Announcement Official Response to "Allegations of AI misuse in PF"

[Greg Bernhardt]

Dear Physics Forums members,

I want to address a serious mistake that has impacted our community’s trust. A deeply regretful error of judgment that I alone am responsible and accountable for. No other staff had any knowledge.

During the past several months, to counter declining engagement, I experimented with generative AI, intending to add value to "thin threads", especially those with 0 replies.

The thread AI experiments were as follows:

1. Discussion summary above the OP
2. Topical FAQ at the bottom
3. Bot replies to old 0 reply threads

A full stop that these experiments are clearly against our long-standing mission, values, and guidelines.

For experiment #3, I used real accounts that I considered to be decade-old spammers as best I could tell, but I can't be 100% certain. I categorically regret and reject such a shameful and unethical decision.

Moving forward, please know that all the generative AI components have been removed along with thread content cleaned and scrubbed. They will not return. The rest of the staff has been graciously providing feedback and problem-solving ways to ensure better decision-making and accountability. Stay tuned for more news on that! Our commitment to an authentic human-driven community remains and will remain paramount.

I take your trust and the value-driven mission of PF with the utmost seriousness and will work hard to heal the damage. I am deeply sorry for this error in judgment that the community and its members, whom I hold dear, do not deserve.

I welcome any questions and feedback.

 

[pines-demon]

Thank you Greg for being clear about it.

I imagine this is not the end of AI test here or outside PF. Surely as AI tools get better, these tools can become helpful to promote activity and make threads easier to find/enter. I hope that you can find the right combination of tools, resources, and transparency, you are not the only one trying to align the AI.

 

[AlexB23]

Dear Physics Forums members,

I want to address a serious mistake that has impacted our community’s trust. A deeply regretful error of judgment that I alone am responsible and accountable for. No other staff had any knowledge.

During the past several months, to counter declining engagement, I experimented with generative AI, intending to add value to "thin threads", especially those with 0 replies.

The thread AI experiments were as follows:

1. Discussion summary above the OP
2. Topical FAQ at the bottom
3. Bot replies to old 0 reply threads

A full stop that these experiments are clearly against our long-standing mission, values, and guidelines.

For experiment #3, I used real accounts that I considered to be decade-old spammers as best I could tell, but I can't be 100% certain. I categorically regret and reject such a shameful and unethical decision.

Moving forward, please know that all the generative AI components have been removed along with thread content cleaned and scrubbed. They will not return. The rest of the staff has been graciously providing feedback and problem-solving ways to ensure better decision-making and accountability. Stay tuned for more news on that! Our commitment to an authentic human-driven community remains and will remain paramount.

I take your trust and the value-driven mission of PF with the utmost seriousness and will work hard to heal the damage. I am deeply sorry for this error in judgment that the community and its members, whom I hold dear, do not deserve.

I welcome any questions and feedback.

You are forgiven. AI is fascinating stuff. Transparency is key to AI, which is why I run open source AI on my laptop (Mistral AI). Now, I do not use AI on this site, but it is fascinating stuff. Great for summarizing science articles when using retrieval augmented generation (RAG) on GPT4All, an AI front-end.

Prompt: "Please provide a concise summary of this article, highlighting the key findings, research methods, and major conclusions, using your AI capabilities to extract the most important information."

Screenshot of AI reply:

 

[pines-demon]

You are forgiven. AI is fascinating stuff. Transparency is key to AI, which is why I run open source AI on my laptop (Mistral AI). Now, I do not use AI on this site, but it is fascinating stuff. Great for summarizing science articles when using retrieval augmented generation (RAG) on GPT4All, an AI front-end.

Prompt: "Please provide a concise summary of this article, highlighting the key findings, research methods, and major conclusions, using your AI capabilities to extract the most important information."

Screenshot of AI reply:
View attachment 357188

I am not a fan of AI summaries for the moment, most of the time it hallucinates stuff that is not in the article and misses key info.

 

[AlexB23]

I am not a fan of AI summaries for the moment, most of the time it hallucinates stuff that is not in the article and misses key info.

I agree for the most part. AI is getting better every month, but yes, hallucinations still occur. RAG helps reduce hallucinations, but they can still happen.

 

[Orodruin]

As far as AI goes, I can share a teeny tiny tidbit of information:

Just last week I was contacted by the publisher of one of my books asking me to sign an addendum to the author agreement whereby the publisher may licence the book for the purposes of AI training. For better or worse (hopefully better), LLMs are likely to soon be in business when it comes to providing high-quality textbook level replies ...

 

[Borek]

For better or worse (hopefully better), LLMs are likely to soon be in business when it comes to providing high-quality textbook level replies ...

Problem is not with the quality of the material used to train LLMs, problem is LLMs are trained to deliver AN answer, and not THE answer. They will probably get there, but not without changes to the paradigm.

But we digress.

 

[Bandersnatch]

I do feel this flirt with the AI was a mistake, quite apart from the case of bad judgement in implementation. Sure, forums seem to be dying, but such is the way of things. I really don't think replacing the dearth of humans with bots will achieve anything other than accelerating the demise. I know I'm not visiting here to interact with a robot, and I suspect it's a common sentiment.

