New Anthropic research paper: Scaling Monosemanticity.

The first ever detailed look inside a leading large language model.

Read the blog post here: anthropic.com/research/mappi…

A very productive week in Kurdistan with CJBKR with positive exchanges and meetings. We are very grateful for the ongoing support from ETTC UK in Erbil Kurdistan Regional Government for our mental health project with OCTC Oxford
kr-uk-impakt.org/kmhp

Reminder:

A new study led by King’s College London has demonstrated the potential of Foresight, an AI tool, to predict the health trajectory of patient by forecasting future disorders, symptoms, medications and procedures.

Full story 👉 ow.ly/be0550RcXXP

Future capability to produce whole forecasts at low-cost will galvanise research on diseases and support all manner of healthcare planning x.com/skynews/status…

Our #GenerativeAI produces forecasts from text in health record across any disorder, test, medication, treatment or complication across all disease groups into the future.
This patient #digitaltwin can provide insights and 'What If' scenarios.
Video youtu.be/6i7faJV-hg4
/2

'within 5 years, deep learning is going to do better.”

Reality has been rather different. Beyond a few applications — for example, in mammography and colonoscopy — evidence of AI’s superiority over doctors remains thin.'

Headlines outpacing hard work
on.ft.com/3PCHNrM

NEW Research: Foresight—a generative pretrained transformer for modelling of patient timelines using electronic health records: a retrospective modelling study. #GPT #EHRs Zeljko Richard Dobson

Read it here: buff.ly/3VrzumE

Using CogStack deep phenotyping AI means any hospital can produce such large scale data for #GenAI

In fact, there isn't one Foresight model; we made one from each #NHS hospital using >10 years of data!
King's College NHS NHS Maudsley King's College London Institute of Psychiatry, Psychology & Neuroscience Guy's and St Thomas'
/3

Unsurprising. As anyone who has used LLMs know, it requires some cognitive effort to detect the confabulations. It is sometimes faster to just write the code from scratch than to debug output of a coding copilot. statnews.com/2024/03/13/hos…

Congrats 🎉. Looking forward to training the next generation of #NHS data scientists for data-driven #healthcare and future safe development of #healthAI kcl.ac.uk/news/kings-cen…

I have some preference to the term Confabulations as LLM outputs are similar to the fluent semi-factual veneer of believability of conversations with the Delirious patient.

Over the ☾ to be collaborating with the Responsible Technology Adoption Unit to ensure the fairness and ethical use of generative AI and LLMs technology, as part of the CogStack programme in healthcare. Thrilled to contribute to the validation of #GenAI and #llm for patient care. 🌟

The new name of Responsible Technology Adoption Unit really outlines the intention for safe adoption. Honoured to have CogStack #GenAI and #LLM being rigorously evaluated for safe risk-managed adoption.