Replying to @_khaledsaab
@KhaledSaab11 et al. @GoogleDeepMind upgrade their medical AI AMIE with multi-modal reasoning and a state-aware diagnostic framework, outperforming primary care physicians in a randomized, double blinded diagnostic study.
Read more here: gstatic.com/amie/multimodal_…
Zhou et al. curate an expert-labelled dataset and formulate a novel framework Dual-Inf to improve LLM diagnoses and explanations for automatic differential diagnosis.
(corresponding author: @RuiZhang1229)
🧵 5/5
A study from @LeapLabTHU found that reinforcement learning with verifiable reward (RLVR) improves a model’s ability to select correct reasoning paths on the first try, but does not create new strategies beyond those already present in the base model.
🧵 4/5
@ShiwarskiLab developed a comprehensive tissue engineering platform to fabricate organ-on-a-chip devices. Using FRESH bioprinting to extrude cellulose-based scaffolds dubbed CHIPS, they printed a pancreatic-like system that excreted 8 ng of insulin in 24 hours.
🧵 3/5
Another fun week of decoding for the team!
This week we looked at a novel gene editing paradigm, drug factories and advice for life (from a molecular biologist of course! @LifeSciVC)
Read BioByte 082 here: decodingbio.substack.com/p/b…





