3 ms·

SEEKING FREELANCER | Remote Refine and iterate on gpt agents. An example project - prove that it's impossible to make a GPT agent that can do the following ta

by tikkun 3y ago

SEEKING FREELANCER | Remote Refine and iterate on gpt agents. An example project - prove that it's impossible to make a GPT agent that can do the following task: Please make me a table of all quest diagnostics locations within 10 miles of [zip, city], as well as the yelp review scores for each location. Do it only via browsing, no APIs. But the constraints: No API usage (other than google search API), no hard coding of anything to do with quest diagnostics or yelp. Should be a general browser agent. Must have no errors in the result. Must complete the task within 10 minutes. Inspiration: * BabyAGI (seems to make progress, but doesn't finish the task within 30+ minutes, and partial progress shows the list of locations isn't quite right. Doesn't succeed with GPT-4 or GPT-3.5) * Auto-GPT (doesn't succeed with GPT-4 or GPT-3.5 and doesn't seem promising) * Amoni (fails with an error) * A human doing it manually (flawless, does the whole thing in a few minutes) But the steps the human takes to do it manually seem to be things that GPT should be able to figure out. Your goal would be to prove that it's impossible (and hopefully to make it work in the process!).

tikkun 3y ago
To apply, send 2 things: 1. Links to the 1-2 solo projects you've made that you refined the most over the longest time / most iterations 2. How would you approach proving whether it's currently impossible to make a version of BabyAGI that can succeed at the task mentioned? Preferably, send these by June 3. But if it's past that time, please still send it in, as I'm hiring multiple freelancers on a rolling basis. Send to [email redacted] No other info needed, and please be concise in your email. If you already posted in this thread as a freelancer, please still reach out - I rely on people reaching out. Other possible projects: more agent projects, using LLMs to identify and measure the spread of complex ideas online, tools and auto graded competitions to compare LLMs. If you're interested in one of those in particular, include a few bullet points of thoughts you have on them. Communication style if we work together: no phone calls, mostly email for communications. Occasional Loom videos or Otter audio notes. Can do discord though email is easier.
- sgu999 3y ago
  > 2. How would you approach proving whether it's currently impossible to make a version of BabyAGI that can succeed at the task mentioned? What do you mean by "proving"? Proving anything on the behaviour of a non trivial neural net is not a small endeavour, and that's assuming you have access to the weights. Are you expecting for someone to test all the currently known tricks on a BabyAGI-like setup to know if that's enough to achieve your task? That's likely enough to tell you that it's not doable with a given budget, but that's not a proof.
  - tikkun 3y ago
    > Are you expecting for someone to test all the currently known tricks on a BabyAGI-like setup to know if that's enough to achieve your task? That's right. I mean proving in the sense talked about in this HN comment https://news.ycombinator.com/item?id=36083070 https://news.ycombinator.com/item?id=36083070, not in the sense of a mathematical proof.
    - notomorrow 3y ago
      What is your background/budget/size of team to achieve such project?

by tikkun 3y ago

tikkun 3y ago

To apply, send 2 things: 1. Links to the 1-2 solo projects you've made that you refined the most over the longest time / most iterations 2. How would you approach proving whether it's currently impossible to make a version of BabyAGI that can succeed at the task mentioned? Preferably, send these by June 3. But if it's past that time, please still send it in, as I'm hiring multiple freelancers on a rolling basis. Send to [email redacted] No other info needed, and please be concise in your email. If you already posted in this thread as a freelancer, please still reach out - I rely on people reaching out. Other possible projects: more agent projects, using LLMs to identify and measure the spread of complex ideas online, tools and auto graded competitions to compare LLMs. If you're interested in one of those in particular, include a few bullet points of thoughts you have on them. Communication style if we work together: no phone calls, mostly email for communications. Occasional Loom videos or Otter audio notes. Can do discord though email is easier.

sgu999 3y ago
> 2. How would you approach proving whether it's currently impossible to make a version of BabyAGI that can succeed at the task mentioned? What do you mean by "proving"? Proving anything on the behaviour of a non trivial neural net is not a small endeavour, and that's assuming you have access to the weights. Are you expecting for someone to test all the currently known tricks on a BabyAGI-like setup to know if that's enough to achieve your task? That's likely enough to tell you that it's not doable with a given budget, but that's not a proof.
- tikkun 3y ago
  > Are you expecting for someone to test all the currently known tricks on a BabyAGI-like setup to know if that's enough to achieve your task? That's right. I mean proving in the sense talked about in this HN comment https://news.ycombinator.com/item?id=36083070 https://news.ycombinator.com/item?id=36083070, not in the sense of a mathematical proof.
  - notomorrow 3y ago
    What is your background/budget/size of team to achieve such project?