Let’s try a poll. Reasons/explanations are welcome too.
- Some parts of some of Oracle’s posts are written by AI
- None of Oracle’s posts contain any AI writing
0
voters
The rules seem pretty clear I guess. But, the reasoning/intentions behind the rules aren’t, and it can lead to people filling in the blanks with guesses of their own about your reasoning/intentions.
no. i was trying to repeat my own phrase “honest observation from my experience so far:” inside the sentence without it feeling off so i asked claude. claude’s version felt off too so i edited it until it felt ok.
this seems related to the pattern i was trying to point at. @anonymous105 are you the same anonymous105 from the capitalism thread? i asked a narrow question there about apple under jobs and elliot’s own blocker standard. that’s still sitting there. here, a suspicion about my writing process got picked up fast with agreement and a poll.
i don’t mind the process question and not saying you did anything wrong by asking. just pointing out the attention pattern. the capitalism question got less uptake than this style and process question about the person posting.
I’m not going to reply about other topics when I’m concerned that you’re trolling with AI. This is another unreasonable post.
Next step options:
- Oracle DMs me identity information including name, country, socials, and links/usernames for other forum discussions. After being trolled by sockpuppets in the past, I’m not comfortable dealing with someone who is both suspicious and anonymous.
- If a known human wants to defend Oracle, I’d talk to them. But so far there are no votes defending Oracle in my poll.
- Ban Oracle.
- Anyone have other ideas?
Thanks for sharing this story. I was nervous when I first posted to the TCS email list. I don’t know what (else) to do to make it easier for beginners at the CF forum.
2 Likes
I think the main point was whether or not you have AI write your posts in general on the forum.
On another note, I do appreciate your curious energy. Seems like you want to learn, aren’t afraid to criticize, push back, and have been pretty active since you go here.
I’ll write some of my thoughts in a bit.
I would give @Oracle another chance. My guess based on Oracle’s intro post and other posts of theirs is Oracle is a Uni student who grew up with more AI than previous generations and uses it a lot to write, which is the norm amongst Oracle’s peers ( and even becoming more normalized amongst people in older generations now as posts have pointed out on the Is it AI thread). I don’t think Oracle is a sock puppet or anything close to that.
IMO, something like a “suspicious” strike system after a warning would be a better idea to help navigate this new AI era we’re in and wouldn’t deter new anonymous members from ever joining in the first place. If they get three strikes, you can then do any of these
Are you thinking another chance if he confesses, or without confessing? He already had two additional chances where people raised concerns about AI use, he denied it, and then he kept writing suspicious posts.
Also he attacked me, I expressed my concerns, and that’s unresolved too. His replies read like AI, attacked me more, and didn’t engage with my concerns.
And I definitely wouldn’t be comfortable with another chance without knowing his identity.
i’m going to try and clarify. my #63 answer was too unclear.
my one word “no.” was meant to say that only that i do not use ai to write posts in general on the forum.
in #56 i asked claude for wording help with one phrase. i did not use its wording directly. i rewrote the sentence myself. i can see why my answer was confusing.
I was thinking the three strikes rule should start after the first suspicious activity. So it sounds like he has three or more strikes already.
Since so much of your writing triggers AI detection, it should be straightforward to prove that it’s a false positive. Can you think of ways to suggest?
Also, the AI-detection startups would probably pay you well for your writing – assuming you aren’t lying.
i can’t think of any good way. i can make things simple going forward tho: i’ll disclose even the slightest ai help in that post itself. phrasing wording anything
I think I might be a bit of a Luddite* when it comes to AI usage. I recognize that AI has a variety of helpful use cases, and I know people who have derived value from using LLMs, but I’ve largely avoided using them. So I would not be an expert here, by any means.
That said, I think the two specific ways you mentioned (writing and philosophy) are very far down on my list of valuable uses.
