In the video you posted your cursor was hovering over the message box
Huh, that’s a good point.
I guess the on-going theory on Reddit is untrue then.
Or faith, perhaps, rather than compassion. It’s interesting that people will default to a faith or trust in a system over a person, nevermind the fact that the ‘system’ was made by fallible people in the first place.
What makes this one sort of ironically humorous is that people seem to rush to its defense because they think it suggests a problem with the language model. It isn’t, it seems to be a mundane UI/UX bug. But people tend to rush to the conclusion that ‘the problem isn’t ChatGPT, it must be you that’s the problem. You’re either lying for faking this somehow’.
People have put a lot of faith in this system, very quickly, and in this case that faith is ‘leaking’ toward disbelieving the possibility of stuff like interface bugs, because they can’t separate the language model and the web interface itself, etc.
I think we’re in for a very weird time in the upcoming years. We’re in the early days yet…
1 Like
So, I have been tracking submissions to the ChatGPT subreddit, and started asking people if they were using Opera browser, and they all were. Also, doing some searching for Opera + GPT bug has turned up other examples in the wild. I also found that many Opera users have been unable to put line breaks in ChatGPT by hitting shift+enter, which may be related somehow.
So this appears to be an Opera issue.
1 Like
Nice find.
The interesting part is that it happens (what appears to be) intermittently.
After a quick search I am seeing what you see. A lot of people are complaining and noted that disabling the Opera GX “AI Prompts” feature fixes the issue.
It seems that Opera is injecting some of their own code into ChatGPT to accomplish this functionality (Admittedly I haven’t looked at how AI Prompts works but it seems like you can highlight text and then import to ChatGPT for a discussion)
Funny enough, this was also reported. Opera does inject what seems to be a control element. EDIT Ok actually “shodan” seems to be some sort of tracker…? Not sure if it’s related to Opera?
Ok, so it seems like it somehow is related based on this comment:
well. im not sure what are AI Prompts extension bc i do not have it, only some setting. also opera adblocker and opera wallet has been there since you download browser (these are not new). You can only turn off these but not remove what i know.
but you can disable ai prompts: opera://flags/#shodan-extension - opera wallet: opera://flags/#native-crypto-wallet
My god. I wonder if they are injected this into every page and selling the data in exchange for this “AI Prompts” service? This seems very shady.
The fact that I can’t find any sort of technical details on this is insane. They don’t ever mention any connection to “shodan” either. If it’s what the first result is, that’s messed up. The worst part is that some users report the injection occurring even with AI prompts turned off. Look at this sketchy console logs:
Users also report that this AI Prompts has complete access and full rights. Now, being fair, it is needed to accomplish what it’s doing. BUT, this has so many potential security & privacy issues that without any sort of transparency has me very worried.
I would need to go to the red flag wholesaler to raise enough right now.
Yeah, I came across that link too, but wasn’t sure if I’m allowed to post links yet.
I was considering downloading Opera to test it out, but I think I’ll pass.
In case anyone comes across this - I’ve submitted this as a bug to Opera, Bug #BS-69463, if you need to reference it.
Here are some recent examples of the bug I’ve come across.
https://www.reddit.com/r/ChatGPT/comments/167mpqa/thanks_just_what_i_asked_for/
https://www.reddit.com/r/ChatGPT/comments/167j1me/im_sorry_what/
https://www.reddit.com/r/ChatGPT/comments/167cq0l/quick_note_there_was_no_previous_prompt_before/
https://www.reddit.com/r/ChatGPT/comments/167bxm5/what_is_this_bug/
https://www.reddit.com/r/ChatGPT/comments/16792sh/hey_guys_how_do_i_fix_this_issue_chatgpt_gives/
https://www.reddit.com/r/ChatGPT/comments/16793vj/that_is_kind_of_not_what_i_asked_for/
https://www.reddit.com/r/ChatGPT/comments/1671z41/is_chatgpt_completely_lobotomized_now/
https://www.reddit.com/r/ChatGPT/comments/166k3lr/chatgpt_hallucinated_into_coming_up_with_its_own/
https://www.reddit.com/r/ChatGPT/comments/166i4tx/not_even_close_to_what_i_asked/
https://www.reddit.com/r/ChatGPT/comments/166g1hp/well_here_goes_my_answer/
https://www.reddit.com/r/ChatGPT/comments/165jkl5/chatgpt_keeps_giving_responses_from_other_users/
https://www.reddit.com/r/ChatGPT/comments/164u80v/uhuh_yeah_exactly_what_i_was_talking_about_there/
https://www.reddit.com/r/ChatGPT/comments/164dsg5/okay_time_for_bed_chatgpt/
https://www.reddit.com/r/ChatGPT/comments/163hrub/what_did_i_do_to_it/
https://www.reddit.com/r/ChatGPT/comments/163iufb/uhhthanks_i_guess/
https://www.reddit.com/r/ChatGPT/comments/163lw0p/what_the_heck_chatgpt/
https://www.reddit.com/r/ChatGPT/comments/1633wzd/what_on_earth/
https://www.reddit.com/r/ChatGPT/comments/1630gya/chatgpt_is_broken_i_ask_a_math_question_and_it/
https://www.reddit.com/r/ChatGPT/comments/161ki1h/wtf_just_happened/
https://www.reddit.com/r/ChatGPT/comments/161sxw8/what_is_happening_here/
https://www.reddit.com/r/ChatGPT/comments/15xkv7z/uhh_sorry_what_when_did_i_talk_about_popcorn/
https://www.reddit.com/r/ChatGPT/comments/1683dw1/my_boyfriend_almost_broke_up_with_me_because_of_a/
1 Like
Today I disabled the AI prompts feature (because it’s annoying). I still have the ChatGPT option in the sidebar though. Just now, I asked it a question, yet I still got the messed up response. This one doesn’t look like a suggested prompt, just judging by the wording.
Is this the same bug or something else?
Photo: /a/xtCIzx8 (add imgur dot com in front)
Try removing ChatGPT from your sidebar as well. As mentioned above even removing AI Prompts doesn’t actually disable the injection.
You should be able to confirm that it’s gone by running
document.querySelector("#shodan-injected-chatgpt") && alert("I am still here!")
I did that, but just now it happened again:
Photo: /a/EfjzT7s (put imgur dot com at the beginning)
Interestingly, just after I hit enter, my input message appeared under a blank message from ChatGPT. It also showed a “only one response at a time” error message. Both of these disappeared after 1-2 seconds.
Oh god I thought I was the only one! I always type my prompts on Discord or a text editor because of it. Pressing shift + enter just submits the prompt instead of adding a new line. I’m also an Opera user (regular Opera version 99.0.4788.77)
The thing is, the behavior is always changing. Sometimes line breaks work, sometimes they don’t. This also includes the response behavior. Sometimes it glitches out and generates both responses at the same time. Sometimes when I reply to the given response, instead of responding to that it goes and begins generating a second response to the initial prompt instead.
Sometimes ChatGPT works smoothly. But then within a few hours the behavior completely changes. Has been this way for months, I would say since at least June.