HarryOru @ HarryOru @lemm.ee

Posts

0

Comments

33

Joined

2 yr. ago

I hope this doesn't come off as defensive, but I'm not sure how many times I need to repeat that I'm not at all disputing or discrediting the existence and danger of Eliza, other tools like it, and generative AI in general. Like I said, I very strongly believe that this type of manipulation absolutely happened, is still happening, and was happening even way before LLMs became a thing. I absolutely believe the public should be loudly and urgently made aware of this.

If there is anything that this Substack proves regardless of its veridicity, it's that this is just the beginning. Very soon, the internet as you and I know it will entirely cease to exist and the very concept of "truth" will follow. This is the reality we are witnessing today. So trust me, I completely understand and share your discomfort at seeing that "nothing is happening" and no one is (at least publicly) looking into this.

Still, if the meaning of "truth" is what's being weaponized here, I cannot in good conscience claim that, to my honest understanding, this specific Substack is currently proof of anything. And please do not take my word for it, keep sharing it and looking into it because I'd be the first to celebrate being proven wrong about this.

To answer your question about whether they could just do it via API, I mean, it's certainly possible, but in the same way that breaking into your own house through a window when you have the front door key in your pocket is possible. And hey, again, don't take my word for it. That is just the most likely conclusion to me, but logic doesn't apply to these people and their schemes the way it does for you and me.

EDIT: forgot to say, thanks for providing that link to your thread and collecting the relevant information. I'll make sure to read it.

Very odd though, no? Why would they use a Trump bot as the public example?

If I wanted to assume bad faith, and at this point I will every time a claim is made with no verifiable proof, I could very easily interpret that post and the presence of a Trump character file as an advertisement basically telling potential users "hey look, this is what he used to steal the election, it's possible, it's simple, it's relatively cheap and you can do it too." And if you look through the GitHub Issues for the project you'll see that quite a few people are already currently using it for who knows what purposes.

As someone who works in the field I also believe there are several red flags in the way the article is written and a bunch of contradictions. For example, "hiding crumbs in the code" would generally imply something slightly more sophisticated than a public file that the author himself tells you how to find in the repository with even a screenshot of it.

The main issue for me though is that it is highly unlikely that the owner of X would need to use an external tool that establishes a finicky integration with X from the outside to generate fake content when he obviously has control over the internal private code and data of X itself. It's beyond naive to think that if they wanted to create fake accounts, fake posts or manipulate the content users see, they would need this Eliza thing to do it effectively. This tool is clearly aimed at people who don't own the platforms they intend to use it on.

Ultimately though, while I don't believe this source and story are to be trusted, I still think everyone should know that tools like this actually exist and that something resembling the proposed narrative is extremely plausible regardless.

That is absolutely true, but the substack post as it stands provides no proof of the events it describes. Like I said, I'm not claiming that interference didn't happen, I'm not even claiming it didn't happen almost exactly the way the post describes, but none of it is verifiable through the provided "source", which is just a public GitHub repo.

This isn't actual proof of anything though. This is a publicity stunt by someone trying to promote this Eliza project, which has nothing to do with X because actual X code used for the described nefarious purposes wouldn't be on a public GitHub repository. More importantly, someone who "can't sleep at night" due to their involvement in election interference wouldn't link to the tool they used to do it and instructions on how to use it.

Having a .json "character file" for a Trump chatbot filled with Trump rhetoric is not proof of anything the writer is claiming. Anyone could do that, including LLMs.

I'm not saying similar technologies/tactics weren't actually used over the last few years, in fact, I strongly believe they were. All I'm saying is that this article is nothing but bait. Fact-checking is more important than ever now.

He did the Nazi salute twice, intentionally.

He said "my heart goes out to you" because of plausible deniability. He was giving his followers an argument to deny that he did it intentionally, both to themselves and others.

He has not directly denied it nor apologized for it.

European leaders are acknowledging it for what it is. Neo-Nazis are acknowledging it for what it is, and they loved it.

People who are sincerely on the fence about this, admitted they actually exist, need to wake the fuck up.

I know this is a joke, but as someone fluent in two languages, trust me when I tell you that translating anything Trump says effectively is borderline impossible, and that's a huge problem. It plays in his favor because people in non-english speaking parts of the world don't get to fully see how ridiculously incoherent he actually is; they only get the sanitized version of his ramblings.

Personally, it's been REALLY hard these past few years hearing and seeing the shit coming out of his mouth in its original language and at the same time not being able to at least share my worries with other people in my country without sounding crazy. It's very easy to underestimate the gravity of the situation from the other side of the world if you can't grasp how low the standards have sunk and still believe DT is just an average Republican president, and not living proof that the actual meaning of words has no value anymore.

