Journal Club

Smoothing Lexis diagrams using kernel functions: A contemporary approach

Philip S Rosenberg1 , Adalberto Miranda Filho1,*, Julia Elrod1,*, Aryana Arsham2, Ana F Best3,* and Pavel Chernyavskiy4,*

Statistical Methods in Medical Research (2023)

 https://doi.org/10.1177/09622802231192950 

Hackathon

GPT Functions, https://openai.com/blog/function-calling-and-other-api-updates 

Examples / experiments:

• Lee's chaining functions: https://observablehq.com/@siliconjazz/gpt-function-loop 

• Jonas https://episphere.github.io/gpt
  try, for example,
  await gpt.completions("say hello to C4B on thanksgiving day of 2023",null,null,null,"https://episphere.github.io/gpt/functions/testFunctions.mjs")
  or
  await gpt.completions("say hello to functions after checking the weather forecast for Friday",null,null,null,"./functions/testFunctions.mjs")
  Note how at the end it is up to you feed a function role back o the message array:
  "messages": [
  {"role": "user", "content": "What is the weather like in Boston?"}, {"role": "assistant", "content": null, "function_call": {"name": "get_current_weather", "arguments": "{ \"location\": \"Boston, MA\"}"}},
  {"role": "function", "name": "get_current_weather", "content": "{\"temperature\": "22", \"unit\": \"celsius\", \"description\": \"Sunny\"}"}
  ]

• Aaron ...

• ...

...

Upcoming hands-on GPT - discuss

Hi Oscar,

Adding a Bioinformatics resource discovery and orchestration focus section (#5), and a discussion (#7) at the end.

> To enhance the learning experience, I believe that using familiar examples is always beneficial. As such, I certainly can prepare a cancer dataset utilizing TCGA; it could be in a TXT format.

No need to prepare data - discovering data and workflows is part of what GPT does through APIs. Doing it from familiarity RStudio is covered by the new agenda: 

LLM GPT programmatic and hands-on

Learn how GPT works, and how you can make it work for you.

1. Why APIs?

So you can use your favorite language, as part of your own analytics workflows.
And such that you can have someone else pay for it. We'll return to this at the end #5

2. Get and manage API keys
   OpenAI will give you $5 worth of tokens to get started
   https://platform.openai.com/account/api-keys

3. Understand API markup with role-playing
   System, user, assistant and function roles
   https://platform.openai.com/playground

4. Going from words to numbers
   Using the embedding API to map prompts to a numeric coordinate space, where you can approach it with regular computational statistics and machine learning.
   https://platform.openai.com/docs/guides/embeddings/what-are-embeddings
   
   

5. Bioinformatics resource discovery and orchestration
   How to operate GPT from the command line and from RStudio. Install TheOpenAIR package (http://openair-lib.org) and register your OpenAI key with the function openai_api_key.

6. Money and governance
   Who pays and who governs 

7. Discussion: What LLMs are and what they are not
   GPT provides the missing resource discovery and orchestration language. The analytical workflows, such as Bioinformatics, are external to GPT ... until you weave them in. This will be illustrated with applications to genomics. A brief demonstration of GPT functions, the next frontier, concludes the workshop.

Contention point: without APIs there is no data science - or scalable bioinformatics for that matter. There is just bondage to an unFAIR user interface. If we fail to pass this message all we achieve with the workshop is entertainment. We'll discuss this point today at Cloud4Bio, please feel welcome to join us 🙂

__________

Icecream in the terrace 1-3 pm