I find "big tedious refractors" using coding agents to be a either the best way to spend your tokens or a quagmire that's going to eat up your whole day. 50/50 on which one it'll be.

Less true today but LLMs are still worse at modifying existing code than producing new whole files.

Icon!

yes but maybe it does enough new things to worth a gander?

as a spacesim junkie, I'd avoided this so far as it seemed to be a modern reskin of freelancer (arguably, the best of the genre) which I've played until revulsion. Have you played Freelancer? Should I try Underspace still? Does it do something new/interesting?

though I suppose I'm not really responding to the "repair" or "community notes" part of your proposition. Intriguing pearl

btw addons.mozilla.org/en-US/firefo...

I get the impetus but...you'll probably want to have a sharded spaces for discussions. Each subreddit or alternate implementations like hackernews are contexts that frame the discussion. Not sure how well it'll work without sharding considering moderation, usefulness and all that.

this is just reddit

People keep blaming the user base but I think this kind of behavior is the outcome of the medium. This isn't any much different than what went on on Twitter or what's going on now (just that the politics are different).

I had the same issues. Question is, is there something special on BG2 that one should consider skipping BG1 for it?

This was just what I was saying. None of the political channels that I'd wasted hours into showed up and yet, that one video of Top Gear (I checked) I've seen puts them in my top 5.

Nonsense, what limitations do you think needed to be highlighted that would have improved the message? Or maybe can part on us your secret knowledge that would redeem it?

I've had those before, annoying little buggers. I keep trying to bite them off but it always ends in tears

People are mad about AI everywhere in my experience, including tech forums.

Cool idea but to click on that tangent, what were the issues of SOLID? Been a minute since I've taken a look.

Hmm, will have to look into that (and sorry, my tired ass neglected to read the full thread again). Small question, did you try this fine-tuning on SOTA offerings from OpenAI/Anthropic or is it on the self-hosting friendly, small models? Trying to gauge the expected cost before I rec this to team.

By LLMs, you mean machine learning? Are Hive AI doing gen AI? We tried moderation with off the shelf LLMs at $dayjob and it didn't go well. Not to mention how slow it was.

People with high visibility are by their nature special, especially when your service is about social graphs. It will create "mass resentment" if your judgment is off in applying the rules to them. What's right in most cases depends on context and not necessarily the minutia of the TOS.