I threw together a little prototype that uses GPT-3's logprob output to flag unlikely tokens, so they can be double-checked by a human proofreader. It's sort of a turbo-spellchecker: in addition to catching "They had fuor eggs", it can also catch "They had for eggs", "He asked me to prostate myself to the king", and even the somewhat subtle bug in this C code:

    for (int y = 0; y < HEIGHT; y++) {
        for (int x = 0; x < WIDTH; x++) {
            buf[y * HEIGHT + x] = 0;
        }
    }

Hope you like it, and curious to hear people's thoughts, feedback, or questions! :-)