Károly Zsolnai-Fehér discusses reported improvements in a DeepSeek model and explains why token price alone does not determine the cost of completing a taskCost per completed task measures the total AI, tool, infrastructure, retry, and repair expense for each verified useful outcome.. He describes his own observations that different models can use different numbers of tokens for the same work.
Károly Zsolnai-Fehér discusses open weightsAn open-weight model makes its trained parameter values available so others can run or adapt it under the release's licence. and the hardware required to run large models. He describes how subsequent smaller models may make similar capabilities more accessible.
Károly Zsolnai-Fehér presents learning from self-feedbackAn AI feedback loop occurs when AI outputs or their consequences become inputs that influence later model or agent behavior. as a possible explanation for the reported gains, while explicitly acknowledging that the explanation is uncertain. He uses an iterative game-development example to explain the idea.
Watch on YouTube
