4 ms·
A few comments: 1) "Teaching" GPT-3 is misleading, unless the model is explicitly finetuned on that prompt schema. The default davinci model for the OpenAI API
by minimaxir 6y ago
A few comments:
1) "Teaching" GPT-3 is misleading, unless the model is explicitly finetuned on that prompt schema. The default davinci model for the OpenAI API is the same model for everyone using it; the few-shot prompts merely guide GPT-3 toward a domain and generation schema.
2) This approach is essentially a classification problem, with extra steps. That's not new with GPT-3; causal language models (including GPT-2) use classification tasks as a way to quantify model performance (see the papers). Hugging Face transformers is a good library to finetune Transformers specifically for classification tasks.
3) The full string "yo be real" is inefficient if you're just checking for positive/negative responses. (Token efficiency will likely be an interesting area of optimizing GPT-3 performance)
4) That said, using token prediction probabilities can be useful for assessing model performance (I explicitly duplicated the web UI's implementation for my new client: https://twitter.com/minimaxir/status/1288125135526875146 https://twitter.com/minimaxir/status/1288125135526875146). However, the prediction probability is often very dependent on previous tokens, including the prompt, so the results are not universally consistent. But there might be more clever outside-the-box approaches using the token probabilities and heuristics...
tbh, I wish OpenAI would also return attention weights for each previous token in order to be able to more easily-visualize which tokens have a high influence on generation.
- arram 6y agoTeaching is used in the colloquial way, not in a technical ML sense. The model isn't being retrained or fine tuned.
- typon 6y agoThe prompt is essentially fine-tuning with extra steps
- minimaxir 6y agoNot really. Fine-tuning fundamentally changes the model weights to be more amendable to a particular use case/domain; the few-shot prompts for GPT-3 is just a guide (and it's very easy for the model to ignore said guides and go on a tangent). If you could finetune the 175B model, you'd likely get even better results for these Q&A prompts. (unclear how the OpenAI API is implementing its finetuning demo but I believe it's not on the 175B model).
- IshKebab 6y agoI think they only fine tune the top layers.
- msapaydin 6y ago>essentially a classification problem, with extra steps. I think it is a question answering problem, with the extra of being able to say "I don't know", rather than a classification problem. The current transformers models for question answering such as huggingface pipeline implementations do not handle the "I don't know" case and this is an attempt to see whether gpt-3 can be trained to have this extra functionality easily. >results are not universally consistent. I think the relative probabilities are what matters and those may be more consistent than the absolute probabilities.