shom

Here are four of the ten looping Claude user quotes on anthropic.com homepage... Mind you, these are not dynamic, they chose these explicitly. Are they trying to represent user sentiment accurately or are they reading these very differently than I am?

I went there after watching this talk: "Nicholas Carlini - Black-hat LLMs", from one of their engineers. There's definitely good work by talented and conscientious people that's going on there.

I'm rewriting this post because I'm cynical of corporate motives but I also don't think that interpreting everything cynically is helpful. Even after the VC funding runs out (hopefully before we destroy the planet and society), these tools won't disappear especially for malicious actors. So if they're also building tooling to mitigate harm / defend against threat actors, do I dare to hope they're reading the quotes the same way I am? Or is it more of:

I feel like I'm creating more dependency than knowledge.

#AI #Anthropic #Claude #Blackhat #LLM #SoftwareSecurity #Cybersecurity #ThreatActor

7:01pmWeb
Show thread65dBnoise3h ago

@shom
Conscientious people always exist, even in the worst of environments. Many a time they are useful in providing the facade a company may need to move attention away from problematic areas. But can those good & conscientious engineers define or even influence company policy?

By their nature private companies try to maximize profits. If there is no regulatory obligation for them to do otherwise, they can correctly assume that it's not their job to self-regulate. Even SCOTUS doesn't.

1/

Show thread65dBnoise3h ago

@shom
Cynicism assumes there are moral criteria that apply. Companies operate on the basis that those are embedded into regulations and the law. But that's not what the public is lead to believe; the public expects somehow companies to act morally, as if they were fellow citizens. They are not. Hence the dichotomy.

The only way for the system to work as expected is to codify people's expectations to regulations and laws. But that would be the case only in a democracy.

2/2

Show threadshom2h ago

@65dBnoise

Many a time they are useful in providing the facade a company may need to move attention away from problematic areas.

This is absolutely the reading I generally have (not exclusive to AI companies). For profit institutions will always prioritize profits over anything else (most of the time short term profits over even self-survival) and there are no shortages of examples.

I'm struggling with how if all the consentious people are rightfully not engaging and there is total regulatory capture then what does that look like in the long run. I'm quite befuddled by how callously a lot of the public uses AI without any thought given to any of the moral and ethical impacts: I'm talking about frivolous use here all the way from benign generating videos for a laugh to the revolting non-consensual porn. I don't see this same public holding politicians to account to create and enforce regulations.

I'm not discounting use of LLM as tools altogether, I'm in data and analytics and providing an interface to our business users to help craft queries so they can solve their needs (with governance and validation) isn't a thing without merit.

I'm not arguing a specific point as much as thinking out loud about how do I talk to less technically plugged-in people about LLMs now. They're actually useful in some contexts now and dismissing then outright isn't going to help those folks avoid the downsides. Hanving these conversations, thinking critically, understanding hallucinations in the machine and brain atrophy in people and then being able to non-judgementally inform and engage folks is a lot of work. I think it's work worth doing, and in a way I'm just taking Mr. Rogers' advice and "look(ing) for the helpers".

Show threadAlex White1h ago
@shom if you read the page that section links to they disclosure large amounts of both positive and negative feedback, including common concerns around AI, from a survey they conducted. So the negative comments are purposeful.
Show threadshom25m ago
@alextheuxguy yeah thanks, I was reading that. It's good of them to feature both positive and negative feedback. Regardless of the intent that is a good thing. Thanks for pointing it out.