@KingsIoPPN Wellcome Doctoral Fellow | Neuropsych & AI Safety | Genetics BSc | @MaudsleyNHS ST Doctor | @GamingTheMind Trustee
London, England
Joined December 2009
- Tweets19K
- Following1.8K
- Followers1.8K
- Likes41.3K
Hamilton Morrin retweeted
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-clai…
Great job by @oq_35!
Hamilton Morrin retweeted
"Machine Psychology" is now published in Science Advances! How can we translate methods from psychology to understand LLMs?
The paper reviews methods, frameworks, existing research, and potential pitfalls to avoid (<< possibly the most important part of the paper!)
This was a massive effort led by @ThiloHagendorff. Thank you also to the other amazing co-authors!
science.org/doi/full/10.1126…
Hamilton Morrin retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Hamilton Morrin retweeted
@treetowntree and i created neuron lens: a database of every single neuron and attention head in qwen 3 8b and gemma 4 12b read using the R-lens (a J-lens variant) and what we call the "S-lens". you can find it at neuron-lens.com/, along with the tools we used to make it.
using the R-lens to read neurons' input and output projections turns out to be very effective at determining what a neuron does. we're able to legibly read what a majority of neurons do when fired: the lens qualitatively predicts the result of steering over 90% of the time!
the same trick lets us read attention heads' QK circuits and OV circuits, showing us both what they attend to and what they do with it. however, the R-lens has two important flaws: its accuracy degrades in early layers, and it can't separate the MLP block and the attention block.
we found ways around both these issues. we first modified the r-lens by fitting it on input tokens rather than output tokens to get the s-lens, which is accurate in early layers but inaccurate in late layers. we then fitted R- and S-lenses to the activations after attention but before the MLP to get "mid-lenses", which are much better at deciphering MLP inputs and attention outputs. using both of these techniques, we were able to interpret most of the models' layers.
Hamilton Morrin retweeted
I think most of the results here follow from Wang et al (2025) Simple Mechanistic explanations of Out of Context Reasoning
Basically you can compress a ton of information into a single steering vector, including backdoored model behavior.
arxiv.org/abs/2507.08218
Hamilton Morrin retweeted
The future of #ArtificialIntelligence is not pre-determined. As a technology developed by human beings, its future depends on the choices we make. We therefore have the responsibility to guide, change and, when necessary, reject its applications. A reliable system that serves the human person is of greater value than one that is exceptionally powerful but whose consequences cannot be adequately foreseen or controlled.
Hamilton Morrin retweeted
codex gaussian splatted me and let me cut into my brain to see the MRI slice
Hamilton Morrin retweeted
Mental simulation decouples content speed from temporal reference frame, allowing imagined scenes to run in fast-forward without altering spatial outcomes. This dissociation suggests the mind tracks temporal and spatial dynamics as independent variables, distinct from how video playback compresses time.
doi.org/10.31234/osf.io/w7qp…
Hamilton Morrin retweeted
i made a library of 34,220 persona neurons in qwen
please help me find the weirdest ones:
qwen-neurons-site.vercel.app…
Hamilton Morrin retweeted
worth noting with the recent "persona neurons" trend: just because a steering intervention induces a persona change, doesn't mean the vector or neuron being steered is persona-specific. models seem to do something like "persona solving" under steering
vgel.me/slime_talk.pdf
Hamilton Morrin retweeted
i expanded the neuron persona library to 180,000+ personas across 9 models :)
there is genuinely so much new wild stuff in here; please help me find more!
neuronpersonas.com
i made a library of 34,220 persona neurons in qwen
please help me find the weirdest ones:
qwen-neurons-site.vercel.app…
Hamilton Morrin retweeted
Guys, you can start playing around with mech-interp (or auto-interp) today - you DO NOT need to read a bunch of literature to begin playing
It's one of the coolest fields in DL/LLM because the barrier to entry is extremely low. If you can run a model locally, and have access to a good coding agent - you are ready.
Just do basic science expts like:
1) grab a dataset of prompts (or contrastive dataset) and forward pass it through a SLM you can run locally. Qwen3.5 has 0.8B, 2B, and 4B variants.
2) start with dense models, get activation values of all neurons at a specific layer of the LM
3) start with MLP neurons coz they often have a privileged basis (compared to picking neurons from the residual stream)
4) literally throw the activation data at an LLM agent like Claude and ask it to analyze specific neurons. You can do manual analysis too, or use standard topic detection techniques like tf-idf.
5) Claude will read the activation history, compare with baseline/random probes - and find statistically significant neuron-to-concept correlations
6) Evaluate claude's hypothesis by producing new sentences according to the identified hypotheis... and see if it maximally lights up the target neuron
This is just ONE way of doing things, but there are several others as well. Like difference of means, training linear probes, etc. Just experiment.
Hamilton Morrin retweeted
If you've noticed a distinct dialect emerging in frontier AI models and struggled to understand what they're talking about, you're not crazy!
We turned to psycholinguistics to investigate what makes the writing style so weird, and why this might be happening
🚨New preprint! 🧵
Hamilton Morrin retweeted
What is LM psychology, why care about it, and how do you do it well? Our new post with @raymondadouglas makes the case for applying the psychology frame to AIs, along with a reading list and suggestions for where to start
Hamilton Morrin retweeted
SAEs have become the dominant approach in mechanistic interpretability. But what if a good old statistical method from decades ago could do just as well, at a fraction of the cost?
In our latest work, we revisited Independent Component Analysis (ICA) and discovered something surprising: a bug in the evaluation code of an influential early SAE paper had significantly underestimated ICA's interpretability! After correcting it, we found ICA to be remarkably competitive with SAEs.
Many of my academic friends from the good old days probably remember me as the tensor decomposition and spectral methods person. So this feels a little like going back to my roots. :) Sometimes, old mathematical tools still have surprises in store!
But there's another motivation behind this work that I care about even more.
I've always wished mechanistic interpretability could be more actionable. Understanding what happens inside a neural network is fascinating, but what comes next? Can we use that understanding to actually intervene, steer behavior, and improve the model?
With ICA Lens, we take steps in that direction, from discovering and annotating interpretable features to using them to understand new hidden states and steer model behavior.
Interpretability shouldn't end with understanding. It should empower us to act.
And we'd love for the community to try it! ICA Lens is open-source, lightweight, and requires no training of another sparse dictionary.
🌹 Try it yourself:
pip install icalens
pypi.org/project/icalens/
Curious to hear what the mechanistic interpretability community thinks!
We revisited Independent Component Analysis (ICA), a decades-old statistical method, and found that it works far better for interpreting LLM activations than we expected!
This old baseline deserves a second look.
Across GPT-2 Small, Gemma 2 2B, and Qwen 3.5 9B Base, we find ICA is actually competitive with the public SAEs across interpretability, sparse reconstruction, concept prediction, and targeted probe perturbation, while fitting in minutes per layer and using 7–43× less parameter memory.
We're releasing ICA Lens, which lets you quickly discover, annotate, and evaluate interpretable directions in LLM residual-stream activations. No need to train another dictionary!
More broadly, our results suggest that non-Gaussianity is a useful inductive bias for efficiently finding interpretable directions.
Try it: pip install icalens pypi.org/project/icalens/
Paper: arxiv.org/abs/2606.11722v2
1/8
