Add in Generative AI to the "What is Machine Learning good at?" section

[MikeTrizna]

The current version of the episode (https://github.com/carpentries-incubator/machine-learning-librarians-archivists/blob/e51b3327aa1ae9a4fb470ab751ac31af477bc02c/_episodes/04-what-is-ml-good-at.md) does not include any mentions of Generative AI. We need to add that in here -- in the context of Tasks (as mentioned in Issue #87 )

[MikeTrizna]

Here is an outline, pulled from new material I added to a workshop I gave last week (slides 29-35 here):

• Generative AI
  • Generative AI models learn the patterns and structure of their input training data and then generate new data that has similar characteristics.
• ChatGPT and LLMs
  • ChatGPT is a Chat interface to a Generative Pre-trained Transformer (GPT) machine learning model. A generative model is one that generates believable text or images.
  • GPT4 is currently the state-of-the-art Large Language Model (LLM): a text-based machine learning model that is trained on huge amounts of text to "learn" relationships between words.
• LLMs: Emerging properties
  • As the scale of the model increases, the performance improves across tasks while also unlocking new capabilities.
  • But many of these tasks are actually ones we covered in NLP tasks.
• LLMs: Additional applications
  • Text generation
  • Code generation
  • Multi-turn chat conversation
  • Evaluation and scoring of text inputs following provided criteria
  • Complex instruction following
• LLMs: Hallucinations
  • “Hallucinations,” i.e. ChatGPT can make things up completely.
• LLMs: Prompt Engineering
  • Prompt engineering is the process of strategically designing and refining the input prompts given to an LLM, to elicit the most accurate, relevant, and useful responses.
  • The effectiveness of an LLM's response heavily depends on how the prompt is structured -- different prompts can lead to vastly different outputs, even with the same underlying question or task.
• LLMs: Retrieval Augmented Generation
  • [This slide only had a diagram]

[mark-bell-tna]

Another important aspect of LLMs is fine tuning which I think is the area the humanities could have a big impact by curating datasets for improving model behaviour. We cover some of the issues around data in the ethics section but perhaps we could do a separate mini-episode about its importance in ML based systems.
We also discussed adding something around the ethical considerations of LLMs which are outside of what is already covered.