I won’t use Deepseek Harness for a newbie like me.
Waiting for the big guys to make a batch of plug-ins. Let’s wait for the reviewers to say which one is better. I will use it then; Currently, Hermes/workbuddy is enough for a novice like me.
Deepseek just killed the coding-agent industry.
A frontier lab just gave away for free the exact layer companies charge $200/month for. no subscription. no vendor lock-in. no closed source. i went to check the repo, because if this holds, the coding-agent subscription model
Karpathy's prediction about RL is coming true now!
He called reward functions unreliable and argued that a single reward number is too low-dimensional to teach an agent what "good" means for complex tasks. To solve this, Agents need a knowledge-guided review as a
Composio tested 8 harnesses, DeepSeek V4 Flash ran 30 hard tasks, and Pi directly won the highest pass rate, with a success rate of only 2.8 cents.
Many people will find it strange: Why is the simplest Pi the most powerful? Pi engineers Mario and Armin had a thorough chat this time: 1. Code is the truth When writing code, no additional memory system is required. As long as the code itself, git
You have reached the end of the archive
All of deepseek