Please stop asking AI for legal advice.

[tantramar@zeroes.ca]

@fsinn @theleftistlawyer I’m really just going to lean in on calling these willfully stupid people “chum” — in the shark-baiting sense, not the “buddy” sense.

[fsinn@mas.to]

@tantramar @theleftistlawyer Some otherwise very smart people are unbelievably blind about this.

[alisonw@fedimon.uk]

@theleftistlawyer
"Computers can't make mistakes"? Seriously? Whoa.

[tantramar@zeroes.ca]

@fsinn @theleftistlawyer I know many smart people whose brains turn to mush the moment there are computers involved.

[tantramar@zeroes.ca]

@fsinn @theleftistlawyer Also? Smart chum and not-so-smart chum taste the same to sharks.

[andres4ny@social.ridetrans.it]

@jmccyoung @theleftistlawyer Have you ever used _ducking_ autocorrect?

[cholling@bytes.programming.dev]

@theleftistlawyer I wrote a blog post about this very fallacy some years ago: cholling.com/posts/self_driving_fallacy/

[thriftwicker@mastodon.social]

@tantramar @fsinn @theleftistlawyer

That is a lot nicer than cannon fodder, I suppose.

[whitecattamer@mastodon.online]

@theleftistlawyer I genuinely cannot fathom how these companies aren’t getting inundated with lawsuits. Just constantly getting served. Buried under mountains of papers, and that’s just the list of plaintiffs.

[tantramar@zeroes.ca]

@thriftwicker @fsinn @theleftistlawyer

[nicelymanifest@mastodon.social]

@theleftistlawyer What will exacerbate this problem is the chronic loss of critical thinking across the land.

Too many allow themselves to be driven by the herd - they see justification because 'most people do that'.

Including wearing headphones when crossing the road and not looking as your eyes are still on your phone.

Hear no car.
See no car.
...

[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)