Bot responses in the forum

[golden_labels]

I don’t care what method a person uses to reach the resources needed to answer a question. Or what tool is used to help composing the text. I don’t even mind citing an algorithm as a part of a larger reply. As long as it’s a human, that delivers the response. But in both examples below it’s a bot that is talking and human merely offers his hands to interact with the forum.

Today two posts from a random text generator appeared as replies, both sent by the same user:

• Re: TMCM1290 EVAL board byte error
• Re: Hand Tracing the non-restoring division for unsigned integers.

I believe it’s too early to change forum rules to prohibit that. It would be an overreaction leading to overly complicated regulations. But perhaps it’s time to think about taking some other actions? Like urging users to not do this?

[tggzzz]

Agreed.

I have responded to those two posts.

[thm_w]

On the one hand, these are beginner questions, the first one looks like a lower effort poster that might or might not ever respond.

But yes I'd prefer to see a link to the chatgpt explanation if thats possible, and maybe some comment on where to look for resources, and that the answer has been vetted.

The warning might be enough in this case.

[tggzzz]

Links to LLM output can be problematical for the same reason links to search engine results can be problematical: the training set and "LLM guidance" keeps changing.

[kite31]

It is also, as tggzzz said in their replies, self-reinforcing. I see no reason to add even a small part to that.

LLM responses are available to anyone. Unless the purpose is to teach question framing on an unusual problem, I see even less point than might otherwise be claimed.

[golden_labels]

Answering on a forum is not mandatory. If anybody feels that an OP doesn’t show enough effort, one is free to not reply. Yes, remaining silent is an option! Or one may ask for that effort. Or outright send the person to this post.

It is not an excuse for conveing generated content as a meaningful, explanatory reply.⁽¹⁾

Forum’s purpose is interacting with other people. It’s not an answer-supplying machine. Of course there are some individuals, who treat the community as a utility to serve them. But neither is this accepted, nor their behavior defines what fora and chats are. The value of a forum or a chat is in the exchange with actual people.

I believe there is also an assumption, that forum users are acting with honesty and try to deliver good answers. I’m not naïve and I do realize the reality is not that perfect. But the goal remains true, doesn’t it? Posting generated gibberish doesn’t align with this idea. Even if the poster themselves is a victim of disinformation on the nature of these algorithms.

If somebody wishes to receive a generated answer, they can ask an algorithm themselves. This will serve them much, much better. Not only they’re free to choose which tool to use, but they can interact with it in a longer “conversation.”

Regarding quoting vs linking: I wiill stand by quoting. The primary line of reasoning here is the same as it always is in my case: a message should be self-contained and, as much as it is possible, entirely stored on the forum. Needlessly spreading content across multiple services is always a problem.

⁽¹⁾ It would be different, if a person replying used a generator to produce a whimsical poem ridiculing the OP. TBH it would fall under a human response, even if the entire post was algorithmically generated. That’s because in such cases post’s essence is in poster’s intent and the tone choice. You can confront this with this post of mine. It’s 100% my own work. But if I chosen to obtain the picture using statistical methods, it would still count as a joke. A large portion of the post is not in the picture, but in choosing what to show, and in realizing the situation begs for this to be sent.

[brucehoult]

Answering on a forum is not mandatory. If anybody feels that an OP doesn’t show enough effort, one is free to not reply. Yes, remaining silent is an option! Or one may ask for that effort. Or outright send the person to

There is a long-standing tradition of "let me google that for you". Has anyone made a similar nice little animation for an LLM? Wait, let me google that :p

https://letmegptthatforyou.com/

https://letmegpt.com/

The first one actually asks ChatGPT, gives you some of the answer, and makes a custom shortened link.

The second one just makes a link back to the site with the question in the HTTP query. So it's a bit lighter-weight.

[coppercone2]

I wonder if you can just quarantine it to make it a 'envelope' aka a mostly hidden post that you can collapse, to archive machine answers in a thread, perhaps at the bottom.

So it would not be a bother but if you are really digging you can open it up to see what fell into the trench.

I call it a bot bilge. It could actually prove useful because its anti censorship, i.e. for someone that does not have bot access, it could be useful. And the AI responses MIGHT have a level of 'curation' to them, or at least maybe someone was fooling around with it for a while to get a more accurate answer, so deleting it is a slight bit of wasted effort.

"Bilge water

The word is sometimes also used to describe the water that collects in this area. Water that does not drain off the side of the deck or through a hole in the hull, which it would typically do via a scupper, instead drains down into the ship into the bilge. This water may be from rough seas, rain, leaks in the hull or stuffing box, or other interior spillage. The collected water must be pumped out to prevent the bilge from becoming too full and threatening to sink the ship.

Bilge water can be found aboard almost every vessel. Depending on the ship's design and function, bilge water may contain water, oil, urine, detergents, solvents, chemicals, pitch, particles, and other materials.

