Am essential part of AutoGPT is going to be knowing what code is safe to run, which is conveniently a core feature of @darklang
@paulbiggar @darklang I am weirdly less hot on autogpt despite being super bullish on this current phase of ML. Compounding probabilities just feel like they’re gonna render it useless for real tasks with real consequences
@cartermp @darklang My sense is that a human-in-the-loop is going to be essential for keeping it on track, but that it will improve and need less supervision over time (esp if you keep track of the humans responses).