3 ms·
I think the jump from ChatGPT to GPT4 was also something like 130 ELO points. This roughly equals 2/3 preference for the new model, 1/3 preference for the old m
by starchild3001 2y ago
I think the jump from ChatGPT to GPT4 was also something like 130 ELO points. This roughly equals 2/3 preference for the new model, 1/3 preference for the old model. That's roughly how much top models have improved for an "average query".
Now, none of us enter average queries :) We all have specific use cases in mind. So any specific user's mileage will vary.
I personally feel substantial improvement. I don't feel the need to google every LLM answer 5 times to feel confident about it :)
- dartos 2y agoThat’s really great for you. I have been mislead several times in very subtle ways when trying to ask specific questions to chatgpt and Claude. It actually burned me at work. Made me look bad and caused my project to take an extra a sprint because of incorrect info about the web history API that came with working examples. I just can’t ever trust information from these systems, unless they include sources to first party docs.