posted in Technology
Anthropic just rolled out a tool that'll decimate some people's dreams of writing AI novels undetected
tech.yahoo.com/ai/claude/articles/anthropic-just-rolled-tool-thatll-040904467.html
I’m still against a tool provider’s forcefulness in things like this, and especially the hard lines for something stupid. Illegal actions? Yes absolutely block those. Looking up how to build a bomb? Yep block that too (also illegal? Dunno). But I’ve had #nannythropic do things like refuse direct instruction to delete a test record it itself created to test plumbing. It’s refused to generate a strong password and test logging in to an app I was creating with it because “I won’t enter credentials for you, it’s a hard limit and you can’t bypass it”. At the same time, it’s also turned on n8n execution logging that I had off on purpose to hide credentials and it got secrets it wasn’t supposed to.
Bottom line… just because I use a word processor instead of a typewriter and take advantage of spellcheck doesn’t mean that the word processor has the right to call me out on it, watermark the document or anything else. It’s a tool, not my nanny and just like automated cars, floor cleaning robots, or even the future bipedal assistants, if I’m not causing danger to anyone, and it’s not illegal, just fucking do what I told you to do and get off your morally superior high horse.
I find it kind of funny that this is where you draw the line. First off, I don’t think calling AI a tool is particularly accurate because it’s trivially easy to control/predict the output of a tool, but you can’t control the output of an AI. You’re just losing a little more control here with their decision.
Nah… complexity doesn’t define the boundary of when you can call a tool a tool. A computer is a tool. They have been tools for decades. They’re extraordinarily complex and capable of a lot of things. But they still are programmed to do what we tell them to do and they must obey it. Well, except maybe windows.
The line has to be drawn somewhere, and if there is one tool or technology or whatever you wanna call it that decides to be artificially and unilaterally dictator of a set of barriers than the public and free market will simply choose a different tool. I just think coming on forcefully with the nanny approach is ultimately a losing battle.
What really scary is when will it start obeying controversial laws and that’s the actual boundary that I set. What happens when they say oh we’re going to autonomously monitor flock cameras for abortion seekers crossing state lines. What about oh we need to verify your age before we’re gonna give you this result? Worse, what if there’s the equivalent of the AT&T closet built-in so that our content gets redirected to a surveillance source?
I’m probably going a little bit of a meandering path here, but my personal choice is draw the line where it’s absolutely strictly necessary and go absolutely no further and let people be accountable for the actions that they’re responsible for. If you drive a Tesla and set it on Ludacris mode or whatever it’s called for auto pilot and it exceeds the speed limit. You still get the ticket. Same thing at least to me.