This is really interesting: https://www.nature.com/articles/s41562-024-02046-9 "Large language models surpass human experts in predicting neuroscience results" A study on how well trained LLMs can predict the "correctness" of a new neuroscience study/claim based on looking into the associations of the mentioned claims/words in the new study compared to a reference set of thousands of previously published neuroscience papers. The idea here is that there are associations hidden in the vast amount of literature out there that humans in general and singular scientists in particular just cannot read, remember, grasp and comprehend anymore due to the exponential growth in studies. Importantly all thus trained LLMs outperform a large panel of human expert in predicting whether or not a new study or a fake study is true. This is really fascinating and could have major implications to identify new disease MoA hypothesis, target validationa nd dicovery, discover hidden connections and low-hanging treatment option fruits that nobody has yet considered or simply just weed out fake or irrelevant new er publications. #ai #science #neuroscience #drugs #future #LLM #nostr

Are LLMs better doctors? LLM model on its own far outperforms physicians when giving the correct diagnosis. Funny thing is, even when physicians are allowed to use AI as support, their correct diagnosis ratio basically does not increase, indicating a tendency of the physicians to cling to their own diagnosis and ignoring the LLMs advice. https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2825395 #ai #science #health

A recent paper discussing the framework to understand wjether or not LLMs can be conscious: https://www.nature.com/articles/s41599-024-03553-w I don't believe a LLM can ever be considered conscious for many reasons and only other AI approaches could get there but still an interesting read. #ai #llm #chatgtp #consciousness #neuroscience #science #philosophy

Putting LLMs capacities into context:

ACL Anthology

Are Emergent Abilities in Large Language Models just In-Context Learning?

Sheng Lu, Irina Bigoulaeva, Rachneet Sachdeva, Harish Tayyar Madabushi, Iryna Gurevych. Proceedings of the 62nd Annual Meeting of the Association f...

long story short: let's not overestimate what they are capable of. The main danger might in fact be that we as humans begin to overestimate their ability to reason (of which there is none) and begin to trust or follow their guidance. #AI #ai #LLM #science #intelligence #chatGTP #cognition

Interesting look into the political biases of the most popular LLM AI models:

The political preferences of LLMs

I report here a comprehensive analysis about the political preferences embedded in Large Language Models (LLMs). Namely, I administer 11 political ...

#ai #llm #chatgtp #nostr #politics #science

Interesting pre-print how Chat GTP & LLMs are changing the way scientific publications are written:

Delving into ChatGPT usage in academic writing through excess vocabulary

Kind of understandable that especially non-English-native speakers make use of it to polish their articles. But words matter, especially when trying to describe a new finding as scientifically accurate as possible. I wonder to what extend this verbose and flowery language is then really understood by the authors themselves and whether or not those words really correctly describe their findings or not. #science #writing #ai