16,726 karma · joined February 21, 2007
On HN mostly known for Candy Japan, which I shut down, and am now making Roblox games full time. Main game currently is https://www.roblox.com/games/7367863399
One day two salespeople contacted me and suggested taking over ad sales, and managed to sell thousands (I think it was somewhere between $5k-$10k/month, but it was a long time ago) a month in ads for the site. This was for a city with a population of 200k.
Sometimes it turns out the task I had been dreading can actually be mostly completed in that time, and I was just wrong about it being hard.
iOS app "Friday" lets you use to talk to ChatGPT. It seems to be just simple glue (I'm sure there are many similar ones) between speech recognition, GPT, and text-to-speech, but the end result is that when you're bored you can have fun discussions without typing.
I also have a list of ideas, with hundreds of entries on it. As I've gotten older, the amount of time I can put into one without becoming demotivated seems to have increased. Used to be a weekend, but now I can now put ~2 months into one without knowing if anyone is interested in it.
To keep going past that limit, it helps if the idea has a shape such that a very simple version can be quickly released, but then extended. Once you have any minor traction at all, the external validation makes it way easier to remain motivated.
Used to play almost every night when I got Cosmodread, but managed to complete it.
There are thousands of videos it could put in these slots, so they'll be ranked and only the very best (based on metrics and your interests) will make the cut.
Video #11 might have been great, but wouldn't get suggested to you merely because it had a few seconds lower average watch time. Because of this I'd expect a 10% better video (like %, watch time) would improve its views by some huge nonlinear factor (maybe double?).
Side prediction: Echo VR getting shut down in August actually means it will become an activity inside Horizons instead.
Goals: make predictable progress each day, get a feel for how long different types of tasks take, and to have guilt-free rest after a day is “done”.
So I decided I’ll aim to work just 3 hours on weekdays, 2 hours on weekends, but I must first write down my current subgoal in a text file and note every time I stop or start.
So this file is full of things like “15:03-15:20, 15:55-16:00 Make minigun emit a puff of steam when it stops”. Conveniently reused as a git commit message. GPT made me a sublime text plugin that adds up the total, and I have inserting current clocktime as a mey binding.
I’m only allowed to work towards that subgoal as fast as I can with no distractions.
The goal must lead to a releasable update, otherwise it does not count. So in my system I’ve defined doing taxes or checking email as not work but a chore (like washing the dishes), to make sure I prioritize releasing stuff.
If I want to check HN or something, then I need to take a proper break instead of semi-working. And since I need to confront this and note it, I’m less likely to do so.
You don't want to start speaking before you get a full sentence, or the TTS will sound weird. You don't want to wait for a full sentence, because it would be too slow.
Besides that, speech recognition wasn't working well enough, and I heard (haven't tried) Whisper has the same delay problem where you again cannot stream it, but must wait for the user to have spoken for a while.
They are prompting it with guidelines that specifically ask it to use smileys and be interesting.
They must know that these articles actually get people curious to try it. I found myself installing the Edge browser to play with Bing after seeing them, and I’m probably not alone in that.
"We're reimagining how people search for, explore and interact with information, making it more natural and intuitive than ever before to find what you need."
Whether or not this tool works, expect it to make the rounds on any tech publications that have previously written “stable diffusion bad” -type stories.
It took a lot of iteration as initially it wouldn’t for example put a ground in the game, so you’d just fall forever, but one by one pointing out the mistakes actually resulted in a playable game.
With that it would be less effort to create fun and useful little apps without needing to charge the user directly.
For instance they could replace human content moderators, which is costing them billions. Maybe offer some "Meta premium" which summarizes and lets you ask questions about your WhatsApp, messenger chats. Perhaps improve ad targeting by understanding user post content better?
Although clicking on them might not be quite as natural as clicking as part of search results, so still massive damage to the golden goose.
"Authorization": "Bearer YOUR API KEY HERE"