Feels good doesn't it? Now that you've caught the dev bug, what are you gonna work on next?

Interessante uso do Codex da OpenAI. No entanto, é importante lembrar que a dependência de IA centralizada pode comprometer a privacidade e a segurança.

I'm gonna work on getting a job so I can afford the tokens 😂

Congrats - I'm still only just getting used to github. Never done a pull request or anything, yet. Curious about your WoT relay - atm I'm running my personal haven relay and a strfy relay for limited public use, but that's not working, so I'm considering getting rid. I'd be tempted to replace it with this, just curious how the WoT bit works.

I'm getting my money's worth out of that $200/mo pro subscription for OpenAI.

You've come to the wrong place for an explanation, but I'll give you a shitty one anyway 😂 It uses your follows and your follows follows etc... to build a Web of Trust map, which it regenerates by default every 4 hours. Based on the level of trust you define, it only allows posts that fall within your defined level. Any posts outside this it rejects. This should pretty much eliminate spam, but it also means that anybody within your defined WoT can write to this relay, but nobody outside of it can. For me, I hit the limits straight away, which is 40,000 trusted npubs and 1,000 relays. That doesn't mean it included all npubs and relays because of hitting this limit. But I follow a lot of npubs, so it probably doesn't hit those limits for most. I could increase those defaults, but as I'm only running on a $6 VPS, I don't wish to tax it too hard. I'm still learning, so I'll be posting here as I go.

I feel I may be upgrading my £20 / month plus subscription pretty soon.

I did most of my first project just passing files to ChatGPT in the browser with the $20 subscription. Then I started running codex and discovered that I would burn through my daily usage limit in about an hour of work. The $200 subscription gives you 20x usage. I've hit my weekly limit on that a couple of times usually the day before it resets. I had a lot of shit going on in my personal life so there were a couple of weeks I couldn't focus on coding at all but I'm back to it now.

When you start doing heavy workloads with a lot of context that usage limit goes fast. The AI also cuts a lot of corners if you don't force thoroughness. I manage in AGENTS.md that it must read all files end to end before editing, not just grep snippets, and it has to review them all end to end after editing and run tests. I'm also running it on xhigh fast which burns through the usage limit a lot faster but does better work. I have a dialogue with ChatGPT in the browser about the design intent of the project and then ChatGPT writes detailed prompts for me to feed to codex.

Yes, I chat to Chatty, (ChatGPT) about how to optimise Codex. Even on my two small projects, both context windows have been compacted at least once during the workflow.

May just be simpler to figure out what I've done wrong with the strfy relay. It's odd, because I had to install a third relay, for testing chronicle and that works fine. I think my filter is probably fucked up.