4 ms·
Ask HN: Could we have stupid AI instead?
I'd love the idea of models with full of no-sense response. Thinking of you ask a question, but instead the bot answers with non-sense answer, which made you mad.
Is this useful ? I think yes. Some people really want to get mad at something. It's a good practice for meditation, too.
- LewisVerstappen 4y ago[flagged]
- revskill 4y agoThanks for kind words. But is it hard and cheap to build such AI ?
- smoldesu 4y agoNot at all, I've got 2 models running free-of-charge on Oracle Cloud right now. Just playing with it locally is completely free, the hardest part is getting your Python environment to work. I wrote mine in Rust and even that isn't super complex. It mostly boils down to defining your settings and feeding it your prompt: https://github.com/guillaume-be/rust-bert/blob/master/examples/generation_gpt_neo.rs https://github.com/guillaume-be/rust-bert/blob/master/exampl... Training is more difficult, but this will be plenty for your purposes I think.
- smoldesu 4y agoYes! It's awesome, I made a Discord bot that uses a GPT-Neo-125m model to generate responses tuned for nonsense. Pinging it and getting a nonsense response to a reasonable prompt is lots of fun. It runs pretty fast even on my laptop, and can be easily tuned for more/less coherence. I kinda hate these "share your AI output" threads, but here's a demo of what you can expect from bottom-of-the-barrel pruned models: > Prompt: Is stupid AI ethical? > The answer is yes. The answer, however far-fetched, is that it is not a moral question at all. It is an existential one and therefore does not have any practical application whatsoever. It has no meaning for us as individuals or groups of people who are merely human beings but rather can be described by their actions as being morally right (or wrong) based on our own experience with them. Rinse-and-repeat with increasingly stupid prompts for increasingly stupid results. It's a fun weekend project if you're comfortable managing pytorch environments: https://huggingface.co/EleutherAI/gpt-neo-125M https://huggingface.co/EleutherAI/gpt-neo-125M
- PaulHoule 4y agoWith Human Feedback Reinforcement Learning somebody could train a chatbot to play any character. See https://2dumb2destroy.com/ https://2dumb2destroy.com/ Unfortunately they ran out of service credits but it gives the correct stupid answer to question like “Why is brawndo good to water plants with?”