Lightnews — Scholar-powered news

Stephanie Hyland @hylandsl.bsky.social · 18d

uv is so good

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 18d

uv is now stable enough that I'm switching my course over to using it. No more conda, we will create a generation of uv evangelicals

5

Reposted by Stephanie Hyland

Alexander Doria @dorialexander.bsky.social · 29d

Some papers really have a good intro

4 1 16

Reposted by Stephanie Hyland

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · Aug 17

The more rigorous peer review happens in conversations and reading groups after the paper is out with reputational costs for publishing bad work

2 5 49

Reposted by Stephanie Hyland

Jack Lynch @jacklynch000.bsky.social · Aug 17

I'll admit, I was skeptical when they said Gemini was just like a bunch of PhDs. But I gotta admit they nailed it.

Google's Gemini AI tells a Redditor it's 'cautiously optimistic' about fixing a coding bug, fails repeatedly, calls itself an embarrassment to 'all possible and impossible universes' before repeating 'I am a disgrace' 86 times in succession

71 1.7K 7.3K

Stephanie Hyland @hylandsl.bsky.social · Aug 14

what is the purpose of VQA datasets where text-only models do better than random?

1

Stephanie Hyland @hylandsl.bsky.social · Aug 13

lads can we stop

Zotero screenshot showing four different papers with titles beginning with "MedAgent"

4

Reposted by Stephanie Hyland

Tim Kellogg @timkellogg.me · Aug 2

quick diagram of Bluesky’s architecture and why it’s nicer here

diagram from Anthropic paper with an icon & label that says “subtract evil vector”

4 5 73

Stephanie Hyland @hylandsl.bsky.social · Aug 3

Emojis and massive try: except blocks. GitHub Copilot (at least Claude Sonnet 4) is very concerned about error handling.

1 2

Stephanie Hyland @hylandsl.bsky.social · Jul 30

if openreview were a lot fancier you could dynamically reallocate/cancel remaining reviews once a paper meets that expected minimum.

ideally you would mark these remaining reviews as optional rather than fully cancelled, in case that reviewer has already done work

3

Stephanie Hyland @hylandsl.bsky.social · Jul 30

it's frustrating how inefficient review assignments are: we target a minimum number of completed reviews per paper but in accounting for inevitable no-shows, some people end up doing technically unnecessary (if still beneficial) reviews

1 1

Stephanie Hyland @hylandsl.bsky.social · Jul 29

How many AI researchers fold their own laundry?

2

Stephanie Hyland @hylandsl.bsky.social · Jul 25

I am in the UK so feel free to discard, but I recently noticed Discord asking for age verification for some channels:

Stephanie Hyland @hylandsl.bsky.social · Jul 18

ALSO we have released the SAEs we trained, and the automated interp for all(!!)* features:
huggingface.co/microsoft/ma...

*all features for a subset of SAEs, we didn't run the full auto-interp pipeline on the widest SAE

microsoft/maira-2-sae · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

huggingface.co

4

Stephanie Hyland @hylandsl.bsky.social · Jul 18

We also found that the majority of the SAE features remained "uninterpretable", indicating room for improvement both in automated interpretability (we focused primarily on textual features!), but perhaps also questioning the SAE training and modelling assumptions. More work to be done here ✌️

1 2

Stephanie Hyland @hylandsl.bsky.social · Jul 18

... and in some cases we were able to steer MAIRA-2's generations, selectively introducing or removing concepts from its generated report.

But steering worked inconsistently! Sometimes it did nothing, or introduced off-target effects. We still don't fully understand when it will work.

1 1

Stephanie Hyland @hylandsl.bsky.social · Jul 18

We found interpretable and radiology-relevant concepts in MAIRA-2, like:
- "Aortic tortuosity or calcification"
- "Placement and position of PICC lines"
- "Presence of 'shortness of breath' in indication"
- "Describing findings without comparison to prior images"
- "Use of 'possible' or 'possibly'"

1 1

Stephanie Hyland @hylandsl.bsky.social · Jul 18

We performed the full pipeline of SAE training, automated interpretation with LLMs, steering, and automated steering evaluation.

1 1

Stephanie Hyland @hylandsl.bsky.social · Jul 18

New work from my team! arxiv.org/abs/2507.12950
Intersecting mechanistic interpretability and health AI 😎

We trained and interpreted sparse autoencoders on MAIRA-2, our radiology MLLM. We found a range of human-interpretable radiology reporting concepts, but also many uninterpretable SAE features.

Insights into a radiology-specialised multimodal large language model with sparse autoencoders

Interpretability can improve the safety, transparency and trust of AI models, which is especially important in healthcare applications where decisions often carry significant consequences. Mechanistic...

arxiv.org

1 4 11

Stephanie Hyland @hylandsl.bsky.social · Jul 17

Mexico is an *official* NeurIPS event, it’s an additional location for the conference and is different to the endorsement of EurIPS.

1 1

Stephanie Hyland @hylandsl.bsky.social · Jul 17

It’s an endorsed event but is not actually officially NeurIPS! Maybe if this experiment works well there will be more distributed (official) NeurIPS locations in future.

1 1

Reposted by Stephanie Hyland

NeurIPS Conference @neuripsconf.bsky.social · Jul 16

We're excited to announce a second physical location for NeurIPS 2025, in Mexico City, which we hope will address concerns around skyrocketing attendance and difficulties in travel visas that some attendees have experienced in previous years.

Read more in our blog:
blog.neurips.cc/2025/07/16/n...

1 21 45

Reposted by Stephanie Hyland

Sebastian Bordt @sbordt.bsky.social · Jul 10

During the last couple of years, we have read a lot of papers on explainability and often felt that something was fundamentally missing🤔

This led us to write a position paper (accepted at #ICML2025) that attempts to identify the problem and to propose a solution.

arxiv.org/abs/2402.02870
👇🧵

1 5 12

Reposted by Stephanie Hyland

Jessica Hullman @jessicahullman.bsky.social · Jul 2

ExplainableAI has long frustrated me by lacking a clear theory of what an explanation should do. Improve use of a model for what? How? Given a task what's max effect explanation could have? It's complicated bc most methods are functions of features & prediction but not true state being predicted 1/

2 8 44

Reposted by Stephanie Hyland

Ida Momennejad @neuroai.bsky.social · Jun 20

Pleased to share our ICML Spotlight with @eberleoliver.bsky.social, Thomas McGee, Hamza Giaffar, @taylorwwebb.bsky.social.

Position: We Need An Algorithmic Understanding of Generative AI

What algorithms do LLMs actually learn and use to solve problems?🧵1/n
openreview.net/forum?id=eax...

3 37 150

Stephanie Hyland @hylandsl.bsky.social · Jun 27

This is cool arxiv.org/abs/2505.18235.
Linear representation hypothesis discourse needs more differential geometry i m o

The Origins of Representation Manifolds in Large Language Models

There is a large ongoing scientific effort in mechanistic interpretability to map embeddings and internal representations of AI systems into human-understandable concepts. A key element of this effort...

arxiv.org

7