I encountered yet another wall-of-text blog post and wanted to use a LLM to summarize it/extract the single point the author was making.
This was a miserable failure that ended with me disabling the AI features in Firefox (and copy-pasting it into ChatGPT).
- I am asked to choose an AI chatbot, with 5 options and no extra information on each of them.
- The first one I try is useless without logging in/creating an account.
- No indication which of the ones available I can actually use without making an account.
- No indication which of them have a reasonable privacy policy, e.g. not using data for training and not storing my prompts.
- All "help" I get from Mozilla for choosing is the privacy policy (which is a wall of text that I'd need an LLM for to summarize), this page with generic advice and links to more walls of text and a link to more "helpful tips" from Mozilla which are unrelated (different set of chatbots) and outdated/wrong (Le Chat seems to be one of the few that can be used without an account).
- I pick Le Chat and ask it for a summary of the wall of text. It provides a summary of its system instructions (including the part about not revealing them, lol) because apparently the page wasn't fed to it as the context. Apparently, that requires me to use the "summarize" button which doesn't let me customize the prompt.
- I use the summarize button. I receive a message stating that since the page is long, it's only a partial summary. Which makes the feature pointless because I'm using it precisely because the page is so long. The summary of course only summarizes the waffling because the actual point of the article is somewhere in the missing part. I also don't want a summary of the waffling, I want the actual point, so I need to be able to customize the prompt.
- I disable the feature, copy-paste the page into ChatGPT with a "what points, if any, is the author of this wall of text actually trying to make" prompt, and receive my answer.
- I try to repro for this post, but can't find a way to re-enable the feature short of endless digging through about:config
So, suggestions:
- On the choice page, clearly show whether it can be used directly, free-but-account-required, or paid, plus a a one-line summary of the key privacy aspects. Yes, that requires checking the privacy policy. Yes, that's work. If you want to provide value to users, this is an opportunity to do so.
- Add a checkbox "include the content of the current page" so we can actually use the LLM in the context of the current page. For anything else, why would I use a browser sidebar rather than a bookmark?!?
- Make sure "content of the current page" means "the entire medium post" (this one was ~2200 words long). Walls of text are what LLMs are good for.
- Make the setting for enabling/disabling this discoverable.
