3 ms·
So, uh, GPT-4 outperforms at labeling. What is that labeling used for? "Employing Surge AI's top-tier human annotators at a rate of $25 per hour would have co
by rossdavidh 4y ago
So, uh, GPT-4 outperforms at labeling. What is that labeling used for?
"Employing Surge AI's top-tier human annotators at a rate of $25 per hour would have cost $500,000 for 20,000 hours of work, an excessive amount to invest in the research endeavor. Surge AI is a venture-backed startup that performs the human labeling for numerous AI companies including OpenAI, Meta, and Anthropic."
What could go wrong? Using GPT-4 to perform labeling used by OpenAI in order to train...uh, wait.
- realusername 4y agoWe're also using computers to build better computers, it makes sense
- rossdavidh 4y agohttps://en.wikipedia.org/wiki/Positive_feedback https://en.wikipedia.org/wiki/Positive_feedback
- mztwo 4y agoYep - you highlighted exactly what raised my eyebrow as I was writing the article.
- nethdeco 4y agoAnd the noise would keep adding up.
- mclightning 4y agoThis is a bigger problem than people realise. Think about it, how many millions of articles are posted online produced by OpenAI's GPTs to date... Good luck clearing out the training data for GPT-5. True human content will get gradually scarce. We steer it for sure for our posts, but it is still GPTs that do the heavy lifting. OpenAI's own classifier fails to detect GPT-4 generated text at the moment.
- pixl97 4y ago>classifier fails to detect GPT-4 generated text That's because beyond the 'As an AI language model' and a few key words it can be nearly impossible to detect GPT-4 especially if any prompt is used to intentionally keep it from being detected. Human like text is a solved problem. There is no more getting better at detecting AI written text, there is only classifying more humans incorrectly at this point.