Skip to docs navigation

02.02 Use Large Language Models for Ideation (as of Fall 2023)

What are LLMs?

Large Language Models (LLMs) A LLM is a type of artificial intelligence (AI) that can process and generate human language. LLMs are trained on massive datasets of text and code, which allows them to learn the patterns and relationships between words. These became part of the public consciousness with OpenAi’s ChatGPT, Google’s Bard, and Meta’s Llama 2.

Although using an LLM can be quirky now, it is hard to imagine a future where generative AI technology such as LLM’s will not be more intertwined will all human activity including creative activities.

Quick question, do you use a movable type press to print out documents? Do you type on a typewriter? … Of course not, unless you have a specific conceptual reason to avoid a technology there is little reason to not use new tools to become more productive and more creative.

In a few months to a few years, the idea that one would not use generative AI tools as well as currently unknown AI tools will likely seem strange. Therefore, it is best to learn and grow with a new technology.

LLMs for Content Summary

LLMs can be used to create summaries of long documents, such as research papers, books, or artist statements. While this can be a time-consuming and labor-intensive task for humans, LLMs can do it quickly and efficiently. LLMs can be used to create personalized summaries for individual users. This can be done by taking into account the user’s interests, preferences, and background knowledge. Another way is to train an LLM on a specific data set unique to the user. Training LLMs is out of the scope of this course module but it will likely become commonplace in the near future.

Since LLMs can be used to extract key information from text, such as the main points, supporting evidence, and conclusions, we can potentially leverage the power of LLMs to go through documents and texts we have generated to find hidden connections we may not be aware of. As long as you stay within the token or prompt limit, you can feed an LLM multiple documents. For example, you can feed it multiple versions of your artist statement, a written report you made, other notes and docuemnts from your computer and sketchbook. After this information is in the “chat” you can talk to the LLM about it.

LLMs for Ideation

A September 9, 2023 article in the Wall Street Journal titled, “M.B.A. Students vs. ChatGPT: Who Comes Up With More Innovative Ideas?”external link (paywall) archive of articleexternal link recounts research into the ability of LLMs to generate innovative ideas. On three measures, the amount of ideas, overall quality of the ideas, and the amount of really amazing ideas, GPT 4 beat the M.B.A. students. In a discussion of the article on Y-Combinator there is open debate about how soon the same will be true for computer coding.

Right now there is an opportunity to enhance your creativity, ideation, and thinking by using generative AI. It is not a replacement for human thought yet, but it is now good enough to be useful.

LLMs Are Good/Bad?

Everything about generative AI is sunshine and rainbows, except that it is not. LLMs operate on a statistical prediction model and can hallucinate, make stuff up, or even tell lies with confidence. This could be dangerous if one seeks advice from a large language model about what mushrooms are edible in the woods. In a recent court filing, a lawyer was scolded by a judge for using a LLM and submitting its output, including made up court cases, to the court without fact checking it. Caution is to be applied to the use of generative AI but that is no reason to stick one’s head in the sand.

Some are not heeding the advice of caution such as the online site Gizmodo that fired its Spanish language staff and replaced it with a LLM. 1 The unethical nature of this behavior aside, it already is creating article translations that are inaccurate and switch languages in the middle.

Large Language Model Ideation Exercise

Choose a LLM of your choice and try asking it questions about art. See how the conversation goes. Ask it for new ideas. It will likely not be very helpful without giving it some information about yourself or your art. Use the information from the previous Creative Practice Awareness Exercise and feed it to the LLM.

Ideas to try

Ask the LLM:

  • to summarize your writing
  • write 5 questions about your art
  • write 5 followup questions to those questions
  • give 5 reasons why you should NOT answer the questions
  • generate new ideas based on your input
  • rewrite those generated ideas with more detail included materials used, dimensions, concept and more
  • generate 5 new ideas unrelated to the previous ideas

Remember that the LLM is not “thinking” the way humans do yet. It may give the illusion of responding in an intelligent way but it is just predicting the next likely text. NEvertheless, since it was trained on so much human text, this predictive power can be incredibly useful. It is sometimes a good idea to ask a LLM to summarize the current “chat”. This will through all of the text from the conversation back into the LLM and can lead to new insights.

Bard and ChatGPT Videos