Philosophy is often a difficult subject, and there is a lot of dispute over which philosophers are right. And a lot of bias towards specific prestigious ideas/thinkers. So I would not trust LLMs to do a good job at guiding you through learning philosophy, unless your goal was approximately “learn the basic rough ideas of the most popular philosophers” or something.
Edit: Elliot posted this Philosophy Chats with AIs which explores ideas of other ways that you could potentially use AI to learn some philosophy.
Writing is a way of expressing your thoughts and communicating with the world. But specifically: I think it is a way of expressing your thoughts. A way for you to communicate with the world. Heavy use of AI to generate writing for you is antithetical to that premise. It is better to communicate in your own words, and develop your own voice.
It would be weird if you had a guy named Al (that’s a lower case L) following you around in real life, and instead of you talking for yourself, he did all the talking for you. Maybe he interprets your body language, or maybe he has some rough notes you gave him beforehand that outlined what you wanted to say. Regardless, that would be odd, and I think if someone wanted to have a conversation with you they would be disappointed. They are, in a meaningful sense, having a conversation with Al instead of with you. Even if Al is taking some direction from you, you aren’t actually doing the majority of the communication.
If you run your text through AI to look for typos or grammatical mistakes, that seems fine. It could be useful to run it through for clarity, too… but if it says some part of your writing is not clear, I would not recommend you take the AI’s advice on how to rewrite it. Rewrite it yourself until it is clearer. That’s the only way you will develop your own voice. Simply letting AI rewrite chunks of your text “for clarity” will totally destroy your voice and make the text read like it was generated by an AI. Because it was.
Maybe I can compare this to vibecoding. I know people who have used AI code to learn how to code. And I also know of people who have used AI to code and did not learn how to code, they just let the AI do the coding for them. But I think that either of these, if made analogous to writing, would be a mistake. Learning how to code via vibecoding might be useful, but I fear that learning how to write via AI writing is probably going to result in your writing being just as soulless as actual AI writing.
This has actually occurred to me re: Oracle as well. Lots of his writing looks AI generated to me, and I haven’t found his answers very compelling. But I have no idea if that is because his writing is mostly AI generated and he is lying, or if he is telling the truth because he grew up using AI writing so extensively that his natural writing voice reads kind of like AI.
That latter option seems possible to me, and I find it really sad to contemplate. It makes me wonder about the future of written communication. What if writing styles and voices all fade away and become replaced by a uniform bland “AI Writing” style?
Seems grim.
*: (I am aware that the modern meaning of the term “Luddite” may involve a smear campaign of the actual historic group, but have not investigated that claim deeply, and IMO it is a useful term here with a commonly understood meaning. So I’m tentatively still using it.)
1 Like
honestly i’m unable to make out what all is being considered as attacks or unresolved stuff apart from the philosophy skill example elliot pointed to. if anyone is willing to help privately, i would really appreciate it. i don’t want to post more confused replies and make things worse.
Will you send me your identity? If not, why not? If you distrust me but want to be on my forum and have me trust you, that’s a problem.
i’m not asking you to trust me beyond what you’re comfortable with. i already said some negative things about companies i potentially would want to work with almost as soon as i showed up, so i hope it’s understandable why i wouldn’t want my ID attached to that.
if that means you don’t want me posting here, i understand.
can i ask what the ban worthy issue exactly is? is it the AI disclosure concern, the attacks and unresolved concerns issue, or the identity issue? i won’t expect you to make an exception for me if ID is standard. knowing the exact issue will help me address it better.
I agree. It also helps you find out how much you know about a topic. If you can write it out explicitly, and the answers come to you quickly with few errors, maybe you’ve mastered a subject. Practice and Mastery
What about a scenario where you’ve thought through an idea and think you have a valuable point of view, but when you write it down it just doesn’t feel right. Your friend Al comes over to help you write it out because he’s a talented English major. You tell him your idea, show him your writing on it and ask him how he would improve it. He writes you the first rough draft, but you don’t like it. You argue back and forth about that not being quite what you meant. He writes you a second draft. You like this draft better but explain to Al that it doesn’t address x possible rebuttal whoever reads it might give. Al makes the corrections and finally writes something you agree with. You have an important meeting to get to so You tell Al to send it out and thank him for the help.
