Don't Chat with AI!

Attorneys, don't chat with your new AI chatbot. I'm not saying don't use it. I specifically mean don't "chat" with it. Instead, 𝘀𝘁𝗮𝗿𝘁 𝗼𝘃𝗲𝗿.

If you've just begun to use a general purpose LLM chatbot (either ChatGPT, Claude, etc. or legal-specific ones, CoCounsel, Harvey, whatever), it can be tempting to engage with it the same way you might with a younger attorney or new paralegal. You give it an assignment. It does a good job, but not a great one. You attempt to talk to to it, to give it "feedback" to help it improve.

Great thing to do with a younger colleague; potentially massive waste of time with an AI.

Unlike humans, AIs tend to get worse as the conversation gets longer. They forget things you told them at the beginning, they undo good work they did before, they decide to take their own path. (OK, maybe some humans are like that too.)

So what can you do? You can 𝘀𝘁𝗮𝗿𝘁 𝗼𝘃𝗲𝗿. Yep, unlike humans, LLMs don't find it demoralizing or exasperating to be asked to start over from scratch and try a task a completely different way. They just do it, and fast. And that's often the key to getting better work product out of them. Don't continue the conversation; start a new one and ask them to do the task over again.

Here are three specific ways to "𝘀𝘁𝗮𝗿𝘁 𝗼𝘃𝗲𝗿" with an AI to try to get better results:

☑️ 𝗥𝗲𝗽𝗵𝗿𝗮𝘀𝗲: Open a new chat and just ask the question a different way. I wouldn't say this is a high percentage strategy, but it's low cost. Worth trying at least once.

☑️ 𝗠𝗼𝗿𝗲 𝗜𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻: Give the AI more information to help it complete the task: a good example, a website to read, some documents with background information. Called "context" in the LLM world, the information you give the AI as part of your question is probably going to have the biggest influence on your results.

☑️ 𝗦𝗺𝗮𝗹𝗹𝗲𝗿 𝗧𝗮𝘀𝗸𝘀: Try to break your task up into multiple smaller tasks. For example, don't ask it to "summarize the [very long document]." Instead ask it first to "outline the major sections of the document," then ask it to summarize a particular section, etc. You get the idea. This helps to cut down on the AI's tendency to go off and do its own thing.

You'll find other ways as you get more familiar with using AI, but the key point is what I started with -- don't get into long discussions with it. As soon as the discussion starts to go off the rails or gets frustrating, 𝘀𝘁𝗮𝗿𝘁 𝗼𝘃𝗲𝗿!