Then simple question, why not use First Gentleman? Kamala Harris' husband was Second Gentleman and would have been first if she had won.

Going by your logic, that would be offensive to me, as a gay man.

But this kind of hypersensitivity, especially when applied regardless of context or intent, is one of the main drivers of the reactionary sentiment that is allowing the right to win elections all over the world. It's one of the reasons why the "free speech" argument has worked so well in their favor.

Currently, the only ones benefitting from this "moral high-ground" stance are fragile little baby egos like Musk and Trump who can rest assured knowing their opposition would never stoop as low as them, while they get to freely spew as much intentionally evil shit as they want.

A society where no one is ever offended by anyone is a utopia. It's as desirable as it is unattainable. I think the best thing we can do at the moment is focus on fighting back, not fighting against each other.

Missing Teen's Friends Go On Tv To Plead For Her Release, Gossip About Ugly Classmates

or

Expert Wasted Entire Life Studying Anteaters

"I waste my time in the morning and evening
\ Caught in a feeling
\ I lose my mind looking up at the ceiling
\ It's just a feeling, it's just a feeling..."
\ Part III - Crumb

"It's not meant to be a strife
\ It's not meant to be a struggle uphill
\ If you're bleeding, undo
\ If you're sweating, undo
\ If you're crying, undo"
\ Undo - Björk

"Past love, come back to yourself
\ Don't keep reaching out to him
\ He can't help you now
\ It's a past life, so come back to the time
\ It's been far too many nights, and you still cry"
\ Past Love - Kimbra

I'm not sure what you mean. Flexible OLED displays have been around for a while and foldable devices are just an example of the technology in use, but we've had them in consumer products way before that (phones with curved edge displays, for example). The potential for flexibility has always been intrinsic to OLED displays because they don't need a backlight. The reason our phones don't bend and flex like the "device" in the video isn't because of the display, but because the battery, processors, ram, speakers, ports and all other components are not flexible and won't be for a while. The device in the video does not include those, there is a ribbon cable coming out of the bottom connecting the two screens to the actual hardware.

I've had this issue with Samsung phones and tablets I owned in the past. Working as an app developer I still see this kind of problems on heavily oem-customized versions of Android.

Personally I "solved" this by switching to Pixel phones which in my experience don't slow down even after 3-4 years of usage and updates. I believe this is true in general for phones that stay as close to AOSP/stock Android as possible.

I can't speak about Flutter or React Native, but what I can say is DON'T use Xamarin Forms/MAUI. As a native Android developer I had to start using Xamarin after changing jobs and it's been one of the biggest regrets of my career, honestly. Literally nothing works like you would expect it to. I understand the idea of writing the same code twice is intimidating, but trust me, nothing beats native development. Nothing. I can say with 99.9% certainty, you will regret not going with native if (or when) your app requires any vaguely complex feature to be implemented into it. Swift and Kotlin are similar enough that you can literally write the same app natively for both platforms faster than it would take you to write them in any cross platform framework (or at least Xamarin/MAUI), unless you're making an extremely simple app with no customizations whatsoever.

I kind of gave up on Japanese years ago, but I've been trying to learn Korean for a couple months now. I started with Duolingo (great for basics like the alphabet, and in general for reminding you to practice everyday), then added YuSpeak (similar to Duolingo but way better paced/structured and with some useful systems to keep track of your weak points and review words and concepts effectively. Their Japanese course has even more features).

Although the apps are a great way to start, like other commenters said, they really promote memorizing over understanding/using the language, especially Duolingo. So the greatest addition for me so far was actually ChatGPT (GPT 4). Now I can learn a new concept on YuSpeak or Duo and then immediately go ask ChatGPT to go over it together and make up some exercises for me to do. It really works surprisingly well, and in just the few days since I've started using it, it feels like my ability to write and read the language has improved exponentially compared to when I was only using the apps. I also fare much better in the apps themselves.

Of course ChatGPT is not 100% accurate, as we all know, but it opens up so many learning avenues that it doesn't really matter, especially if you're a beginner to intermediate student. Obviously it also can't completely replace an actual teacher (on the other hand, it's available 24/7, unlike an actual teacher, and it's much cheaper too) and it won't help with listening or speaking, but using it in conjunction with apps and a YouTube video here and there really creates a pretty effective and proactive learning environment.

TL;DR: Check out YuSpeak and consider using ChatGPT 4.