LeicaLatte

I got out of GitHub last year and self hosted a vps that runs gitea.

My work llm provider is GitHub copilot and we are completely blocked since morning. My question is who’s comparable to copilot vscode and cli. Multi model is table stakes.

If china captures the market now, well deserved. Way cheaper compared to us providers.

My setup is very similar.

After you log in you can unlock keychain by running this command

‘security unlock-keychain’

Completion API is only half the product. Teams integrate these tools into production with nothing but policy documents between them & incidents. The guardrail system does not exist for a reason. No business model. No investors. No revenue stream.

Tesla didn't just build FSD. They had to build a verification layer: driver cameras, steering sensors, attention alerts. That parallel software makes autopilot road ready.

Instead of reinventing terminals, what if we standardized on structured data exchange between CLI tools (like JSON/msgpack) and let terminal emulators render it intelligently? The backwards compatibility problem that nixpulvis mentions is real - any "terminal of the future" needs to run existing tools well first while gradually enabling richer experiences.

A simple "END OF ARTICLE" delimiter in prompts might prevent most of these embarrassing slips.

Great timing - launching right as the transformer bubble might be peaking.

YouTube's messaging is the more frustrating part about all this. Panic might drive more creators toward direct monetization, that might just be the better net outcome.

Mobile apps often sort by creation time in the UI (chat messages, activity feeds). Since clients only see the masked version, there might be a need to expose a separate timestamp field.

I think the 73% non-work usage ratio will flip again within 2-3 years, but not because consumer usage shrinks. As AI becomes embedded in workflows through APIs the "work" category is set to expand dramatically.

Curious about the interaction between this memory behavior and fine-tuning. If the base model has these emergent memory patterns, how do they transfer or adapt when we fine-tune for specific domains?

Has anyone experimented with deliberately structuring prompts to take advantage of these memory patterns?

Liking the package management from first principles as a systems-level optimization problem rather than file scripting. resembling a database engine - dependency aware task scheduling, cache locality, sys call overhead - they are all there.

Very cool. I would probably build this using SDUI server driven abstractions. Curious if you considered that approach to build bitrig or parts of it.

I just hate arrays.

Loving all these Claude Code as a Platform products. Hope Anthropic doesn't ban them once they become popular citing some vague TOS violation.