Build a simple browser agent as proof-of-concept for the new function calling API.
1. Enter URL 2. Tell it what you're looking for
It will browse the site, click on any links, and return the results.
This is all you need to provide to the API.
You then need to watch for any "function_call" response, run the actual function yourself, and send back the results to the API.
I didn't record this session, but this was one magical. It used the search feature to find Andrey's profile. I saw it live tapping each search result page until it find the list of users, clicked Andrey's profile, and summarized his recent work. 🤯
This isn't new. ChatGPT has this built in with their browser functionality.
What is new, is how easy it now is for developers to implement similar functionality.
Having GPT control a browser is just the first step. There's *so many* options here.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Some thoughts on originality, competition, and one of my biggest weaknesses as an entrepreneur…
In 2010, I launched BetaList because I recognized how difficult it was for founders to get publicity, particularly for those outside of the Silicon Valley bubble.
It turned out to be my first successful product.
Product Hunt launched a few years later. A similar product with a different take: Reddit-style upvotes.
Combined with their access to SV influencers, funding, and talented team, they quickly took over the top spot in the "daily list of new products" category.
I never shared this before publicly, but I owned @WIP on Instagram (probably worth $10k+) for about a minute.
I figured with all the recent handle talk, now would be a good time to share the crazy story…
June 2018 I was considering getting more active on Instagram with WIP. I owned @wipchat there, but I thought it would be much nicer to get @wip instead.
So I checked who currently owns, to see if there's *any* chance I could somehow get.
I was expecting it to be either suspended or used by some big brand. Instead, it was just a regular person with a bunch of family photos.
Turns out it was owned by a Dutch guy that lived in the small city (200k residents) I'm from. Odds of that happening are less than 0.1% 🤯
You might be able to copy a (successful) product, but you probably can't copy the conditions that made it succesful.
So stop copying other products thinking you'll get the same success.
Study how they learned what to build given the conditions at the time. Copy that process.
Example:
Everyone building AI avatar apps today. You're too late. The time was a month ago when the technology was accessible enough for developers, but not yet mainstream.
You can copy @levelsio/@dannypostmaa's product, but you can't copy last month's conditions.
Instead, study their process and copy that. My take-aways:
- Have your distribution channel ready (e.g. grow a following)
- Boil down a product idea to its bare essentials
- Ship a lot of different ideas cuz you can't predict a hit
- Only ideas that can be shipped/tested quickly