To be honest, I think providing inaccurate information is largely an alignment issue. For instance, when I ask:

GPT-4 responds

But again, if i asked it for the lyrics like you did, it does indeed make something up:

So its basically trying to give you what the lyrics should be, even though it doesn't actually know. Its trying its best mate. Dont be mad a the machine, but instead try to find where you went wrong asking the machine the question. For these systems, prompts do indeed matter.

I'm not saying these system are perfect, but these strawmans (which are largely upvoted by the unknowing) i think are largely driven by a mixture of bias and fear. Understanding how these models arrive at specific outcomes is up for debate. Researchers understand the general concept of attention, as well as how to implement and train models that use it. However, interpreting the attention weights and understanding their exact influence on the model's decision-making process can be quite challenging. There is still much to learn about how attention mechanisms and other components in deep learning models work together to produce results. Pretending otherwise is intellectual fraud. Whenever i see people that claim they 'know' exactly how these machines work, i know they're full of shit.

is a handwoven way of saying "I dont know, but really i do" without a true understanding of what the hell youre talking about. It is true that these models are, in essence, complex mathematical constructs that rely on statistical patterns in data. This description, however, doesn't capture the full depth and intricacy of their inner workings. Even experts in the field don't have a complete understanding of every aspect of an LLM's behavior, particularly when it comes to interpretability and explainability.

While I agree with you that while US social media platforms like Meta and Twitter definitely have their own challenges, the Chinese government has a history of exerting control over media and shaping narratives to advance their agenda. It's also worth noting that the US government does not own Meta, whereas the CCP can certainly exert control over tech companies operating in China.

I agree that blindly consuming propaganda, whether it be from the US or China, is not healthy. However, it seems like your reply does not take into account the points that were discussed and instead resorts to ad hominem attacks. Let's focus on having a productive conversation about the issues

While you raise valid concerns about the questionable efficacy of banning TikTok due to concerns over targeted spying and mass surveillance, we cannot overlook the potential for China to have access and control over one of the largest propaganda machines in history. This is a significant concern, as TikTok has the ability to shape opinions and spread disinformation to millions of people worldwide.

Additionally, it's important to recognize that China has long restricted access to international social media platforms in the name of national security. China places great importance on controlling the flow of information, and if we want to preserve our democracies, we must also consider similar measures. This does not mean that we should inhibit freedom of speech, but rather restrict the narratives of foreign, non-democratic governments through social media platforms. While we should approach the US-China relationship with nuance, if the public is readily embracing the narrative of spying, we need not dismantle it if it serves the goal of protecting American citizens.

We should not underestimate the potential for propaganda and censorship. It's crucial to address these issues while avoiding severe harm to US-China relations. However, this requires a degree of give and take, as we cannot claim that we're not allowing China to compete while at the same time, they restrict access to our media companies. Ultimately, we must work towards a solution that prioritizes the safety and security of citizens both in the US and abroad.