DeepSeek talks so cutely. Holy shit.
423 pieces of news and 15 desks were handed over to two models, jev and ds, to perform the same set of screening and judgment.
Come and experience jev vs deepseek You can intuitively see that jev does not need to generate a long answer, but directly returns a structured decision. Jev speed is about 780 judgments/sec and P50 latency is 476ms. After Jev was completed, DeepSeek only ran about 14% of the complete task volume.
I paired my cognitive framework "Assess Decide Do" with a JEV implementation.
In the movie below, "Assess" and "Do" are LLMs, they output text. "Decide" is JEV, it evaluates the state (which is dynamically updated by LLMS) and route to: - Assess, if we need more clarity, so
Deepseek V4.1 flash — the $0.07 model beating $4 giants
There's a tiny open-source AI that costs pennies and keeps up with the most expensive models on earth. It's DeepSeek V4.1 Flash. Free weights. Download it today. → 552B parameters but only wakes up 8B to read. The rest
You have reached the end of the archive
All of deepseek