Santiago
banner
santiagosam.bsky.social
Santiago
@santiagosam.bsky.social
Previous SWE Tech Lead, now trying AI Engineer | Working with computer vision and LLMs to build something great. Instagram: santisamc
Another thing that sometimes helps is to ask if “understands” the problem and have explain it (of course, understanding is a relative term because is an LLM lol). Asking to just go step by step through the problem/solution is also good, because it gives focus.
February 7, 2025 at 4:02 PM
Claude has been great at coding, but o3-mini-high is worth a try! With ChatGPT, I've had better luck when I've given it all the context up front - details of the problem, documentation , anything that might give good context. If it starts to spin up, I reboot and try a clean chat again.
February 7, 2025 at 3:59 PM
The same thing happened to me with Deepseek. Try o3-mini-high, it's very good, maybe it's too early to say but so far it's the best model I've tried for code. Most of the time it gets things right the first time.
February 2, 2025 at 6:57 PM
It's really interesting, the opposite happened to me, ChatGPT was able to help me solve a strange bug with Deepstream that Claude couldn't. In my experience it depends on the problem, it is good to have both.
November 25, 2024 at 1:24 AM