Enjoy having your code be trained on and stored and inspected
And you think none of these companies don’t do this lol? This is the trade-offs. Unless self-hosted models get to a stage where the reasoning and output is on par with some of these hosted models, anything you send can be stored, used to train and inspected. How do you think llms reasoning got to this stage so quickly?
I think there’s a much higher probability of a foreign company from a country rife with industrial espionage who explicitly has it in their terms to use it in a way I wouldn’t like than an american company who explicitly excludes it, yes. Seems pretty simple to me.
How do you think llms reasoning got to this stage so quickly?
Seems like you don’t know how much money they’re spending on people to write good code datasets or how RL guided COT with verifiers works or that in standard chatgpt there are explicitly no such guarantees?
oh, you mean deepseek? yes, I agree but ultimately, it’s down to the end user. I bet majority don’t really care I definitely don’t especially when equated to costs. I use deepseek for other non-critical stuff. Topped it up with $10 a month later it’s still $9.02. Try this with claude api and come back. The quality of the responses was sometimes even better than Claude’s. so having the option is good.
well, yes. It was initially RHLF i remember a year ago now they just use synthetic data from output the model themselves generate. they don’t necessarily need your code.
No, I’m not talking about RLHF. Looks up how reasoning models are trained these days. And I agree users are free to opt in but it should never be a standard model for an enterprise application and it should come with some kind of disclaimer
Echoing lots of the sentiment said already.
I’ve found 3.5 Sonnet to be completely unusable on slow request traffic, to the point now where I’ve just given up using it as it’s just not responding. Using another model isn’t a viable alternative for my requirements, as 3.5 sonnet is just so much better than other models at coding tasks.
I can completely appreciate that there’s a huge demand for Claude 3.5 Sonnet, but I signed up for Cursor to use this specific model, and despite being a pro user that’s run out of premium request, I am unable to use the service how I previously did, which is a big shame as I really have enjoyed using Cursor over the last couple of months.
I will likely come back to using this service if these issues are fixed, but in the current state of things, I’m going to sign up to another competitor.
The CHAT tab is working pretty fine.
It seems COMPOSER tab is the problem, or the agent.
1 Like
I have the same problem, it keeps me waiting for 5-10 minutes, it’s been like this since the night!
1 Like
Subscription cancelled here. I’ll maybe come back when I can use composer more often than once every 10 mins.
3 Likes