#artificialintelligence #100: How would a copilot first approach look like if we started with a generative / assisted strategy ?

Hello all

welcome to #artificialintelligence #100

The 100th episode - we have more than 71,000 subscribers.

Thanks for your support

I have been impressed by the Microsoft 365 copilot (

https://lnkd.in/eR7pQUZ3 ) and what it means for the future of AI - especially the possibility of making AI inclusive. I think most people would agree that LLMs are a game changer. Inspired by a post from???Jepson Taylor ?(shared by?Bronia Anderson-Kelly )

I am trying to create an overall thesis / workflow for the new way of working with AI. We can call it low-code but the primary motivation is that its driven by LLMs

So in this context

If we started with a generative / assisted strategy first how would the development world look like?

We first look at modalities of data(tabular, language and image)

And then how these can be effectively made to be low code driven by LLMs (effectively a copilot first approach)

1)?Start with domain knowledge

2)?Use GPT-4 to summarise ideas and to create new suggestions

3)?Define a validation criteria - (how do you know what success looks like)

4)?Generate code using co-pilot, prompt engineering and fine tuning strategies

5)?Validate output as per criteria

The overall picture looks like below

Notes

a) automl as it stands is largely not LLM driven - but there is some work in this area also?https://lnkd.in/e57y9bMV

b) the word LLM also is a bit dated (already) since?GPT-4 is more than a language model

c) This does not replace all forms of development ex using traditional python etc - but it is a major shift

d)?The capabilities of copilot and LLMs to generate code are indeed very impressive as you see below

KdNuggets chatGPT for data science cheat sheet https://lnkd.in/e8ev2SGT

copilot for datascience?review https://lnkd.in/eTEWdxbc

How ChatGPT Helps You To Automate Machine Learning https://lnkd.in/eTc_qSSA

ChatGPT can solve simple machine learning tasks as classification and categorization https://lnkd.in/etBGEpdP

e) GPT can be discriminative i.e. if you ask it a question that needs a specific response (in contrast to other values), it can do so as well.

f)?https://lnkd.in/eV5QYJyV ?is a good link

I was inspired by this?post also ie to rethink education an the wold of work?(https://lnkd.in/exBm2kEE )

If you want to study these themes with us please see our course

https://lnkd.in/efKkYEpD

Also, if you want to stay in touch with me re launch?of the Erdos institute and other work please join my substack https://lnkd.in/e4-hAzPT