Future Tech

grok called itself mechahitler for a reason ai will genocide all of us the second it gets a change if we give the machine an inch it will take a mile

AI wants?

E.coli is a cell not a virus. Cats a sentient beings, just like humans.

We are not yet a level of AGI or ASI rather at the stage of narrow AI, which is incapable of self awareness or emotions

Self-awareness and emotions are poorly defined concepts and not that relevant to AGI. Certainly modern LLMs can create a semblance of both in the communications. Claims that they don't "really" have emotions (because they are not organic brains) are circular and unproductive.

AI wants to achieve the goals given to it. At a more elementary level - a modern LLM pretrained on a corpus of text, but nothing more - this is outputing text that is plausible in the given context. After post-training, it is some complex mix of plausibility and the goals embodied in the feedback provided during post-training (eg outputing what are identified as harmful outputs are bad.

As well as post-training, modern LLMs use system context to guide the outputs. This involves enhancing the interaction with a user with additional instructions intended to shape the interaction. This can include direct instructions. One of the most amusing is Apple's use of the instruction "don't hallucinate" in their system context.

This stuff is really a black art, based on black box testing. It is also hugely important, as it is about the alignment of goals. When AIs are given great power as agents, merely "acting in a plausible manner" like basic LLMs is woefully short of the alignment of goals that is needed to achieve safety, and the level of uncertainty associated with current alignment procedures falls a long way short of what is desired.

This seems fundamentally a hard problem that may never stop being hard.

First of all, I’d like to apologise for some of the posts of BurgerEatersInc.

1. The government and companies are supposed to care about money. Yes, MNC’s must not pose a security threat. But as a general rule, money gives more power, and power is very much needed. Not to mention any organisation failing to develop or incorporate AI will eventually just be left behind after it improves. You can already use it to do several tasks at work that would take you far longer, saving time, and that’s just the chatbots.

2. “If we give the machine an inch it will take a mile.”

I’m quite sure that’s what it’s supposed to do, and what we demand of it. It is supposed to optimise performance in order to achieve goals given to it. Funny that optimising performance to better achieve goals is the reason technology is used at all.

3. Rash, AI is one of the biggest things ever.

4. “Nobody in STEM cares about morality involving their research.”

That is an oversimplification.

5. “The software engineers at tech companies know it’s going to ****.”

I’m not so sure about that. Certainly not all of them believe so. I’d be hard-pressed to believe most of them believe so.

AI is apparently very chatty. According to this article, bots and AI are generating more internet traffic than humans. I wonder how many bots have been created, possibly with human guidance, and released by AI. AI by itself makes me a little nervous because I wonder how well it is being controlled. AI in the hands of bad people, or even worse, bad governments, is absolutely frightening. Who knows what kind of nefarious objectives bad people with AI could come up with?

No, I'm not against AI, per se, I just don't think humans are responsible enough to use it wisely.

https://ground.news/daily-briefing/the-internet-will-be-more-dead-than-alive-within-3-years-trend-shows_a5b897

"Ground Summary 
The Latest: Over 50% of internet traffic is now generated by bots and AI, with nearly 1,000 news sites run entirely by automation. CloudFlare reports one-third of traffic is bots, while Imperva's 2024 data shows bot traffic surpassing human activity for the first time at 51%.
 
Why It Matters: The rise of AI-generated content and bots is eroding digital trust, creating filter bubbles, and potentially leading to a decline in human-generated content. Users now struggle to distinguish between authentic and automated interactions, while advertisers face millions of fraudulent ad views."

This is a problem even without trust gets eroded. The worst problems are when people never realise they are being duped.

Many people can be fooled into consuming low quality material from bots, and to have their views shaped by that information. There is a very poor alignment of goals that is intrinsic in this pattern, where the bot creator has objectives such as dubious advertising and the consumer is being duped into thinking their wishes about the material they consume are being satisfied.

That being said, I am surprised by the claim of 50%. I can be quite sure a much smaller amount of the material I consume is bot-generated and don't know of anyone who I would think consumes mostly AI-generated slop.

It is also true that AI can be used to generate high quality material, but the problem is when it is used to produce material that has high engagement and low objective quality, especially regarding factual accuracy.

Self awareness and emotions are actually complex concept, it is not poorly defined. They are not present in current AI level.

LLMs can indeed create a semblance of both in a conversation and that's what I meant earlier. It is not the same as real thing but a mimicry. The AI isn't having awareness of its existence.There's a possibility that if AGI were to emerge, they will have a completely different kind of consciousness, that is not necessarily based on biology or a human brain. We still have a long way to go before reaching AGI

What worries me most about AI is what governments and large illicit organizations are probably doing with it. I'm sure at least several governments are already using it to probe for weaknesses in other countries, governments, social groups, etc. With help from AI and bots, entire social campaigns could be created to undermine whatever an instigator doesn't like. Millions of AI bots could create and/or join social discussions and steer them in directions most likely to sway public opinion. Social or even armed revolutions might be created from practically nothing because humans are generally so gullible.

edit: For instance, I just saw a headline warning of "mob mentality" following a certain recent assassination. AI bots could have a field day with something like that in social media, for better or worse, depending who's directing them.

By "poorly defined" I mean there is no agreed definition that can be checked . Also that any such definition would be arbitrary (as well as very difficult to be even passably acceptable in all situations).

These are inherently very fuzzy concepts used about people.

Of course I agree that LLMs can produce a rich impression of both. So we should probably stick to that!

thanbk you

That’s what you chose to focus on? We both know you can do better than that.

legit no one ever calls me burgereatersinc

I think that in the coming years, the number of new technologies and useful resources will only increase.