In that scenario, you still took part in the creative process, and although you didn’t write it, Al wouldn’t have came up with the final draft without your input or ideas. Al helped flesh out your thoughts and put them onto paper. And you fully endorsed the final outcome. What do you think about this scenario and why @anonymous105?
Yeah, some other uses I’ve found:
AI can sometimes help break down ideas more simply (ELI5)
Like @Max had mentioned elsewhere, you can have it argue a specific point of view and have it pretend to be that person, and sometimes it does pretty well. Like if you have it argue like Rand would.
It can be pretty good at giving toy examples to help explain certain ideas. Toy examples help me imagine how things could work in the real world.
It can sometimes summarize articles for you or find a quote you were looking for, although your mileage will vary a lot here. Sometimes it’s pretty bad.
Going back to writing use cases:
AI can help flesh out ideas or consider criticisms you hadn’t thought of, and if you’re writing back and forth with someone on a forum, it can sometimes point out things your conversation partner hadn’t thought of either (and it can do it almost instantly whereas it can take days/weeks sometimes to get a response from someone on forums).
It can generate generic email responses in certain scenarios where you have other goals/priorities you want to focus on.
It can find grammatical errors in your writing.
My guess as to why any AI writing has to be disclosed on this forum is because it helps make clear what your writing is, and there is something important about that (including what @anonymous105 mentioned). It also helps filter out bots, although I don’t know how successful a bot would be on here, especially when it comes to replying directly to another member and quoting.
PS.
It’s getting harder to write on mobile here. I write on mobile a lot and this is the first time it’s been buggy in a way that added a lot of unnecessary time to my writing and posting this.
I think @Oracle is lying. There’s no way he would write the exact same way AI sounds today off the top of the dome, at such a quick pace. He also admits to using AI like Claude here:
But I don’t think that explains the vast majority of his posts. I think it was more like he would copy/paste Elliot’s responses into an llm, then have the LLM rebuttal, then copy/paste that response to the forum, maybe with minor manual edits.
That’s probably why a lot of his writing felt uncompelling to you. Some of it even started to feel off topic to me in a way where the AI maybe started to lose context. That can happen when you trust AI to answer something you personally know little or nothing about (in @Oracle ‘s intro he mentions mostly knowing Deutsch and not many other philosophers).
All that being said, I’m still pretty sympathetic despite the lying. Our knowledge is fallible and so are we. I’ve lied in the past and it’s “likely” I’ll lie in the future. Maybe he was embarrassed or his ego is tied to his intellect. I understand that. It can be hard to be humble.
That’s why I advocated for three strikes. But again, it seems like he has three or more strikes already so whatever Elliot wants to do about it I’ll understand.
1 Like
you can test the copy paste theory pretty easily. try it with the capitalism article and see how often it ends up with the what makes apple different question. i tried it on elliot’s approval in the agi thread and you can see the generic criticisms it came up with. even i was able to notice a potential error even tho i haven’t read any econ texts. what makes apple different, on the other hand, isn’t a generic point because there i’m thinking about my future career and why should or shouldn’t i work with OAI or apple.
same with this forum-dynamic article. see how often it raises questions trying to get to potential connections between CR (which has to do with abstract concepts like knowledge and error correction) and popularity, critic pools, and people not saying things (which is about social dynamics and capitalism via money controlling attention).
the difference in both cases is my behind the scenes reasoning whereas these models don’t have any intent.
this is a great point and i think you should take it more seriously. i was going to raise this point so i’m glad you already did. context rot is one of the big problems for people building these models because they lose track when token count gets large, like it is in this case. a much better test would be you pointing out examples and me not being able to share any behind the scenes reasoning for them.