I'm confused. Lately people here have been telling me Google had become unusable and they nowadays use ChatGPT to get their information...
Now Google brings AI to search and that's wrong, too?
rufus
joined 2 years ago
https://en.wikipedia.org/wiki/Scientific_method
No. Science isn't done by a vote of majority. It's the objective facts that matter. And you don't pick experts or perspectives, that's not scientific. It's about objective truth. And a method to find that.
We're now confusing science and futurology.
And I think scientists use the term "predict" and not "forecast". There is a profound difference between a futorologist forecasting the future, and science developing a model and then extrapolating. The Scientific American article The Truth about Scientific Models you linked sums it up pretty well: "They don’t necessarily try to predict what will happen—but they can help us understand possible futures". And: "What went wrong? Predictions are the wrong argument."
And I'd like to point out that article is written by one of my favorite scientists and science communicators, Sabine Hossenfelder. She also has a very good YouTube channel.
So yes, what about DNA, quantum brains, Moore's law, ... what about other people claiming something. That all doesn't change any facts.
Thanks. Yeah I know what the ICD Codes are used for and what's out of scope.
Thanks for clarifying which platforms and models you're working with. I'll look up the two I'm not familiar with.
Nice talking to you. Yeah I've met her. I also need to tell her that I appreciated the conversation. Maybe I've come off a bit argumentative. I'm going to do that tomorrow, it's getting late here.
I've tinkered around a bit with LangChain, vector databases and stuff myself. I started last year. But I'm more one of the "open-source" and Linux types of guys... So I don't really like proprietary services like ChatGPT. I've used the so called "open" models that I can run on my own hardware. Like Meta's Llama and others. But I lack a proper graphics card to run that at home. And last year when I started, everything was happening quite fast. (It still does.) And I just couldn't keep up. So I scrapped the project. But I'd like to start fresh and make another attempt at creating something.
With the worth, that's an interesting way to look at it.
I don't think you grasped how exponential growth works. And the opposite: logarithmic growth. It means at first it grows fast. And then slower and slower. If it's logarithmic, it means at first you double the computing power and you get a big return... Quadruple the performance or even more... But it'll get less quickly. At some point you're like in your example, connecting 4 really big supercomputers, and you just get a measly 1% performance gain over one supercomputer. And then you have to invest trillions of dollars for the next 0.5%. That'd be logarithmic growth. We're not sure where on the curve we currently are. We've sure seen the fast growth in the last months.
And scientists don't really do forecasts. They make hypotheses and then they test them. And they experimentally justify it. So no, it's not the future being guessed at. They used a clever method to measure the performance of a technological system. And we can see those real-world measurements in their paper. Why do you say the top researchers in the world aren't "well-enough informed" individuals?
Fair enough. To make a friendly gesture from my side... Would you be willing to label yourself if I were to agree to label myself as human? I think that'd be a fair deal.
-- rufus (human)
(Edit: I agree with the open source bit. I don't know if I've linked the Eric Hartford article in this discussion... I think it sums up the underlying issue very well: https://erichartford.com/uncensored-models )
Sure. Multimodality is impressive. And there is quite some potential there. I'm sure robots / androids are also going to happen and all of this has a profound impact. Maybe they'll someday get affordable to the average Joe and I can have a robot do the chores for me.
But we're not talking about the same thing. The video I linked suggests that performance might peak and plateau. That means it could be very well the case that we can't make them substancially more intelligent than say ChatGPT 4. Of course we can fit AI into new things, innovate and there is quite some potential. It's just about performance/intelligence. It's explained well in the video. (And it's just one paper and the already existing approaches to AI. It doesn't rule out science finding a way to overcome that. But as of now we don't have any idea how to do that, instead of pumping millions and millions of dollars into training to achieve a smaller and smaller return in increased performance.)
Hmmh. I'm a bit split on bio implants. Currently that's hyped by Elon Musk. But that field of neuroscience has been around for some while. They're making steady (yet small) progress. Elon Musk didn't contribute anything fundamentally new. And I myself think there is a limit. I mean you can't stick a million needles into a human brain everywhere from the surface to deep down, to hook into all brain regions. I think it's mostly concerned with what's accessible from the surface. And that'd be a fundamental limitation. So I doubt we're going to see crazy things like in the sci-fi movies like The Matrix or Ready Player One. But I'm not an expert on that.
