT researchers release a repository of AI risks | TechCrunch
A group of researchers at MIT and elsewhere have compiled what they claim is the most thorough databases of possible risks around AI use.
techcrunch.com
January 5, 2025 at 9:03 PM
"The AI risk repository, which includes over 700 AI risks grouped by causal factors (e.g. intentionality), and domains (e.g. discrimination), was born out of a desire to understand the overlaps and disconnects in AI safety research"
#AIEthics
techcrunch.com/2024/08/14/m...
#AIEthics
techcrunch.com/2024/08/14/m...