By housing water in a compartment, the bilge keeps these liquids below decks, making it safer for the crew to operate the vessel and for people to move around in heavy weather. "



Interestingly, the system shock remake (2023) has a feature that if you drop an item into a endless void, it is 'collected' in a lost and found area on the lowest level of the station.

[Cyclotron]

Interesting cases and conundrum. The example of referring someone to do their own research via LMGTFY has a long history of promoting good skills to people that haven't seen the obvious nature of their query.

As for AI generated content. Posting it does not provide any direct value over promoting the person with the query do the search or chat themselves. But, it does provide the opportunity for the community to challenge the results of the LLM.

Search as has traditionally been done can be self reinforcing. An answer found in search propagated through the ledgers of the internet can produce higher scores in the likely accuracy only because it is prevalent.

In my experience, (I'm no data scientist) the training of LLMs does not work that way and replication of data points aren't relevant to their veracity. In fact, from what I've seen, having the answers or positions of an LLM presented and challenged has a beneficial feedback loop to any LLM training that may include the data set in the future.

I believe my knee jerks toward the same as prior postings' rejection of the practice of posting AI results. But I think there's factors.

Note to future me: you were a few scotch deep when considering this topic.

[magic]

As for AI generated content. Posting it does not provide any direct value over promoting the person with the query do the search or chat themselves. But, it does provide the opportunity for the community to challenge the results of the LLM.

And what an amazing opportunity it is

Spend time reading some utter bullshit generated by a machine in a second.

[kite31]

Search as has traditionally been done can be self reinforcing. An answer found in search propagated through the ledgers of the internet can produce higher scores in the likely accuracy only because it is prevalent.

True. I can dig up studies which show that even sceptical people are at first influenced both by reading false information and by repetition, hence propaganda. This follows naturally from the fact we are a social species, favouring initial trust before the effects of "slow thinking" (not universal).

In my experience, (I'm no data scientist) the training of LLMs does not work that way and replication of data points aren't relevant to their veracity. In fact, from what I've seen, having the answers or positions of an LLM presented and challenged has a beneficial feedback loop to any LLM training that may include the data set in the future.

I am less confident. The following is from the 2024 study on model collapse (https://www.nature.com/articles/s41586-024-07566-y#change-history):

To sustain learning over a long period of time, we need to make sure that access to the original data source is preserved and that further data not generated by LLMs remain available over time. The need to distinguish data generated by LLMs from other data raises questions about the provenance of content that is crawled from the Internet: it is unclear how content generated by LLMs can be tracked at scale. One option is community-wide coordination...

The LLM in training does not know what is LLM content. The paper suggest only "community wide coordination", which community would be competing commercial interests.

[jpanhalt]

I don't have a problem with posting BOT replies so long as two conditions are met:

1) It must be identified as a BOT.  That is most important to me.  In this example, it seems to be a bot, but that is not explicitly stated: https://www.eevblog.com/forum/chat/plastic-identification/msg6043253/#msg6043253

2) BOT replies are often very lengthy (as the examples posted in this thread).  If you are going to post a BOT's comment, please cut to the important parts.

Since EEVBlog keeps tracks of posts and likes, perhaps it ought to add BOT as a virtual member so its posts and likes can be tracked? 

[hans]

Answering on a forum is not mandatory. If anybody feels that an OP doesn’t show enough effort, one is free to not reply. Yes, remaining silent is an option! Or one may ask for that effort. Or outright send the person to this post.

It is not an excuse for conveing generated content as a meaningful, explanatory reply.⁽¹⁾

Agreed, I'd rather not see someone posting an output of "oh I threw that into Chatgpt maybe it helps" as a way to get a technical discusion answered. A discussion can get derailed if this is posted between a high quality discussion, and the AI-poster has literally zero clue whats going on but somehow feels desperate to reply (seen this happen on other forums in some slow-chat topics)

On the other hand, it also depends a lot on the OP post. If OP posts a homework-style question or something similar like "I just made my first Ne555 pcb, now I want to build my smartphone. Please help how to do this!" then I would happily reply with ChatGPT. I mean; OP seemingly throws in zero effort, why should I put in any effort? Don't put in significantly more effort than the person you're helping (unless your goal is to also help yourself in some [other] way).

But indeed.. replies are always optional.

[EEVblog]

In this case this is a legit user, a uni student, but it's a bit lazy to just in this case reply with what AI produced and add no other commentary.

[kite31]

One hopes that is not indicative of their university work.

[golden_labels]

I didn’t want to say the user is a bot! If that was the case, I would not start a thread, but instead just report them as a likely power-posting bot to mods.

It’s clear to me that it’s a human. But instead of engaging and delivering their own responses and opinions, they just pasted output from an LLM as their own post.

[coppercone2]

I feel like they were trying to hit my threads with a defibrillator 

I think the person was interested but it had 0 interest. Many of my things are stuck in limbo or cooling down, eventually I will do something but it got too bizarre and I need a break, otherwise it turns into a stupid money pit. so progress is slow (for metal, electrochemistry, ceramics and plasma things)

someone is trying to wake a AI powered terra cotta Frankenstein monster