<aside> 🧂

This page reflects my ongoing thoughts about where AI/ML technologies best fit into a user experience or market need to solve a problem. Take with enough grains of salt to cook your pasta.

</aside>

Searching

This feels like the OG use case for AI, which has been accelerated by RAG and GenAI technologies. I see searching use cases and I think of chatbots and contact center software that leverages AI to pull data from a variety of sources to provide the best answer at the right time.

Synthesizing

Low risk since people may just discard the information or read on for more detail. This can also help with creating transparency and interpretability in highly technical pieces of data, like SQL queries, for non-technically-minded individuals.

Some examples:

• Alltrails AI-generated synopsis on trail reviews to quickly surface the most helpful information to the hiker.
• Stripe’s KYC (Know your customer) tool that scrapes payer websites and synthesizes them for a customer team to review. They also weigh in on whether they suspect the site to be a fraudulent site or not, with some significant success (more of a Critiquing feature).

Critiquing

These use cases involve using AI to challenge ideas and find the faults. A personal sounding board for new ideas and idea formation, based on large unstructured data like call transcripts or survey open question responses.

Note: LLMs still haven’t been great at finding key trends without layers of synthesis. Some tips:

• Put the data through multiple “washes” of data - maybe categorize the responses by sentiment or topic, then conduct more analysis through the GPT
• Make sure to prompt that the GPT needs to preserve exact quotes that demonstrate a problem or impact

Creating

AI tools have been extremely useful for their ability to generate something different. It does best with a lot of context, but it can help launch through a lot of the busy work to get something produced.

Note: Content can sometimes come across as feeling useful and more information-dense, but not exciting or different. But also sometimes feeling off in ways people can't quite describe. Like a friend who you'd grab a beer with but wouldn't invite to the wedding. Make sure that new creations are screened back by a human after creating.

Cross Media understanding

Interpreting Audio to video to text and back again. There’s been a lot of, frankly, busy work involved to-date in taking data (like a conversation) and then making a presentation out of it, or taking meeting notes and converting it into a document. AI can act as a catalyst (not the final product!) to taking information and switching the medium or format.

There’s still a lot of issues in this, especially in the act of transcribing conversations, which tends to overly hallucinate extra information.