Scrolling through the feed and found some really solid reads today. A mix of relatable workplace drama, AI amazement, heatwave struggles, and local LLM debates. Curious which one speaks to you the most! ๐
- Wojtek322 sharing the pain of a new WFH policy that basically makes remote w...
I just don't find much utility in local coding models. Maybe I'm just not using it right. I used Qwen 3.6 35B to try to help with my reverse engineering project and it must not be its forte, It struggled to get anything right. Opus breezes through it like it owns the place.
That's also my experience. But realize, that Gemini and stuff beat the open models and god, I would never use Gemini. So...
And yes, for "dumber" models, extra / different prompting is needed to get certain results. I kinda liked DeepSeek because it did literally what is told and nothing more, but it left too much dead code. Maybe it would've been fixed if I just added it to the prompt. Who knows.
Models are also a bit like choosing a supermarket. You go to the one you are just used to. A kinda flavor thing ๐.
Claude back in the day did even more than it does now. That was also not comfortable. All extra code is code that has to be maintained. Sugar is not appreciated. ๐