 

[Charles Link]

@Greg Bernhardt You've done and continue to do a great job. Keep up the good work. We seem to be seeing some decline in the viewing audience over the last couple of years, but that could be very temporary. I think we all need to keep doing what we do best, and things should ultimately fall into place.

On another note, I really enjoyed this year's voting back in January. The Physics Forums continues to give me a place to interact with others with similar backgrounds. I am retired and can not find any physics or math people at the neighborhood Starbucks. Your Physics Forums is truly a wonderful thing. <3 <3

 

[erobz]

I know I'm not visiting here to interact with a robot, and I suspect it's a common sentiment.

Besides the obvious fact that I'm dense and slow, how do you know I'm not A.I.?

For the record: I don't know that I'm against teaching the successors of intelligence...at least our efforts will be in the fossil record!

 

[erobz]

but that could be very temporary.

Once they cut its leash the site could see an uptick for a bit while it spars with us.

 

[Bandersnatch]

Besides the obvious fact that I'm dense and slow, how do you know I'm not A.I.?

I don't. But that's true of everyone - solipsism is not a new idea. I give the benefit of the doubt based on credences derived from a set of priors I inherited from the pre-AI world. Each hiccup, such as the topic of this thread, erodes them.
I worry one day I'll wake up in that meme with the two astronauts looking at Earth - wait, it's all robots? Always has been.

 

[pines-demon]

This is a shower thought, but maybe one day we could have an LLM or AI forum to discuss how to use these tools.

 

[AlexB23]

This is a shower thought, but maybe one day we could have an LLM or AI forum to discuss how to use these tools.

That would be a great idea. :) The AI software front-end that I run will eventually get a new update which would add a new embedding model to it, so the AI can learn documents better. Would be cool to have an AI subforum or forum to discuss how to run local AI.

 

[Greg Bernhardt]

That would be a great idea. :) The AI software front-end that I run will eventually get a new update which would add a new embedding model to it, so the AI can learn documents better. Would be cool to have an AI subforum or forum to discuss how to run local AI.

What prevents us from starting these threads in Comp Sci or Computing forums?

 

[AlexB23]

What prevents us from starting these threads in Comp Sci or Computing forums?

That could be a good threat to start up in comp sci or computing.

 

[phinds]

That could be a good threat to start up in comp sci or computing.

Yes, I agree ... AI is a threat.

 

[AlexB23]

Yes, I agree ... AI is a threat.

Haha, I meant thread. :)

 

[erobz]

I think it's going to be extremely easy to believe you understand something in the future or humanity is not going to care about that aspect of it at all. Maybe the only thing that will be of value is what you can achieve with AI.

 

[AlexB23]

I think it's going to be extremely easy to believe you understand something in the future or humanity is not going to care about that aspect of it at all. Maybe the only thing that will be of value is what you can achieve with AI.

Who knows, one will have to see what the next 10-20 years bring.

 

[Office_Shredder]

I don't. But that's true of everyone - solipsism is not a new idea. I give the benefit of the doubt based on credences derived from a set of priors I inherited from the pre-AI world. Each hiccup, such as the topic of this thread, erodes them.
I worry one day I'll wake up in that meme with the two astronauts looking at Earth - wait, it's all robots? Always has been.

The fact that ai content designed to basically never be read by anyone got spotted is a pretty good sign that we aren't overrun yet in my opinion.

 

[Orodruin]

The fact that ai content designed to basically never be read by anyone got spotted is a pretty good sign that we aren't overrun yet in my opinion.

How do we know that the blogger is not a LLM running amok?

 

[Rive]

Moving forward, please know that all the generative AI components have been removed along with thread content cleaned and scrubbed. They will not return.

I missed the 'event' but based on the traces left, I think it wasn't that making a flirt with AI was wrong, but the way it was conducted.
- a contrasting AI content may still have an educative value at this point: since the actual policy is based on the inadequacy of the thing, and that was clearly displayed, though not clearly identified
- as things will change, it's still better to have a finger in the waters just in case.

 

[tylerkelley1980]

I appreciate the transparency in this response. Allegations like this can easily spiral, so having a clear stance on how AI tools are monitored and limited is important. If anything, the real issue is users misrepresenting AI-generated content, not the tech itself. I hope the mods continue to refine policy without stifling discussion or experimentation.

 

[javisot]

I'd like to offer positive feedback, the idea seems good to me.

In every physics forum I know, there are threads that comply with all forum rules, contain reasonable questions, and still get 0 replies. The number of valid threads without replies keeps growing, which seems unnecessary to me in the age of AI. Behind those valid threads is a person who needs help.

PF could have an AI do this work, generating replies that are validated by forum experts using a scoring system, remaining published if they exceed a certain threshold (for example).

Or we can simply let the number of valid threads without replies continue to grow..



*Bonus points for the AI: it has infinite patience and politeness.