GPT-4 Turbo, AI in Medicine Book closing recommendation, dearraying in two steps

Journal Club

GPT-4 Turbo and other OpenAI product upgrades (this Monday!)

Summary: https://openai.com/blog/new-models-and-developer-products-announced-at-devday 

Keynote: https://www.youtube.com/live/U9mJuUkhUzk?si=XofmBo_lY6cFcJP5 

Workshop: https://www.youtube.com/playlist?list=PLOXw6I10VTv-exVCRuRjbT6bqkfO74rWz

Hackathon

Prompt Functionalization

AI in Medicine Book closing recommendation was to embrace, hands-on, the "stages of grief" (Zak Kohane) that using LLMs will almost always trigger.

Functionalizing Language Models - make sure you wrote at least on GPT function, that you projected a (text) prompt into (numeric) latent space. Listen to your fingers.

In case it helps, an example of a GPT function:

// defining obj+fun for getting human genome sequences from UCSC genome resource

mod = 'https://episphere.github.io/gpt/functions/testFunctions.mjs'
obj = (await import(mod)).fetchUCSC
fun = (await import(mod))[obj.name]

// testing it 

seq = await fun({genome:'hg38',
        chrom:'chr1',
        start:100000,
        end:100100
    })

// should return
// 'ACTAAGCACACAGAGAATAATGTCTAGAATCTGAGTGCCATGTTATCAAATTGTACTGAGACTCTTGCAGTCACACAGGCTGACATGTAAGCATCGCCAT'

And an example of embedding onto a feature space:

await (await import('https://episphere.github.io/gpt/min.js')).embeddings('hello world')

Dearraying in two steps (Aaron, Praful)

segmentation, griding

...

WebRTC-based Federated Logistic Regression 

tested with the IRIS dataset (Praful)

...