Our previous interpretability research found that we could artificially dial up or down certain “features” within a model to modify its behavior. Here, we ran larger-scale tests of feature steering, focusing on measuring social biases.
RESEARCH
-
Feature Steering: Quantifying Model Behavior Control Trade-offs
By
–
Feature steering does successfully change how a model responds in some cases. But we hadn’t quantified these effects until now. There’s also a downside to feature steering. Steer the features too much, and the model’s overall usefulness degrades.
-
Anthropic Studies Feature Steering in AI Systems
By
–
New Anthropic research: Evaluating feature steering. In May, we released Golden Gate Claude: an AI fixated on the Golden Gate Bridge due to our use of “feature steering”. We've now done a deeper study on the effects of feature steering. Read the post: http://
anthropic.com/research/evalu
ating-feature-steering
… -
Math Papers Face Extreme Scarcity of Qualified Reviewers
By
–
ML: "We don't have enough qualified reviewers"
Math: "Hold my beer" For some math papers, the number of reviewers who can actually understand and engage with the paper is in the single digits. -
Aya-Expanse: Impressive Multilingual LLM from Cohere
By
–
Aya-Expanse is a really impressive model if you need a highly performant pre-trained LLM (or speech) in one of the below 32 languages. Congratulation @CohereForAI and @cohere Easy to combine with speech and images as well (just play with the HF Space below) Languages: Arabic,
-
Stem cell transplants show no increased cancer mutation risk decades later
By
–
Reassuring: decades after stem cell transplants, no indication of increase in cancer related gene mutations https://
science.org/doi/10.1126/sc
itranslmed.ado5108
… @ScienceTM https://
nature.com/articles/d4158
6-024-03450-x
… @Nature -
Alibaba Presents Ovis: Novel Multimodal Language Model Architecture
By
–
Alibaba presents Ovis a novel Multimodal Large Language Model (MLLM) architecture, designed to structurally align visual and textual embeddings https://
x.com/i/broadcasts/1
zqKVYbPvRPxB
… -
ESA and EUT+ Use AI to Secure Moon Landings
By
–
[#Article] The ESA and the EUT+ alliance harness AI for moon landing safety https://actuia.com/actualite/lesa-et-lalliance-eut-mettent-lia-au-service-de-la-securite-des-alunissages/
…
#AI #artificialintelligence -
AI and Machine Learning Tutorial with Bi/Cross-Encoders Deep Dive
By
–
If you enjoyed this tutorial: Find me →
@akshay_pachaar
Everyday, I write about AI and Machine Learning! Also, check our deep-dive on Bi/Cross-Encoders: -
Combining Bi-Encoders and Cross-Encoders in RAG Systems
By
–
Combining Both in RAG Systems To get the best of both worlds: Retrieve: Use Bi-Encoders as embedding-model to efficiently retrieve top candidate documents. Rerank: Use Cross-Encoders to rerank these candidates for better accuracy. Check this out 7/n