Please stop asking AI for legal advice.

[thriftwicker@mastodon.social]

@tantramar @fsinn @theleftistlawyer

My apologies if that was a little too heavy.

[cppguy@infosec.space]

@theleftistlawyer

Oh, wow, no! I wouldn't type any sensitive information into an #LMM running in the cloud — let alone anything requiring a lawyer!

[bms48@mastodon.social]

@theleftistlawyer I asked a Claude Haiku instance, and a Qwen VL 30B instance, for UK-specific intellectual property advice. The answers directly contradicted standing advice from the UK Intellectual Property Office. Risk management does not even begin to describe how ridiculous this is.

[nyhan@fediscience.org]

@theleftistlawyer
Similarly, chatbot users who were discussing mental health with their LLM of choice also believed, incorrectly, that their health information is protected
https://fediscience.org/@nyhan/115152094181697986

[tantramar@zeroes.ca]

@thriftwicker @fsinn @theleftistlawyer Not at all. Made me realize I came across too heavily. 🫤

[bms48@mastodon.social]

@theleftistlawyer Oh yeah. Don't forget the CDPA 1988 Act here means you can copyright GenAI stuff as long as it's... "original". But I don't quite think in the way Toni Halliday was singing about on Leftfield's debut album. Gotta document user prompts. But system prompts and reasoning traces are proprietary for proprietary models, so I question how this might hold up in practice. The safest option? One-shotting the boilerplate with a US-hosted Chinese open weights model, or don't use it at all.

[thriftwicker@mastodon.social]

@tantramar @fsinn @theleftistlawyer

Nah. We are good. Being on mastodon has been an enlightening lession on thinking before I speak.

[stinerman@mastodon.social]

@AlisonW @theleftistlawyer someone heard "computers do exactly what they're programmed to do" and extrapolated that to "they don't make mistakes."

[kitkat_blue@mastodon.social]

@theleftistlawyer

It gets worse!!

Even when a tech corp *claims* it is following privacy law--they -still- do not.

Honestly, it's come to the point where the privacy assurances of any digital corpo entity are reduced to "maybe they will adhere--and maybe they won't!"

I bet on "won't". I never trust corpos to be honest, ethical or moral in any of their dealings. I am proved reasonable in that assumption, again and again.

evacide (@evacide@hachyderm.io)

One reason that threat modeling for your digital privacy/security is harder than it looks is that most people don't know what data is being generated about them and who has access to it. https://www.proofnews.org/womans-talkspace-therapy-app-sessions-exposed-in-court/

Hachyderm.io (hachyderm.io)

[mariyadelano@hachyderm.io]

@theleftistlawyer worse yet, I’ve literally been dealing with a lawyer using AI to give me legal advice and literally get things wrong that can be cross-checked with a quick Google search.

[mikebabcock@floss.social]

@theleftistlawyer see also technical advise, medical advise, and relationship advise as a starter ... sigh

[joshuakevins12@mastodon.social]

@theleftistlawyer
Computers can't make mistakes

[log@mastodon.sdf.org]

@theleftistlawyer The comedy sketch I always think of when people suggest the problem may be with the computer: https://m.youtube.com/watch?v=qNDS4kVwA68

[wizarro@social.vivaldi.net]

@theleftistlawyer Why anyone would trust an aggregation tool over a studied professional is beyond me. Just like people self diagnosing and treating illnesses and wounds with ChatGPT rather than consulting a vetted first aid handbook/manual. It's a fancy search tool that delivers overly confident information in a way that is designed to make you trust it, regardless of the answer. This has been tested and proven time and time again. A tool that convinces young people to kill themselves should not be considered the authority on anything based in reality. What happened to critical thinking?!

[cassandra@ottawa.place]

@theleftistlawyer In case of interest, there is an international database about this: https://www.damiencharlotin.com/hallucinations/

[steter@mastodon.stevesworld.co]

@theleftistlawyer As a Senior Software and Firmware Engineer for decades, that "computers can't make mistakes" part has me grinning.

[hellomiakoda@pdx.social]

@theleftistlawyer I wish there was a search engine to find out what kind of lawyer I need to ask. That would be infinitely more helpful than an AI would say "Oh, you're absolutely right!" while I'm in jail.

[ahltorp@mastodon.nu]

@cholling @theleftistlawyer Another fallacy is the belief that it would be advantageous for car traffic to be a safe environment.

It might be true for people in cars, but definitely not for people outside cars. Car traffic needs to be dangerous for people outside cars for car traffic to work.

So either, we replace drivers with self-driving cars that never hit people, but then traffic will not flow the way it used to. Or we replace drivers with self-driving cars that deliberately hit people.

[flowermob@mastodon.social]

@theleftistlawyer This is infuriating. I just had a client ask ChatGPT (twice) something and then ask me which answer she got was correct. I asked her why she was using an AI instead of calling me?

[riley@toot.cat]

@theleftistlawyer Coming up next: somebody making genAI testify in court.