how are you going to spend only 30
how are you going to spend only 30 seconds per email??
18 Replies
putting this in a thread to avoid sounding too markety.
we've been building an MTurk compatible API alternative to help with use-cases like this. 😛
it's not too hard to use. You basically combine:
1. User and API activity
2. Previous messages/interactions
3. New features
Throw that all at GPT-4, and use our MTurk compatible endpoint to have a human do a quick proofread and final edit before hitting send.
Ahh okay. Making more sense now
Is this your product?
yep. We were going to keep it a secret to power our other in-progress projects and not release it. Kinda high risk to do it that way though, since we'd be without revenue for another ~6 months.
the nice thing is I don't really care if many other people use it, because that would give us an advantage for our other products if competitors don't use it.
if other competitors start to use it a lot to beat us at original plans and we end up sucking at everything else, well, at least now we have a platform everyone uses. 
clarify, we currently do the "MTurk compatible endpoint" part. You bring your other pieces and your staff to solve the tasks. 🙂
clarify, we currently do the "MTurk compatible endpoint" part. You bring your other pieces and your staff to solve the tasks. 🙂ah gotcha, so it's more abstract for any task?
exactly. The reason we used MTurk's API, is because we can leverage all the existing tools and developer knowledge on it. e.g. for ourselves, we're in the process of adding support to Meta's Mephisto tool (which already supports MTurk). https://mephisto.ai/docs/guides/quickstart/#optional-set-up-mturk
You can use https://parl.ai/ with Mephisto. 🙂
10-minute Quickstart | Mephisto
First, clone this repo to your local system.
I've never used Mturk so I understand the concept but not the pain points. do you have any concerns about getting traction if you go too broad?
The main pain points for us are:
1. I don’t trust someone making $3-$5/hr with important data like emails or any sensitive data. e.g. this ends up happening. https://www.technologyreview.com/2022/12/19/1065306/roomba-irobot-robot-vacuums-artificial-intelligence-training-data-privacy/
2. The pricing of MTurk isn’t great. 20% cut and fees per task.
3. You have nearly no chance of getting highly skilled people to help. (So if your task is something that requires real experts, good luck.)
None of these seem like a uniquely “us” problem. Some companies might not care about leaking private data, so #2 would be the only interesting part for them.
MIT Technology Review
A Roomba recorded a woman on the toilet. How did screenshots end up...
Robot vacuum companies say your images are safe, but a sprawling global supply chain for data from our devices creates risk.
Doing an MTurk API doesn’t seem broad at all, I thought it was too narrow if anything? 😅
for getting initial traction it could be - do you have specific use cases to very specific audiences that you will be marketing or that are built in?
good thread from Berry yesterday about this https://twitter.com/nathanbarry/status/1659566137179537408?s=20
Nathan Barry (@nathanbarry)
My SaaS business hit $1,500 MRR and then got stuck.
Until I learned an important trick…
Here’s the exact method I used to scale to $100k MRR in the next 12 months:
Likes
1072
Twitter
thanks!
Likely going to wait until we have the MVP with all the features we need internally first, then more external audience targeting. 🙂
it's so verrrrrrrrrry alpha but I didn't want to lose out on signups with press release stuff happening this morning. 😛
https://www.ai.moda/en/blog-post/646af91e0d214400017219e9
Apologies for the blog formatting.. we never tested the blog stuff much until tonight.
ai.moda
Introducing YourCrowd: a private MTurk-compatible crowdsourcing alternative. Maintain full control over your data while leveraging your established crowd.
it is blank for me also what in the SEO worst practices is up with your URL?
Ughhhhhhhhhh race conditions..
yeah I didn’t spot the url thing until tonight
Haha that’s intentional. They’re all being fetched with low priority.
I've never seen so many response headers in my life 🤣
This is what happens when the browser security guy (me) has fun with headers.
I use some of them for lazy bot detection.
Ah, our webdev who handles the blog part of the site is currently under airstrikes again, didn't know about it until he lost electricity a few minutes ago.