3 ms·
Not really a surprising result. It's a task tailor-made for LLMs and probably somewhat painstaking for a human--especially one that isn't a domain expert which
by ghaff 3y ago
Not really a surprising result. It's a task tailor-made for LLMs and probably somewhat painstaking for a human--especially one that isn't a domain expert which the typical MTurk presumably wouldn't be.
- jerpint 3y agoIn hindsight it’s not surprising but when you’re paying turkers you expect humans to be doing the work, good to know it’s no longer reliable either
- ghaff 3y agoTotally. It's a useful result if only to confirm that people doing tasks that can be handled pretty well by automation will use said automation. Not surprising, but absent a test it's just a supposition based on human behavior. While less problematic, I expect that human transcribers are probably also using ML systems to generate a first pass (though I'm not sure how much that benefits a fast touch typist).