With that said, I share your excitement for what's about to come. I'm sure there is lots of potential in AI and we're going to see crazy things happen. I'm a bit wary if the consequences like spam and misinformation flooding the internet and society, but that's already inevitable. My biggest wish is science finding a way to teach LLMs when to make up things and when to stick to the truth... What people call "hallucinations". I think it'd be the next biggest achievement if we had more control about that. Because as of now the AIs make up lots of facts that are just wrong. At least that's happening to me all the time. And they also do it when doing tasks like summarization. And that makes them less useful for my every-day tasks.
Highly doubt that.
- I don't think they're gonna allow ERP and that's a big market (erotic role play). The other companies are going to stay. Mind what happened with Replika AI.
- Other companies also constantly innovate and invest. Regularly outperforming OpenAI and also offering a better value for money ratio. Take like Llama 3 or MistralAI or Google. OpenAI just has lots of money and researchers. But they're not leading by much concerning their competitors.
- AI might not grow exponentially
- I'm fine keeping up to date manually. It's a bit of work but has the added benefit that I'm not fed that much false information. I rather believe journalists and scientists. At least until AI gets more intelligent.
- Yeah, I still worry about your ethics. But I dropped the topic since you were not listening. Seems you're already set on your position and not able to accept other positions.
- Hmmh. I also worry about the ethics of the big tech companies. It's not a good thing they get to choose what's right and what is wrong instead of we, the people.
Hmmh. We've seen all kinds of claims and hype regarding AI. I'd like to see and judge for myself. Guess I'll have to wait a few days.
Edit 2024-05-18: And yesterday it showed up in the webinterface. How do I get the talking and the emotions? Is that not available yet? Or do I need a phone app for that?
What I the to be alive.
It's "What a time to be alive..." And you need to link the Two Minute Papers video when saying that 😆
Ah, I missed that. So if it's on camera, I hope they go to jail for that.
I mean there is evil and there are criminals. Always have been. That's how the world is. We need to lock them away or give them therapy. And we need a good justice system and police to handle that. And we need a healthy society to pvent that from happening in the first place.
Your revenge fantasy isn't helpful. That's how it's been before civilization and a just justice system. There have been vendettas, lots of people being killed. Sometimes the correct ones, lots of innocent people have been killed, too. And it leads to more and more violence and crime. We know that from experience. And how does it even help the victims? I suppose not every victim is sadistic and wants to see their violator suffer physically. I myself think locking them away is alright. I'm not getting off to violence. But the punishment needs to fit the crime. They need to get locked away for a good amount of time.
And if we allowed vigilantism and mob justice... Who do you think would suffer the most? Probably black people and minorities, because they get wrongfully accused all the time. And women, because men seem to be slightly more aggressive in general. And it'd end up being a disservice to women. My point is: We need justice. Not yet more violence.
"Just a few days ago a middle aged women was walking [...]"
Wtf? Why didn't you do something? You should have stepped in or called police!
"protected"
What do you mean by protected? They're locked away?! What would be your solution? Vigilantism and vendettas?
4B [...] correct me if I’m wrong [...] murdered [...]
That's not the correct picture of what's happening in South Korea. It's mainly about patriarchy, young men getting ultra-conservative, childish and pricks to women. Murder and violence, too. But it's not like that (murdering spouses/women) happens way more often than for example in America or Germany.
I completely agree. I'm not sure where you live, but also live in a country with a lack of doctors. It's a shame we have to wait months and months for therapy. And that's not healthy, neither for the people in need, nor for society in general. And mental health is such a valuable thing.
I think you're right with your assessment towards "open" models. I think that viewpoint has been popularized by the leaked google memo We Have No Moat, And Neither Does OpenAI last year. And unless we want the path forward to be laid out for us by companies like Google and OpenAI, there's no way around independent and "open" models.
And I'd like those AIs to become more accessible, run on (affordable) consumer hardware. Maybe on my phone. Things like real-time translation would be great and I could have a simultaneous interpreter in my pocket. Or an assistant that can organize my files and keep track of everything. I think there is quite some potential and all of that is doable in the near future. And it'd seriously boost my own abilities if I were for example enabled to read Japanese text or Spanish news.
Concerning the AI at home: I use koboldcpp as of now. That enables me to run smaller models on my old computer. I use that for role play, companionship etc. I'm also having difficulties finding the time to build software around that. At some point I'd like to have a virtual companion that's open-source, keeps me company and can do useful things. I would estimate I'd need like 4 weeks of full-time work to build something like that. But I'm already able to program in Python and have a broad understanding of the required components. Maybe one day I'll find some time, but there's so much other stuff that needs attention.