ChatGPT is supposed to be really smart according to all the rankings I've seen but it just seems worse than the other AI at all the tasks I'm trying to use it for?Where exactly does it outperform?
>>110289332>Where exactly does it outperform?artificial benchmarks and funny curves on flipcharts
>>110289332posting frogs is your only usecase for computers frognigger
>>110289350spbp LMAO>>110289332there's a difference between the free and paid vers. also, LLM corpos sometimes intentionally throttle their models for specific users if they're ebil or just have throttled performance due to costs. GPT 5 had noticeably worse performance compared to GPT 4.some models have a larger dataset compared to others that they've been trained on that could also effect performance.ask your question again but in here: >>110273013no point in making a whole thread about it.
>>110289332It's all a psyop to get you to pay for the premium sub
>>110289332It outperforms in well-specified tasks, which is nerd speak for "if you hold its hand and do all the mental work yourself it will do what you told it to", and benchmark tasks are NOTORIOUSLY specified, often literally including code snippets and telling them exactly what parts of the code to look at, the edge cases to consider, etc. - they are basically asking the model to implement features that are already designed.Opus on the other hand actually understands shit and has the common sense to know what you mean when you don't spell out every little detail like a design spec, because it can design shit itself.
>>110289332The free one isn't any better than self hosted Gemma.The paid one is crazy good.
>>110289602>The free one isn't any better than self hosted Gemma.Okay let's not get carried away. All these <30B models are basically useless.
