Hacker News Reader

Comment by dcre | Hacker News Reader

I don't see how things like Cursor[0] (a VS Code fork with a nicer Codex UI) have a moat or chance of profitability in the face of things like this. If you can run models locally or your employer can spin up their own instance for internal use on a decent server, why is anybody paying OpenAI? And that leaves Cursor with no cash flow to skim off of.

[0]: https://www.cursor.so/

View on HN

You can't run anything even remotely close to GPT-4 locally.

Large enterprises can run a private instance of it, but OpenAI is paid for that (indirectly via Microsoft).

Their value proposition appears to be a fancy UI. Better/cheaper models would only improve their product and grow their addressable customer base.

Anybody can copy that UI once they figure out what works. No moat.

I wouldn't discount good UI.

I don't use Cursor, but consider that GPT-3 sat around for months and nobody was really talking about "AI doom" and "the lightcone" until OpenAI put a good chat UI layer over it.

I would argue that actually capturing the value from these models lies mostly in the UI: allowing the user to seamlessly and quickly extract useful information from the model.

Chat generally being a nice way to interact with the model is one thing, but if anything this proves my point: there are now dozens of little web UIs on GitHub that are as good as the official one or better. You put in an API key and you have fully recreated their UI. I even wrote a CLI[0] in 130 lines of Deno that I prefer over the web UI. People are trying to charge for them but I don't think any will succeed.

[0]: https://gist.github.com/david-crespo/d9dbefe5a50c0f0da9ac3de...

Are you paying OpenAI? I’m not. Facebook’s midterm game on “massive” AI seems to be “commoditize the competition” and ATM it looks like a good bet.

I am paying a couple bucks a month for GPT-4 API calls, but not for Copilot usage.

there's overhead to setting up your own local LLM instance (by that I mean both resource utilization and the setup process itself). so long as this continues (the setup portion at least), there will be people not willing to invest the time and rather pay cursor.

Reply on news.ycombinator.com