@hahnbeeIee @linear Agreed, kinda niche usecase for fast frowing startups tho :)

@raymond_huang26 @keywordsai This is actually a great idea

@TheIshanGoswami Nope just tried it out today - cursor was still quite a bit better

@DarrelMuonekwu wanted to optimise his lifestyle to not get cancer and he came to the conclusion not living is the most optimal way…… what do I do

@vikhyatk yeah chatgpt is getting better and better (UI, models, features etc) im about to cancel my claude sub as well

@AravSrinivas Make it opensource!

@janleike Bro new models pls

@gdb Oh wow, it’s technically an o3 release then

RT @simulacronist: i was recently discharged from the icu at the sf hospital after traumatic brain injury due to a man hitting me on the ba…

o1-high is better than o3-high at creating system prompts atleast. claude with edit_str tool is still the best overall tho

@ankrgyl Do u have any benchmarks for tool calling (something preferably where claude does significantly better than 4o)? Curious how it does at chained tool calls

So, what’s the consensus on o3-mini-high

lmao, it gets confused very easily. especially when multi-turn is involved. its like dealing with someone with memory loss

not sure whats happening but claude on seems very inferior to 4o on chatgpt

@dakshgup lambda labs

Maybe this will do a better job explaining: when you go to DeepSeek's shop, your ice cream order has to travel allllll the way to China before you get your ice cream. But when someone else opens their own shop using DeepSeek's recipe, your order just goes to the shop down the street! Same ice cream, but your order stays right here instead of taking a long trip to China! ice-cream recipe being deepseek's weights and ice-cream being your requests

@RealRichomie How hard is it to hire from China? Especially when I have no connection to china whatsoever

@Benioff yes - data is a moat, but you also have to know what to do with all that data, which I doubt salesforce (or agentsforce?) does