070: Expertise
What happens when the tool can teach its users to use it more effectively?
The ancient parable of “blind experts describing an elephant from different perspectives” feels apt right now when conjuring all things AI. Quite literally, everyone is correct in their narrow opinions and there’s far too much to comprehend.
But the parable would be even better if there was a firehose element.
See? Much better.
As a group of researches put it (emphasis mine) back in July,
“…that such capabilities could be created in hours underlines that we need to start thinking about how LLMs will impact the future of materials science, chemistry, and beyond.”
And by “beyond” I think they meant any topic you desire.
There’s a tsunamic (can I invent a word?) pace with AI that feels very different than the arrival of the Internet, the near endless arrival of mobile, and social. Those new forms of access, connection and publishing absolutely changed culture—we are radically different from ~30 years prior. But those forces and their effects worked more slowly, their evidence accumulating over memes and years, which seem perhaps trivial in this new era. Or as my new friend Ross Patrick puts it,
“We’re teaching each other how to think.”
What feels unprecedented using LLMs is the ability to dialogue with the tool itself, to be able to ask it, “how else can you solve this?” Excel might calculate, but it won’t teach me how or why it did its work. Instagram doesn’t dialogue with its users in the moment of creation. Humans do this, of course, but not the ones I need at 1:14am when I’m busy writing. ChatGPT is happy to oblige at any hour.
Recently I joined a Google Group for AI in Education. Firehoses take many different forms. I’m drawn particularly to examples of engaging LLMs via Socratic dialogue, the Feynman Technique, and other pedagogical methods. The University of Maine offers a robust collection of instruction (pro tip: check Strategies).
Here’s an example of engaging the tool to teach you how to best use the tool.
And this is just one approach. Cut/paste and edit as you see fit. There are other “master prompts” like this one which you can alter to conjure insights from specific industry leaders like maybe choral conducting.
Anyway, here’s how I evolved the task above for a specific prompt…
At this point, I’m not feeling particularly enthusiastic. So we stepped back and interrogated the process a little further.
We’re starting to get somewhere slightly more scintillating.
So we returned to the process defined earlier and tried again but deliberately skipped the socratic dialogue part…
Not bad for 9 minutes of effort.
If a change management assignment rolled through, this could help get the ball rolling. And yes, we’d verify the quotes. More important, in this process, the tool helped me gain a deep understanding of itself.
The elephant describes the elephant?
MORE AI UPDATES
The word on DALL-E3 promises to further shake up generative image making. Can’t wait for the ability to specify text within objects—that will speed up the ad-making experience.
MSFT is revealing further iterations of Copilot. Corporations are about to transform, radically. Hopefully they’re getting the message.
Oh, and we’ll be speaking with ChatGPT soon enough.