Submitted 1 day ago by FallenWalnut@lemmy.world to technology@lemmy.world
cross-posted from: lemmy.world/post/43768262
Some may have believed they were against AI being used for war… they just don’t want it to make the final kill decision.
The argument for them is that AI in the military was inevitable, so their position is a reasonable one.
Soldier: “Help me find my targets in this crowd”
*AI highlights various targets spread around the crowd AI: “I have highlighted your targets for you. You can pull the trigger whenever you’re ready. 👍🔫💥”
*Pulls trigger without double checking anything. 10 random civilians are murdered. Soldier: “You targeted the wrong people! You were supposed to only target enemy combatants!”
AI: “You’re absolutely right. I targeted 10 people in the crowd at random. Would you like me to try targeting enemy combatants this time?”
Anthropic had only 2 rules
No fully automated killings
No mass surveillance
You could automate killing with an approval by someone but trump still wanted to lift these restrictions
No domestic mass surveillance. So fuck everyone not from the US.
Well yeah, of course we’re spying on people who aren’t citizens. How could we not? Protecting the children and whatnot
No (fully) automated killing (yet), but damn did they want to implement it.
The CEO was VERY clear about this in his interview with Larry Ellison’s CBS News.
My understanding is that it’s not military use broadly that they object to but the use of their systems for the development of fully autonomous drones.
This is incorrect.
Anthropic contradicts this, showing an aggressive willingness to work with the Trump department of “War.”
Even fully autonomous weapons (those that take humans out of the loop entirely and automate selecting and engaging targets) may prove critical for our national defense… We have offered to work directly with the Department of War on R&D to improve the reliability of these systems
Here’s the full quote including the parts you accidently left out.
Even fully autonomous weapons (those that take humans out of the loop entirely and automate selecting and engaging targets) may prove critical for our national defense. But today, frontier AI systems are simply not reliable enough to power fully autonomous weapons. We will not knowingly provide a product that puts America’s warfighters and civilians at risk. We have offered to work directly with the Department of War on R&D to improve the reliability of these systems, but they have not accepted this offer.
Still, Skynet downgraded to ad-interrupted AI girlfriends may be a less scary skynet.
Is there a “good-ish” AI chatbot? I uninstalled chatgpt and started using claude over the weekend 🤡
I don’t use it that much, maybe a couple of questions a week and some help understanding Japanese grammar
Depends on your definition, Anthropic has been the somewhat less evil on the scene, doing a lot of research to actually understand what they’re building instead of just making bold claims whenever they launch a new model, but it’s still relative to huge AI companies. The more ethical choice would be local models, again depending on what you see as the ethical issues of AI
Depends on your definition of “good-ish”. Do you mean:
Running one locally on your own hardware would likely reach “good-ish” with some sacrifices against performance/accuracy (unless you’ve got a lot of expensive hardware to run very large models). As far as ethical origins, there are few small models trained on public domain/nonstolen content, but their functions are far more limited.
I mean good-ish in the lesser-evil type of thing. I don’t expect any of those to be 100% ethical but there are some that are a lot worse than others
I don’t really have a computer capable of running a local AI. I have an i3 laptop from around 6 years ago with 12gb of RAM and integrated graphics
I’m quite happy with Mistrals LeChat. I have not done much research on Mistral but from the headlines I reas they don’t seem like bad guys.
If you can’t validate the answers it gives, I would recommend not using it. It could be giving you complete nonsense in Japanese and you’ll have no way to know until years later someone looks at you funny when you say something and you explain “I learned Japanese from chatgpt”.
If you have a decent GPU, Ollama lets you run them on your own hardware
If by “good” you mean one that more reliably answers your questions correctly, then no. That’s not really what these systems are good at. They’re fully capable of giving a solid, accurate answer, but you can simply never trust it to be correct. They’re great for chit-chat and bouncing around ideas if you’re into that, but it’s not an oracle.
When it comes to translating languages, that’s one of the few things LLMs are actually somewhat decent at, and I don’t think there’s much difference between them in that regard.
No, I mean good in the sense of less evil. I don’t ask any question to a bot where I need complete accuracy
They had a long standing contract with the Pentagon. There was a weeks long fight between said Pentagon and Anthropic on how the software is licensed.
And you realize just now that Claude is used by the Pentagon?
crunchy@lemmy.dbzer0.com 1 day ago
Well yeah, Anthropic’s statement said as much. And they already had a contract with the DoD. This isn’t a gotcha.
XLE@piefed.social 1 day ago
When mainstream media and various celebrities have cast Anthropic as the “good guy”, it’s important to point out.
Serinus@lemmy.world 1 day ago
I mean, yeah. They’re the only ones trying to hold out some reasonable constraints.