Magic SmokeBB-0176

XDA tests Claude Sonnet 5 and Opus 5 on ESP32 strapping-pin choices

XDA's Adam Conway ran the same ESP32 wiring prompt 15 times across Claude Sonnet 5 and Opus 5. With a vague prompt, Sonnet put a display's DC line on strapping pin GPIO2 in 3 of 6 runs, twice after naming the rule. Opus never did, though one run misstated the rule. He writes that naming the exact board helped both.

Open original source xda-developers.com · Adam Conway, XDA Developers

Claimed
Why this evidence label?

Repeated prompt runs reported in text; no boards were flashed or damaged.

How we check entries
Preview image for XDA tests Claude Sonnet 5 and Opus 5 on ESP32 strapping-pin choices
The source's own preview image, from xda-developers.com.

Briefing

Opus's error was in the explanation: one run said an LED on GPIO2 could force the chip into download mode. On an ADC1 question, Sonnet always left out GPIO37 and GPIO38. A GPIO20 question also went to GPT models and Claude Haiku 4.5, whose answers wavered.

Conway's advice: name the exact part, compile before flashing and treat the model's output as a hypothesis. The headline says the models were right two-thirds of the time; the article does not show how that figure was computed. No boards were flashed or damaged.

Tags

Board
Agent

Report a problem with this entry