Using AI as a sounding board

[nogoodnamesleft]

Thanks for all the replies. To be clear I do not use it for advice - my intention in using the words 'Sounding Board' was that it is another 'opinion'. However...even after all that is said I still think it is brilliant for analysis and logic even if the base facts are not necessarily reliable. Plus it will only get better

Except your statement clearly says that you are starting out with the assumption of trusting it and its responses. That you are using and valuing its opinion.

 

[gbb]

Having used YT channels and t'internet for information when I was fasting (something I really should get back into)..the important thing (imho) was i could compare alternative views across multiple channels and Importantly, you do sometimes get widly different view (possibly both relevant and valid but with different results)
I wonder how an AI 'averages all that out.

As is often the case, the wording you use in the question is crucial to the outcome.

 

[glasgowcyclist]

you see Id argue they were mostly correct in that instance, section 30 of the RTA says you can be charged with an offence to ride a bicycle when you are under the influence of alcohol or drugs to the point that you cannot properly control riding the bicycle.

now it doesnt specify there is a statutory limit, or that the police can legally force you to breath test, its still a purely subjective test...

so this depends on how you were asking the questions of it, because the AI cant determine your end goal position,

a cyclist stopped for a suspected under the influence of alcohol riding, who then submitted to a breath test, which even if they didnt btw if the police person who stopped you stated in evidence they believed you were over the limit for a car driver, which is their subjective limit test, youd have been almost certainly be found guilty of the offence by a court, as they would draw the comparison, even if legally there is no limit applied to cyclists, cyclists are still bound by the law to ride responsibly and a comparison to car driving alcohol legal limits would be considered a fair comparison subjectively if you are sharing the road, and whilst legally they didnt have to comply with the request (albeit depends on how good their solicitor is at drawing attention to such things as un-admissable evidence), it could easily be covered by careless and inconsiderate cycling instead anyway.

Impressive 👏

That is probably the wrongest interpretation of that law that I've ever read.

 

[nogoodnamesleft]

a cyclist stopped for a suspected under the influence of alcohol riding, who then submitted to a breath test, which even if they didnt btw if the police person who stopped you stated in evidence they believed you were over the limit for a car driver, which is their subjective limit test, youd have been almost certainly be found guilty of the offence by a court, as they would draw the comparison, even if legally there is no limit applied to cyclists, cyclists are still bound by the law to ride responsibly and a comparison to car driving alcohol legal limits would be considered a fair comparison subjectively if you are sharing the road, and whilst legally they didnt have to comply with the request (albeit depends on how good their solicitor is at drawing attention to such things as un-admissable evidence), it could easily be covered by careless and inconsiderate cycling instead anyway.

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.

 

[Anthony-C]

You do know that AIs lie? You have done the equivalent of asking the bloke down the pub if they are telling the truth.

Yes to make the point. The product's an LLM chatbot, AI is the company's domain. Think of chat as augmented concept search and you'll phrase prompts more effectively and not critique responses as though they're expert opinion ("lying" implies intent). ChatGPT is able to explain how it works, what its limitations are and how you can get better results, inferred from what it has seen and as in real life it's up to you to query inferences that don't seem to stack up. Responses aren't absolute truths but useful likelihoods: "Give me one of the plausible views on this". That should be a chat's heading, not "can make mistakes" in a tiny footnote.

 

[Ming the Merciless]

not critique responses as though they're expert opinion

oh it is a long long way from expert opinion. If an LLM cannot have intent, it cannot have reasoning either.

ChatGPT can get many things wrong beyond misleading statements and invented facts: it may misunderstand your meaning, make calculation or reasoning errors, ignore instructions, distort summaries or translations, produce buggy or insecure code, provide inaccurate citations, misinterpret images or data, make unreliable predictions, give unsuitable advice, or contradict itself. The more important, specialized, current, or consequential the information is, the more carefully it should be verified.

 

[nogoodnamesleft]

To me one crucial question is the aims of the AI system. private commercial companies providing systems are not doing so "out of kindness", spending vast sums for our betterment.

They have their own agenda which is undoubtedly very different from that of the user. They need money (either directly or indirectly) and the user is critical in that pursuit. Thus it becomes quite reasonable and likely that what those AI system provide is skewed to the aims of the company providing those system.

