Bringing Continual Learning into Enterprises — Samuel Denton, Applied Compute
Applied Compute uses a technique called continuous learning to improve AI models directly in business production environments.
In one example, they got a Qwen model to complete tasks on SWE bench twice as fast (from 80 down to 40 steps) by showing the model how to behave based on its existing behavior — rather than forcing it toward a "correct answer." Their approach works in two modes: either with old saved data from production (quick to start but limited improvement), or with live feedback from each new run (slower to set up but yields much better results). In a real-world example, they successfully got a model to format hyperlinks correctly in 80 percent of cases by providing smart tips after each attempt — something that both reward systems and standard training failed to achieve.
Vibekollen prepared this summary with AI from the original publication. The content belongs to AI Engineer.
