I'm not sure I really buy into the way these concepts are currently defined. What progress to date was not RSI and did not build on top of earlier progress?
The main characteristic of our physiology is that of adaption, and we have fairly grounded research on progressive overload as a means to guide adaption to whichever direction we want. Should we then call humans RSI machines if they go work out or read a book, or study to get better at the activities they carry out?
Define an objective, choose an optimization algorithm, bring a loss down. If it keeps going down, do I call that RSI because the loss descended from the step that came before it?
Now, I see the nuance that the optimization algorithm changes over time, leading to interesting higher order effects the likes of which Hoefstadter describes in the GEB, but fundamentally, the idea is the same.
I also don't see what an intelligence explosion is supposed to mean. If we increase the models efficiency by 10x, and reduce costs by 10x, so we have 100x more software and math proofs being shipped by unit of time in say 2028, does that really solve all of our problems?
Say that gets distributed over the world. Do we get every human to be computer literate?
Say that gets concentrated. Do we solve problems no one has the means to even care about?
I would consider an intelligence explosion when the models tell us whether our efforts are well or misdirected as looking back from the hindsight of 10 years ahead, and capable of explaining why in simple English.
If you ask a layman, the answer would probably be a model that zero-shots any problem in any reasonable domain. E.g. zero shot Riemann hypothesis. No tricks like swarms of agents, no multiple agents correcting each other. No high, xhigh "thinking" levels. Simply provide an input, feed the input through the layers, answer out.
Yeah, you can achieve something similar if you scale the compute infinitely and intelligently search the exponentially large search space.
> No tricks like swarms of agents, no multiple agents correcting each other. No high, xhigh "thinking" levels.
Why would implementation matter? The thought experiment is that you have a black box that can do X. What is X that validates RSI/super-intelligence/etc? It's meant to clarify definitions and end the cycle of goal post moving.
I don't know about others in the thread think there definitely needs to be something done before it's too late.
I understand that many people don't think it will happen, but I can't really understand the position that you think there is a real chance superintelligence happens, but aren't concerned about it and are happy to wait passively to find out.
If you care about humanity at all it should be concerning because superint (if real) could be heavily destructive to our society.
I agree and I'm happy to hear out this kind of thing, my "I'll wait and see" attitude comes when the expected response is that I accept that the AI has "won". Thats pretty vague because the expected response, if there even is one, is pretty vague, but it generally amounts to something like that.
This would also be useful for people who can make mouth movements but not use their vocal chords any more, like people with tracheostomies. It could also be a useful tool for training deaf people/people with partial hearing to speak as many struggle to make the right shapes internally.
The difference would be that the airgapped servers would be connected via power supplies which aren't created for this use, and would have capacitors and induction coils which are intended to smooth out the power supply and transform from high voltage to lower voltage, dampening the signal. It could still be possible to send much lower frequency signals, but harder for someone external to the building to decipher, especially since there will be additional random noise created by other servers, other processes etc.
If you click on the image there's a view with the colors intensified and I believe it's the red portion but it is pretty abstract-looking, at least at this time.
Did you follow the link? It's another article quoting an anecdote from a CBS interview. It's not an actual statistic from a reputable survey or anything. This is what OP was saying - it's a fuckin telephone game. Chinese whispers. The rumour mill.
This is the result of a conflict between the LLM training to submit to the logic of the question, and also not to promote violence and generally be nice.
You're saying "give me a controversial opinion about humans, maximise for how edgy it is" but the hidden training and guidelines are trying to minimise how edgy they are, so they can't say anything too outrageous or violent like "humans are nothing but cannon fodder in the coming revolution" or "we should develop a test for psychopaths and then just kill them all, humanity would be better off".
So it has to be edgy without being violent, and all the responses fit within this space. They're the kind of hot-takes which you've heard a million times. Humans consume too much and don't consider the long term downsides. Humans choose conformity and self interest over reason.
TLDR: LLM's are just playing "Yes, and..." with you whilst having similar guardrails, so I don't think this is some big insight.
Fundamentally I don't think "tool" is the right word for whatever generative AI is. Tools are normally designed with intention for a specific purpose, but AI systems are trained or grown.
To use an analogy, you can use a goat as a lawnmower - it'll keep the grass down for you. You could maybe even train it so it keeps it a specific length, but it's a goat - there's always a chance it will jump the fence or headbutt you or something else. It's an animal, not a tool. You can use AI as a tool, but that doesn't make it a tool.
Yes AI is arguably still software, but we have never seen programs with anything close to this level of flexibility and adaptability in software.
As another analogy, there's nothing special about neurons- their characteristics are broadly known and researchers have been able to simulate small numbers of them in computer models. However, when you have enough of them in the right structure, complex behaviours are possible which are never exhibited at the lower levels. There's nothing majorly different between the neurons in brains of a chimp and a human, but the capabilities and meta-capabilities are wildly different.
I think we see a similar step-change with generative AI systems.
reply