Hi, if I can't reach a page, my first question is - is it mine or in general? So it just quickly answers if the problem on your side or not. Without switching to the new tab and opening DownDetector (though the sources are different).
For the known sites it also brings a bit additional infos, e.g. the incident description, which could help you to understand, how long to wait, for example.
Interesting approach. But I guess most of the people who uses AI now, have already a feeling, if some text is prepared by AI or not :)
I would point here to another issue - I guess most of the newly created public texts over the internet are at least passed through AI, just to polish and simplify it. So AI-style itself might be not that bad, if there is some interesting and newly invented idea behind. So the bigger problem, from my point of view, is that it's hard to distinguish: if certain AI-generated text has a good idea behind, or it is just a AI-garbage out of top-10 search result from google to certain topic.
That brings me an idea that it could be useful to have a metric like "newness" or "novelty" in addition to marking text as AI-generated. But seems like it is totally another direction :)
To your first point, yes, I agree. I would continue the thought that many of them are irritated with it and would rather not see it at all.
To the second point, that's the idea behind the adjustable threshold. It is indeed hard to distinguish; the extension will miss some gen-AI writing and return false-positives on some human writing. The user gets to decide which side to err toward.
It will never be infallible because the writing won't have C2PA metadata attached to it the way media does. It's just a tool that is a whole lot better than nothing for people who don't want to waste their time reading something written by a robot, or hate the inhuman homogeneity of "AI style." Don't you ever find yourself thinking "nobody talks like this?"
That's an interesting thing... Normally if I have only one session, I just open it and say "go on" :) Claude automatically re-reads previous context and continues from stopped point.
But it is interesting, how could it work (if it works) when you have a bunch of agents, stopped simultaneously? Are all of them have stored session context? (Sorry if questions are dumb, I'm more or less beginner here)
Good question, my usual use case is using claude, hit limits, run `handoff claude codex` then handoff generates a md file from the most recent session that codex uses to continue on.
But i have built a way to provide a specific session id to get context from. It then uses the session id that is printed when you end the session. For example, codex will print something like run `codex --resume <uuid>` to continue this session. That uuid can be provided and handoff will build a context handoff from that session specifically.
So if multiple agents are running you can swap the provider/harness by using that session id.
Long term goals would be to get this working with locally hosted open source models (right now only works with claude and codex) then I can use the SOTA models until limits are hit and continue on with the "free" models.
Hi! The idea sounds interesting, this is kind of regular process. But my question would be - normally I do clarifications within dialog mode, asking AI to add needed details and correcting it to proper direction. Sometimes AI prompt extension goes wrong direction, thus I would avoid automatic use for me.
But it might be most interesting for me as an automatic calculation of "vagueness" and in this case automatically to switch to that dialog clarification mode.
For the known sites it also brings a bit additional infos, e.g. the incident description, which could help you to understand, how long to wait, for example.
reply