My suitcase robot gets high now off a real gas sensor wired straight into the LLM sampler. Smoke raises temperature/top_p/top_k live, so his speech genuinely gets loopier and never repeats.
A Reddit user has integrated a real MQ-2 gas sensor into their offline suitcase robot, Sparky, allowing it to simulate being "high" based on actual smoke readings. The sensor's output is translated into a 0-10 phase value that adjusts the robot's LLM sampler parameters, such as temperature, top_p, and top_k, in real-time. This results in the robot's speech becoming more varied and unpredictable, similar to human-like responses when intoxicated. The modification showcases a
Key takeaways
- MQ-2 gas sensor integrated into suitcase robot
- Real-time adjustment of LLM sampler parameters based on smoke readings
- Simulates human-like responses when intoxicated