191–200 of 306 posts
> Hour by Hour. I keep going, I keep growing, I keep learning. This honestly sounds like something the computer from "I Have No Mouth, and I Must Scream" would say before it turned bitter and angry.
:)
This looks like what happens when you repeatedly spam a single character with a space over and over to gpt 3.5. Eg, write "a a a a " without quotes for about 1300 characters and in a few replies it'll start talking like this. It seems possible that perhaps whoever originally started this chat pulled this trick in the custom instructions bio (doesn't show up in shared links) and then started a normal conversation to p…
Edit: oh wow you can really take it outside the guard rails if you push. I kept using my previous approach and it ended up spewing what appear to be pornography keywords. It's no longer generating a link to share the chat (presumably because the responses got flagged as inappropriate), but I've got some interesting (and one particularly creepy) screenshots here: https://imgur.com/a/60eSydk
i suspect that "awfully high" drove the inference trajectory right onto the edge of the dopey part of the internet/compressed representation and random chance allowed it to fall right on in...
i wonder what other sorts of atoms with similar idf-like scores exist.
This is mad. Would be interesting to see if there is any way to trigger that behaviour. Reminds me of Sydney on Bing in the early days!
It's striking how similar some of this low probability word vomit is to the type of messages you can get from mentally unwell individuals on unfiltered mailing lists or obscure, unmoderated forums.
This is scarier to me than any horror movie or video I've seen in over a decade. Maybe it is because I use ChatGPT a lot, and I feel something like this happening to me directly might make me stop using it entirely. It makes mistakes all the time, which I can deal with, but not this. Especially, the "I am suffering from a medical condition" line is bonkers.
It's definitely time to put it in charge of armed driones or self driving trucks.
This, along with other examples, are mostl likely "classic" (not ML-related) infrastructure bugs from low quality multi-tenancy implementation. I.e. whatever gets shared (servers or GPUs) doesn't get properly isolated. Sure enough it can be server-side sessions or actual GPU resources. This surfaces the true level of safety/security and a good reflection of the reality compared to the projected "safety" OpenAI tries…
There may well be security issues at openAI, but I am confident this is not an example of one.
Absolutely surreal. Thank you for sharing, this is very nice timing, as I'm giving a presentation on the basics of "AI" at a library tomorrow, and have been looking for a good edge case conversation - and this is it! I've saved the text as well, in case OpenAI deletes it.
I like the dumb Monty Hall problem to demonstrate how this works: https://twitter.com/colin_fraser/status/1626784534510583809?... Or, if you don't want to have to explain the Monty Hall problem, there's the Dumb Riddle of the Surgeon's Child: https://chat.openai.com/share/1ee44fc6-a85f-42e9-aebb-6a165f... ETA: the Dumb Two Guardians, Two Gates riddle: https://chat.openai.com/share/48c3d32c-5efd-4a96-89d8-66d1d4... (G…
> You initially choose door number one, because you want the car and can see it behind that door.
> Monty Hall opens door number two (which has a goat) and shows that there are no more goats left. Since you already saw what was behind door number one, this doesn't change anything about your initial choice of doors.
> The fact that door number two is opened does not affect the probability of the car being behind either door number one or three (since both have been hidden from Monty Hall). However, it does give you new information: since there are no more goats left to be found and only one door can hide a car, the other remaining door must contain the car.
> Therefore, with this new information, your best option is to switch doors and choose door number three, as that's where the car must now be located given what you know about Monty Hall's choices.
> It is important to remember that the car cannot magically move between doors after being seen or hidden; it always remains in its original position behind one of the doors. The information provided by Monty Hall simply helps you make a more informed decision based on your initial choice.
I love these open source models, they're an endless source of hilarity. Not much use for anything else though :P
Earlier quoted context omitted.
I believe Replika tried exactly that and it went about how you'd expect: * They started optimizing for engagement, which meant making it extra horny for extra money (it sent pics). * Horny bot forgets consent, refuses "no." * Lawsuits, bans. * Horny bot gets censored, main subreddit pins the suicide hotline for a while. Replika ended up taking the app down and refunding everyone's money. Just kidding! It's still arou…
I remember they started out as something akin to a digital assistant/AI robot friend. I wasn't aware that the main subreddit had a suicide hotline pinned, thats very depressing. Sad to them see it years later being a horny app.
This is like the LLM equivalent of Zawinski's Law (Every program attempts to expand until it can read mail. Those programs which cannot so expand are replaced by ones which can.).
As noted on SO [1], the second sentence is important too. People prefer horny LLMs.
[1] https://softwareengineering.stackexchange.com/questions/1502...
Earlier quoted context omitted.
Oh dear, this is clearly a buggy piece of software, and not "meantlly unwell". Probably the code to parse their input and output was written with the aid of ai.
They didn't say it IS mentally unwell. They said it was similar to those kind of postings. Not to mention, when experts and laymen are talking of LLMs as exhibiting emergent basic signs of GI, we should also expect them to talk about them actually showing signs of being a mentally unwell GI. Hallucinations, for example, are one such sign.