Using AI as a sounding board

[ivegotabike]

What happens if you use another LLM to pick holes in the points put forward by the first LLM?

Does one convince you more than the other?

 

[farfromtheland]

What happens if you use another LLM to pick holes in the points put forward by the first LLM?

Does one convince you more than the other?

Good question.
And I myself am considering trying to engage in some LLM disalogue a la 'Darkstar'.
Garbage in garbage out is the current paradigm. Not sure if I could supervene anyway.

 

[Oldhippy]

I absolutely wouldn't personally.

 

[nogoodnamesleft]

What happens if you use another LLM to pick holes in the points put forward by the first LLM?

That's easy. What happens is the geographic area starts running out of water (needed to keep all those datacentres cool) and electricity bills go up (as all those data centres massively increase demand. All of which contributes significantly to the end of humanity as we destroy our climate ... but at least we have computer based intelligence to watch and record and learn from our mistakes.

 

[PaulSB]

AI will tell you what you want to hear. Don't trust it for one moment.

 

[katiewlx]

I'm no lawyer but I thought the courts (excl. civil cases) had to prove their case against you, and thus "in my opinion and without evidence I felt ..." - that's no more than. "he said, she said".

I'd have thought if you were riding in a "careless and inconsiderate cycling" then alcohol would not be relevant. I'd have thought courts would have been concerned with your riding not any reasons as to why you might have been riding in such a manner. If you were not "careless and inconsiderate cycling" then why were you stopped in the first place?

But I'm no lawyer.

neither am I, but I do know a few of them, purely on friends terms you understand, they charge alot less that way

but the cycling under the influence law, is a purely subjective assessment on the police officer who stops you, their evidence for that charge is essentially in their opinion, and would be based on just observing either your behaviour riding the bike to perform the stop in the first place, or carrying out the discussion with you after the stop. And a police officers opinion does carry weight in a court, especially if the prosecution makes the comparison to a drunk driver.

but what Im saying is essentially if they then additionally provided more evidence beyond their simple subjective opinion, like the results of a breath test, because they simply asked the cyclist to provide a sample for one and the cyclist mistakenly thought they had to provide one, even though it wasnt legally required, that as factual evidence would then carry alot of sway in the courts opinion on the charge. treat it like you dont have to say anything that may harm your defence, but if you do it can be used against you, so you dont have to submit to a breath test as a cyclist, but if you it may harm your defence.

and thats why I think an AI would, depending on how specific the question and parameters were, respond that actually yes whilst theres no actual "drink drive" limit imposed on cyclists in law, if you were found however that may have come about, over that limit whilst riding a bicycle then yes that evidence would be used against you, and so a cyclist is arguably held to the same drink drive limits.

so the AI isnt hallucinating or wrong as such here, its simply providing you with an outcome that it tracks to be the most likely based on the inputs it has, which are failed breath tests can lead to an under the influence prosecution, and a cyclist is not prevented from giving a breath test, which isnt how humans look at the same info.

basically AI is much more literal in how it processes data, it literally relies on mathematical pattern matching, not comprehension...yet.

so you get much more binary analysis of things with an AI

 

[PaulSB]

It's not trustworthy.

I've tested it with questions to which I already knew the answer and it got them wrong.

For example, it claimed there is a statutory limit of alcohol for riding a bike and that police have the authority to require a breath test from a cyclist. It argued with me until I cited the Road Traffic Act.

I would never rely on it for health advice.

Yes, this. I spotted a bit of AI related to upcoming concert tickets. I was very interested, followed the link only to find all the dates were historical and I had attended one of them.

My suspicions aroused I asked four or five questions to which I knew the answer. AI was wrong with each answer. I mentioned this to a friend who is heavily into AI. His response was "those answers were probably correct when the LLM learned them." This was probably true. In other words it isn't intelligent but simply a large reservoir of information.

Another friend, involved in ultra distance cycling, started asking questions about training and nutrition strategy. Over the weeks he became increasingly aware the responses were mainly praising his current strategy and telling him how good he is.

Where I have seen some value is in the detailed analysis of known personal data. I asked Claude to analyse my past two years of Strava date. Six months into the period I had a major RTC that changed my approach and strategy to cycling and training. Claude got the analysis of this spot on, even down to a statement on the lines of "your determination helped you push through very difficult times." This was absolutely true, the only thing that kept me going was sheer grit and determination to get better.

I feel I would trust, to an extent, analysis of known data from a known source. When the data is simply harvested by scraping around the web it's no more than a very fast Google search from which one must draw one's own conclusions.