Bartender change of ownership—potential security issue?

[Synchronicity]

@Glimfeather

the example of Microsoft’s chat bot Tay outputing racist and inflamatory posts with foul language

I guess they went a little overboard trying to make it sound human

[McTavish]

@Glimfeather

Apart from their scrapping of copyright content to populate the LLM base — although Common Crawl would be okay — there is the trustworthiness of the output. Recounted before the problems I expereinced with ChatGPT in particular generating incorrect information. And the example of Microsoft’s chat bot Tay outputing racist and inflamatory posts with foul language does not engender support.

All AIs aren’t created equal. It isn’t fair or accurate to denigrate all AIs because of the actions/results of a few.

[karlnyhus]

@McTavish

But human expert judgement still rules.

Yes, I’m interested in what you know and have formed an opinion on. Not as interested in what you can copy and paste from untrusted sources.

@McTavish

All AIs aren’t created equal. It isn’t fair or accurate to denigrate all AIs because of the actions/results of a few.

Artificial Intelligences are not persons with rights or dignity. Not even close. That day may come but so far it exists only in science fiction novels and the fevered brains of some of its promoters.

[webwalrus]

@karlnyhus

Yes, I’m interested in what you know and have formed an opinion on. Not as interested in what you can copy and paste from untrusted sources.

I think this is the real key. Most forums have a ban on what’s generally referred to as “low-effort content.” MPU doesn’t seem to (at least not specifically), but if we wanted an AI response from a commodity AI, any of us could cut/paste a question into it. And if we (as a collective) consistently wanted AI responses somebody could literally code a bot to reply to every question asked.

[Medievalist]

@McTavish

All AIs aren’t created equal. It isn’t fair or accurate to denigrate all AIs because of the actions/results of a few.

They aren’t AIs. They are LLMs. I have helped creater and use LLMs for very specific scholarly purposes, like comparing an anonymous early text to a parsed contemporaneous collection of known texts, or comparing multiple versions of the same text,

We create the corpus. We know the provenance of all the training materials. We don’t accept the responses as fact (they are often not valid); we research and verify.

LLMs don’t know or understand or learn. They regurgitate based on pattern matching. Commercial LLMs illustrate the basic premise of GIGO; Garbage In, Garbage Out.

Do not want.

[rkaplan]

@McTavish

All AIs aren’t created equal. It isn’t fair or accurate to denigrate all AIs because of the actions/results of a few.

No AI is sentient.

No AI output can or should be trusted without verification.

[Glimfeather]

@McTavish

All AIs aren’t created equal. It isn’t fair or accurate to denigrate all AIs because of the actions/results of a few.

Then there needs to be some objective metric for the veracity of generated output from an LLM-based AI chat bot. Something akin to the TREC metrics of recall/prescion. Until then users, especially uniformed users, will be misled by chat bot generated output and denigration of the bad results a necessary response.

[ismh86]

Ok we’re way off base now. I’m gonna close this one down.

• We’ll all see what happens with Bartender and its alternatives.
• I think folks come here for meaningful discussions and want to hear from people not software.