Stephen Caudill
@mrmemes.eth
I’m pretty sure Cursor has severely limited the context they’re sending in their latest version to save on tokens. It’s gone from generally pretty good using Claude Sonnet 3.7 or 3.5 to downright frustrating over the last few days.
1 reply
0 recast
1 reaction
🗿
@bias
Is there any way to analyze traffic or whatever to get any visibility on whether or not there’s truth to this?
1 reply
0 recast
1 reaction
Stephen Caudill
@mrmemes.eth
You could use Wireshark or similar to find out what they're sending now, then you could diff the payload with an older version that definitely didn't have this behavior, like the 0.4.5 series to see. I don't have the time for that though, and the forums are full of similar anecdotes: https://forum.cursor.com/
1 reply
0 recast
1 reaction
🗿
@bias
Pretty sleezy, dishonest and amateur of them if it’s the truth. 😔
1 reply
0 recast
1 reaction
Stephen Caudill
@mrmemes.eth
I'm definitely speculating. It wouldn't surprise me though... they have a very different business model than their competitors and it doesn't seem like they charge enough to cover sending significant context with every request.
1 reply
0 recast
1 reaction
🗿
@bias
If it was some proprietary hash system they were using for the tokens that could be interesting.. but yeah as you say, it doesn’t seem to add up unless they’re deliberately operating at loss right now
1 reply
0 recast
1 reaction
Stephen Caudill
@mrmemes.eth
Operating at a loss is definitely the standard playbook for a well-funded VC-powered company that's trying to make a land grab.
0 reply
0 recast
1 reaction