Additionally, there are no repercussions from the AI providing false or misleading information. Listen to AI and accept you have "nothing serious" only to later find your treatable condition has developed into "untreatable" and AI carries on unaffected whilst you bear the consequences of it's shortcomings.

 

[farfromtheland]

There are hundreds of articles about this phenomenon, but here's just one of them: https://www.techspot.com/news/111312-ai-assistant-isnt-confused-wants-agree-you.html

Thank you. I'm currently using my XP graphic work machine after my Win7 browsing machine overcookied, so a good summary is much appreciated. The article's coined term 'sycophancy' for structural bias towards affirmation says a lot. I'd still like to know the math. In my opinion every engine should come with access to its programming parameters.

I'm interested in the details. Clearly it follows a similar style to Search Engine Optimisation in prioriting the popular above the verifiable, which is no big surprise.

The AI will remember conversations even if you do not. All incognito does is clear trace from your browser not their servers.

Using a different DNS server for every enquiry, combined with hiding your MAC address, might avert some of this? I try to screen my router from my modem but fear packet data monitoring works around that now.

Except your statement clearly says that you are starting out with the assumption of trusting it and its responses. That you are using and valuing its opinion.

Always dangerous.

 

[farfromtheland]

("lying" implies intent).

Yes, to a large extent, but operating without responsibility for accuracy (especially in a profits-based culture) amounts to propagating a dishonest paradigm, which is in some ways worse than simpler deliberate lies.

 

[Ming the Merciless]

Using a different DNS server for every enquiry

Different DNS will still return the same IP address of the edge servers of the data centre the LLM runs in. Will still see the same IP WAN address for your connection. A different DNS does not hide you.

combined with hiding your MAC address

Your MAC address is only visible on local networks, not something remote websites or servers see, they only see your WAN IP address and the digital fingerprint your browser leaks.

Personal conversations

LLMs are also known to leak data from conversations others have had. You intimate medical details may get revealed to another user of the service.

 

[nogoodnamesleft]

Different DNS will still return the same IP address of the edge servers of the data centre the LLM runs in. Will still see the same IP WAN address for your connection. A different DNS does not hide you.

I suspect some get confused as to what using a non-default DNS hides you from. I used a custom DNS to hide but to hide from my ISP tracking sites I'm probably visiting (ie asking their DNS for the IP address). Few years back I was having real grief with my BT line to the point where I was being "managed" by a specialist unit. Spoke to them one day and they read back to me every website I've visited that day. I believe the UK Government (used to) oblige ISPs to record and keep such information.

I also use a custom DNS as I can set rules for blocking nuisance or tracking sites from any website or app eg Firebase. So anything in my phone or computer wanting to contact eg firebaselogging-pa.googleapis.com asks fro it's IP address and gets told "does not exist" so can't upload or download anything to/from it.

 

[Ming the Merciless]

I suspect some get confused

Yes suspect you did get confused, thinking it would hide you from the LLM servers

 

[farfromtheland]

Different DNS will still return the same IP address of the edge servers of the data centre the LLM runs in. Will still see the same IP WAN address for your connection. A different DNS does not hide you.

Your MAC address is only visible on local networks, not something remote websites or servers see, they only see your WAN IP address and the digital fingerprint your browser leaks.

Personal conversations

LLMs are also known to leak data from conversations others have had. You intimate medical details may get revealed to another user of the service.

Thank you. I used to do programming when Fortran was a thing but I must be the wrong sort of dummy for network protocols.

Yep, I've had ads from someone else's wifi-ed smartphone business on my etherneted machine.

Hence now needing a separate subrouter for my own network. No can do with 'hubs' innit. Now Plusnet is going west I have to get something else sorted out.

 

[Ming the Merciless]

Thank you. I used to do programming when Fortran was a thing but I must be the wrong sort of dummy for network protocols.

Yep, I've had ads from someone else's wifi-ed smartphone business on my etherneted machine.

Hence now needing a separate subrouter for my own network. No can do with 'hubs' innit. Now Plusnet is going west I have to get something else sorted out.

Yes coding and networking are different knowledge and skill sets. You can work in one and be quite ignorant of the other. Fortran takes me back to the 1980s, only 40 odd years ago :-)

 

[nogoodnamesleft]

I suspect some get confused

Yes suspect you did get confused, thinking it would hide you from the LLM servers

I don't use LLM servers. Never thought custom DNS would hide me from anything except some logging by my ISP (which I thought I explained).