eats tokens like crazy since its probably parsing the entire html structure , whenever you use it in a site that has seo tags or inline svgs , goes exponentially.
but its surprisingly stable enough to get the actions done
Kind of don't see the point of this. Just adds a further point of failure (in potential misinterpretation of instructions) and additional cost for literally no reason at all.
Using Puppeteer or something like it is maybe a bit more explicit, but would work much more efficiently.
I used it for some tests in the past. I think I tried the cheaper OpenAI model and the more expensive one. Cheaper took way more steps which ended up being more expensive in the end. But that was a few months ago, not sure if anything changed since then. I guess it's OK if use-cases change and to quickly test things, but for repetitive tasks, better to find different solutions that don't cost 10cents to a dollar to use per instance.
👋 Join WIP to participate
yeah i actually have
x.com/Robimez/status/18817073…
eats tokens like crazy since its probably parsing the entire html structure , whenever you use it in a site that has seo tags or inline svgs , goes exponentially.
but its surprisingly stable enough to get the actions done
Kind of don't see the point of this. Just adds a further point of failure (in potential misinterpretation of instructions) and additional cost for literally no reason at all.
Using Puppeteer or something like it is maybe a bit more explicit, but would work much more efficiently.
I used it for some tests in the past. I think I tried the cheaper OpenAI model and the more expensive one. Cheaper took way more steps which ended up being more expensive in the end. But that was a few months ago, not sure if anything changed since then. I guess it's OK if use-cases change and to quickly test things, but for repetitive tasks, better to find different solutions that don't cost 10cents to a dollar to use per instance.