The HN title frames coding with LLMs as “2x, not 10x,” but the captured evidence is discussion, not the linked article. Individual commenters report outcomes from slower learning to 10x+ claims, depending on the task and workflow.
Why it matters: A useful productivity comparison needs to include planning, review, verification, and whether the task would have happened at all—not just code-generation speed. The comments also suggest matching LLM use to the user's confidence and responsibility for the work.
The captured HN title names a 28.9M-parameter LLM on an $8 microcontroller; commenters discuss speed, offline speech, scaling, and flash questions, but the linked project itself was not fetched.
Why it matters: For people exploring local AI devices, the discussion points to offline speech as a possible use while flagging throughput, scaling, and flash-read endurance as questions to investigate—not established results in this capture.