I was trying to find some info on startup valuations this afternoon. I kept prompting GPT-4, but I couldn’t get the right answer. Turns out, I should’ve let the AI prompt itself.
Or at least that’s the conclusion of a team from VMware. From IEEE Spectrum:
There is an alternative to the trial-and-error style prompt engineering that yielded such inconsistent results: Ask the language model to devise its own optimal prompt. Recently, new tools have been developed to automate this process. Given a few examples and a quantitative success metric, these tools will iteratively find the optimal phrase to feed into the LLM. [VMware engineer Rick] Battle and his collaborators found that in almost every case, this automatically generated prompt did better than the best prompt found through trial-and-error. And, the process was much faster, a couple of hours rather than several days of searching.
Teams from Stanford and Google DeepMind are behind those cutting edge tools to automate prompting. And they’re not alone.
In the Claude 3 Opus as an Economic Analyst demo we covered on the blog yesterday, Claude repeatedly prompts itself.
When I saw the “DISPATCH_SUBAGENTS,” my eyes opened really wide. How long until we don’t say “Jim in accounting, please do this” or even “Claude, please do this” but rather “Claude, prompt yourself in an ideal way in order to do this”?
In time, Claude will probably have auto-prompting built in. So will every other major AI model.
At that point, reinforcement learning becomes the key.
If we can label Claude’s outputs as good or bad, Claude can refine its prompts. Then, we’ll get even better results.
As the models get more and more powerful, Claude may need only a little reinforcement learning to improve drastically. And in time, maybe it will understand our goals so well that it even does the reinforcement learning itself too!
I’d love to tell you I know where all this is going.
I sure would seem smart, wouldn’t I? And we investors love that.
But I don’t know.
All I can do is carefully follow what’s going on, talk to great people, use the products, and think. And you can bet, I’ll be doing that.
Where do you think AI is headed? Leave a comment and let us know!
If you enjoyed this post, subscribe for more like this!
More on tech:
Why I’m Learning to Love Consumer Subscriptions
Save Money on Stuff I Use:
This platform lets me diversify my real estate investments so I’m not too exposed to any one market. I’ve invested since 2018 with great returns.
More on Fundrise in this post.
If you decide to invest in Fundrise, you can use this link to get $100 in free bonus shares!
I’ve used Misfits for years, and it never disappoints! Every fruit and vegetable is organic, super fresh, and packed with flavor!
Use this link to sign up and you’ll save $15 on your first order.
Tremendous 
Leave a comment