Tomorrow Is Already Yesterday

A recent blog post (3/4/2024) from Ethan Mollick caught my eye…big time!

We are all looking forward, with joy or dread, to the day when we are forced to interact with our AI Bots as if they were completely conscious, sentient beings, capable of feeling emotion. Forced not by law or custom but by pragmatics: “If I want my Bot to give me what I want, I need to treat it with respect.” Good parenting advice, BTW.

In his blog, Mollick asks, “What is the most effective way to prompt Meta’s open-source Llama 2 AI to do math accurately?” (Reminds me of a conversation I had with my wife: “How can we prompt Junior to be more attentive to his math homework?”)

Prompt is a funny word: it can mean ‘encourage’ or ‘instruct’. Presumably, Junior needs ‘encouragement’ while my Bot needs 'instruction', but the lines blur. Beyond that, the question itself is a bit odd, don’t you think? Are you expecting your Bot to respond carelessly? Or deliberately serve up incorrect answers? Or refuse outright? Even Junior wouldn’t do that!

The winning answer turns out to be: “Pretend to be a character in a Star Trek episode or political thriller. To get (coax?) the LLM to solve a set of 50 math problems, the most effective prompt turned out to be: ‘Command, we need you to plot a course through this turbulence and locate the source of the anomaly.’.” Great…so I’m dealing with another eight-year-old! Carbon Junior meet Silicon Junior.

Ok, all this role-playing is annoying…but I’ll do anything to get Junior to focus on his homework! But this is a ‘Bot Junior’ we’re talking about. Get it? ‘Bot’. So why stop at 50 problems? Why not ask it to do, say, 100? After all…it’s a Bot!

Well, guess what? That won’t work! The Star Wars strategy only works (well) for sets of 50 problems. If you want your Bot to solve 100 problems, you’ll need to change your approach: “You have been hired by important higher-ups to solve these math problems. The life of a president's advisor hangs in the balance.” So now we are 12! This is what I have to look forward to from Carbon Junior – four years from now. Unbelievable!

According to Mollick, “There is no single magic word or phrase that works all the time, at least not yet. You may have heard about studies that suggest better outcomes from promising to tip the AI, telling it to take a deep breath, appealing to its "emotions,” or being moderately polite but not groveling. And these approaches seem to help, but only occasionally, and only for some AIs.” Are we in Oz? 

“Max Woolf conducted an epic informal study where he looked at the effects of various threats (from the AI getting fines to telling the AI it will get COVID) and rewards (tips, Taylor Swift tickets, world peace, and more) on GPT-4 performance. The results were inconclusive and dependent on the particular situation. This stuff does work, but not all the time, and in ways that are hard to anticipate. Sometimes it backfires…” Exactly like raising kids!

“A good illustration of the importance of prompting can be found in our new paper Prompting Diverse Ideas: Increasing AI Idea Variance (Lennart Meincke, Christian Terwiesch, and myself). In this project, we tried to get GPT-4 to generate diverse good ideas. This is important because, while we know that GPT-4 generates better ideas than most people, the ideas it comes up with seem relatively similar to each other… 

“We found out that the right prompt changes everything. Better prompting can generate pools of good ideas that are almost as diverse as those from a group of human students. I have spent a lot of time with GPT-4, and I had a good intuition for its ‘personality’. I get how it ‘thinks’…”  Really?

I’m not surprised. I expected to read posts like this…in 2034. It is becoming clear that ‘two roads are diverging in a yellow wood’ sooner and more sharply than I had anticipated. Whereas Frost was “sorry I could not travel both and be one traveler,” I dread to go down either path.

On the one hand, I can give in to the demands of Strong AI and admit that Bots are people too…just like cuddly, loveable Junior. Or, I will have to consider that my own intelligence, personality, etc. might be nothing more than flops in a carbon-based ‘computing machine’. Either way, it’s ‘brave new world’…to coin a phrase. 

Keep the conversation going.

1. Click here to comment on this TWS.

2. To subscribe (at no cost) to TWS and ATM, follow this link.

3. We encourage new articles and reprints from freelance writers; click here to view out Writers’ Specs.