Yes! “AI” defined as only LLMs and the party trick applications is a bubble. AI in general has been around for decades and will only continue to grow.

Yeah, they’ll probably have to check everything. Though, I wonder if even just checking that everything is good to go would save time from manually re-writing it all. While it may not be a smashing success, it could still prove useful.

I dunno, I’m interested to see how this plays out.

I think this is an interesting idea. If they’re able to pull it off, I think it will cement the usefulness of LLMs. I have my doubts, but it’s worth trying. I’d imagine that the LLM is specially tuned to be more adept at this task. Your bog-standard GPT-4 or Claude will probably be unreliable.

I can see the allure for places wanting to keep certain trouble-makers out as a precaution, but this gets so close to a privatized social credit score that it’s beyond uncomfortable.

Depends on your skills. Documentation is always useful. If you have language skills, translation of documentation or helping create language packs/translations.

That’s just off the top of my head. I’m sure if I thought about it, I could come up with more.

Yeah, I’m in the same boat.

There’s also PeerTube, the Fediverse counterpart to YouTube. Unfortunately, while there’s some good stuff you can find (and some re-uploads of YouTube), there’s just not as much content. I’d imagine the userbase is pretty small, too.

I use a cheap VPS to host my email server. It’s a bit easier than running it solely at home, but there’s a lot of annoying work to “verify” yourself. Once you get your DNS records good, you shouldn’t be blocked after that (unlike a home server). It only costs me $5/month plus the domain, which I think is money well spent. Doing the admin work to make sure I’m secure still needs to happen, but I don’t mind that work and find it fun.

YES! I study AI, and this is exactly how I feel!

Side note-One of my favorite things to do is ask people what their use case for using AI is, and watch them sputter out “uh…emails and productivity and things.”

The original paper itself, for those who are interested.

Overall, this is really interesting research and a really good “first step.” I will be interested to see if this can be replicated on other models. One thing that really stood out, though, was that certain details are obfuscated because of Sonnet being proprietary. Hopefully follow-on work is done on one of the open source models to confirm the method.

One of the notable limitations is quantifying activation’s correlation to text meaning, which will make any sort of controls difficult. Sure, you can just massively increase or decrease a weight, and for some things that will be fine, but for real manual fine tuning, that will prove to be a difficulty.

I suspect this method is likely generalizable (maybe with some tweaks?), and I’d really be interested to see how this type of analysis could be done on other neural networks.

Blue or green for me. Never could find a proper teal folder.

You’re getting downvoted, but you’re right. And that is the reason that using proprietary software and SaaS is a problem. If I’m only buying the right to use a copy of something as a company sees fit, then I’m not really buying anything. I’m essentially paying a company a tribute to use their software in their way.

Decades ago, it was the same way, but it felt different. We got physical media, and we could do what we wished with the files: modify them, delete them, etc. Hell, the EULAs for some '90s and early '00s software even said you could use the software in perpetuity, and we could use software in anyway we saw fit. The biggest constraint was on selling copies. Back then, and even now, that seems pretty reasonable. (Though, as an aside, it would have been better to also get access to the source code, but I digress.)

Now, we have to use company’s software exactly how they want us to use it. Personally, I refuse to go along with this (as much as I can), so I have migrated most of my digital life to FLOSS.

No lie, that actually sounds kinda good and I want to try it.

Like @Crack0n7uesday@lemmy.world said, it’s on government phones. The thinking goes that TikTok, which is a Chinese company, is exporting too much data from US government devices. In other words, the government is worried the Chinese are spying. Given the amount of data that the TikTok app actually collects, the fear is probably not unreasonable. All corporate-owned social media collects way too much data, but TikTok really is next level from what I’ve read.

I think a big difference, though, is that there is political force to ending TikTok. The US government has no major issues with Twitter, Facebook, Reddit, etc. existing. Remember, there’s actual legislation banning TikTok. Whether that makes a real difference or not, well, I guess we just wait and see. Personally, I think they all should go down in flames.

You’re not wrong. I remember how Bush, McCain, Romney, Obama, Clinton, and others were called Nazis at different points. While it was never really taken seriously then (as it shouldn’t have been), the term has become virtually meaningless. Where the term was reserved for the worst-of-the-worst, for years, it was invoked at the slightest disagreement. Now that there’s a literal Nazi-adjacent person running and getting called out for it, it falls flat.

Building a signup wizard to use that information to select a instance would seem to be the best approach.

That’s actually not a bad idea. I’m not on board with mining contacts, but I think there’s a simple, transparent way to do this that can actually be fun: a personality quiz. Sure, if someone knows what instance to join already, they can override this. But if they don’t, they get like five questions, and then they are matched to an instance.

I think what they’re saying is that Americans don’t pay attention and forgot how terrible the Trump presidency was because it’s been a few years. Most people think that “we’re better now” and any major issues have abated without understanding that nothing has fundamentally changed. Because of all that, Trump will win the election. The DnD portion of the post is just what got OP to think about this.

Sad thing is that there’s merit to the argument. It’s the old trope of “Americans have short memories.”

Except that scaling alone won’t lead to AGI. It may generate better, more convincing text, but the core algorithm is the same. That “special juice” is almost certainly going to come from algorithmic development rather than just throwing more compute at the problem.

I mean, that’s more-or-less what I said. We don’t know the theoretical limits of how good that text generation is when throwing more compute at it and adding parameters for the context window. Can it generate a whole book that is fairly convincing, write legal briefs off of the sum of human legal knowledge, etc.? Ultimately, the algorithm is the same, so like you said, the same problems persist, and the definition of “better” is wishy-washy.