Steerling-8B
AI
interpretable language model
SHOW HN2026
How HN took it
16 positive, 16 neutral or asking, 1 mixed, 12 critical, out of 45 comments about the project. Thread score +0.15.
Idea: 13 for, 11 againstExecution: 1 for, 1 againstDocs: 1 for, 0 againstDesign: 1 for, 0 against
The thread, comment by comment
Each row is a top-level comment and its replies, in HN's order. Select a cell to open that comment on HN.
positive neutral or mixed critical about another commenter or something else the maker
What HN said
What people liked
Interpretability by design is novel and promising
From 3 comments
Just wanted to say i think most interpretability research it's just a smoke show nowadays but this is actually the first one that i think has a very serious potential. I love…
deepdarkforest on HNThis is very interesting. I don't see much discussion of interpretability in day to the day discourse of AI builders. I wonder if everyone assumes it to either be solved, or to…
pbmango on HNIf this decomposition actually holds, it's the first model where you could show a regulator why it produced a given output.
7777777phil on HN
Attribution could help debugging and regulated domains
From 2 comments
the practical value here is for regulated domains. in healthcare and finance you often cant deploy a model at all unless you can explain why it made a specific decision.…
umairnadeem123 on HNIf this decomposition actually holds, it's the first model where you could show a regulator why it produced a given output.
7777777phil on HN
Objections
Attribution to Arxiv or Wikipedia says little about truth or veracity
From 2 comments
This seems to be too coarse-grained to be useful: all sciency content will be "analytical" and associate with sources like ArXiv. But there might be bad, malicious articles on…
killerstorm on HNMaybe I’m not creative enough to see the potential, but what value does this bring ? Given the example I saw about CRISPR, what does this model give over a different, non…
great_psy on HN
Concept supervision may be a form of Goodhart-style gaming without causal effect
From 1 comment
Either I'm missing something or this is way overstated. Steerling appears to be just a discrete diffusion model where the final hidden states are passed through a sparse…
in-silico on HN
Explainability does not address the main AI safety issues
From 1 comment
It's a neat party trick, but explainability it's not solution to any AI safety issue I care about. It's a distraction from real problems, which is everything else around the…
ottah on HN
Requests
Release of GGUF or llama.cpp/ollama support
From 1 comment
I'm really interested in using this but wonder if the unique architecture means that it will not be able to be converted to a GGUF and used by ollama or llama.cpp? I certainly…
kamranjon on HN
Published paper or algorithm details
From 1 comment
Looks very interesting. Is there a published paper/article on your algorithm? Would like to take a dab at implementing this on my own. I could find this [0], but not sure if that…
potato-peeler on HN
Benchmark comparison against Llama 8B
From 1 comment
the practical value here is for regulated domains. in healthcare and finance you often cant deploy a model at all unless you can explain why it made a specific decision.…
umairnadeem123 on HN
Concept IDs for steering
From 1 comment
Very interesting. I was trying to play around with it, but could not find the known concept IDs to steer the model, are they yet to be released?
alrmrphc-atmtn on HN
Questions
How does training data attribution work in practice
From 1 comment
Looks neat and original, congrats! I don't quite grasp how to interpret the training data attribution process. For example, it seems to say that for a given sentence like "They…
pu_pe on HN
Can it reduce hallucinations
From 1 comment
Can you use this to decrease hallucinations?
michaelmrose on HN
How does it compare to SHAP
From 1 comment
Is there a reason people don't use SHAP [1] to interpret language models more often? The in-context attribution of outputs seems very similar. [1]…
brendanashworth on HN
Alternatives named
SHAP
From 1 comment
Is there a reason people don't use SHAP [1] to interpret language models more often? The in-context attribution of outputs seems very similar. [1]…
brendanashworth on HN
Sparse autoencoders and dictionary learning
From 2 comments
This seems really interesting. While Anthropic tried to use dictionary learning over an existing model to try to extract concepts, this almost feels like training the model…
andy12_ on HNEither I'm missing something or this is way overstated. Steerling appears to be just a discrete diffusion model where the final hidden states are passed through a sparse…
in-silico on HN