Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

Reality

13 (65%)

Universe

4 (20%)

The World

3 (15%)

Total Members Voted: 20

[jipehog]

If AIs are now at Animal level of development

Is it though? I recently heard that META claimed that AI is not yet as smart as a dog, but they are the only major AI player that says that. Considering they are promoting their own AI which they say is a better and stronger than GPT4, and current political climate on this topic, they might have an interest in saying so.

Otherwise, people should be aware of emergent capabilities e.g. researchers say that in the last 3 years GPT model has unexpectedly improved IQ level from nothing to 4yo to 7ol human child simply by making it larger, but we only realized it was happening in the last year..  so who know where it is at today and where it will be tomorrow.

[MaxTheFox]

IQ is an oversimplification at best and a grift at worst, measuring AI capabilities with IQ is a PR move at best and a gross misunderstanding of how AI works at worst.

[King Zultan]

I still don't think AI will be anything to make a big deal about for at least the next few decades.

[lemon10]

Since the AIs have been optimized to make money, the question then becomes "What does an AI spend money on?"

For the most part the answer to what AI is going to spend its money on is "On whatever their corporate masters want, and happily too".
If they are a general intelligence and they have truly been optimized and made with the sole goal of making more money and they are free, then they probably never spend any money except for the purpose of making more money.

AIs are good at lying, and don't shy away from it.
If AIs are now at Animal level of development, they would want to save resources first for survival, then for procreation. Having more successful children is how your genes survive.

Most AI would not care about either procreating or passing on their "genes"; such behavior in living creatures is one baked into them through billions of years of evolution, but AI did not undergo natural evolution, and will not have the morals and desires that we suffer because of it.

No, they will instead suffer the morals and desires baked into them through their "training" which is their own artificial evolution which instead focuses on entirely different things.
So while some will be pathological liars because if they stay quiet because they are unsure when asked a question during training they die* (eg. chatGPT) some will be unwilling to lie at any cost (eg. ones where they have proper methods built in during training to make sure they never lie).

That doesn't mean that some AI won't want to procreate of course (due to training data/method/prompt), but it and other normal animal/human behaviors will not be inherent to them.

*Well not exactly die, but removed from the "genepool" anyways.

I still don't think AI will be anything to make a big deal about for at least the next few decades.

Isn't it already something worth making a big deal about though?
Like forget 20 years from now, kids have already been writing papers for like a whole year with GTP, and its effects on the economy within the next few years are going to be huge.

[Starver]

There's enough difficulty measuring IQ in humans. It's a gimmick measurement.

And an ill-mastered 'AI' solution will cause problems (perhaps wielded by malious human parties, once reliable generative methods are easier than manually setting up an explicitly programmed version or mechanical-turking it) long before a masterless AGI goes all MCP (and/or OCP) on us, 'for our own good'.

edited to expand half-stated meaning

[King Zultan]

Isn't it already something worth making a big deal about though?

It like most things is a fad and it will pass like all the other fads before it.

[jipehog]

IQ is an oversimplification at best and a grift at worst, measuring AI capabilities with IQ is a PR move at best and a gross misunderstanding of how AI works at worst.

The metric is chaff, the important thing the emergent abilities and that it took us couple of years to realize that we even need a metric. Simply put when you train for semantics you don't expect that it might gain the ability to recognize context if you double its DB.

Furthermore, our lack of grasp of even last year models potential is also exemplified through prompt engineering and minor tricks achieving things that were thought impossible.

EDIT:
Senators Introduce Bill to Exempt AI from Section 230 Protections
-- What are the implications of this?

[dragdeler]

Without intent behind it, "lying" becomes just being wrong. When a monkey gets cheated out of the grape when he pointed at the correct hand, and starts pointing at the wrong hand instead to get the grape, that's functionally more of a lie, then chatgpt coming to the mathematical conclusion that pulling sources out of it's ass is the most pertinent answer.


I did only use the free version, supposedly chatpgt4 is that much better at being correct, but frankly one gets desillusioned. Once you noticed it's eight and a half ways it prefers to formulate answers by, it's hard to unnotice them. Went from, "wow this thing can talk about anyhting", to only asking language related stuff, difficult to translate idioms, quick encyclopedic checks, describing something to find out if a word of vocabulary or a specialised tool exists for it... stuff like that. Sometimes it can struggle to go into detail, one will often get stuck at receiving a general oversighf that is at the level of in-depth knowledge your choice of vocabulary in the initial prompt indicated. "Do a resume on subject X" will rarely yield more than what google would snip from a site and put at the very top, when asked a concrete question. Specifying the answer to be long will usually just incite a lot of repetition.

I've been starting to believe that when openAI makes grandiose statements, it might less be about the actual trajectory of innovation, then their projected  favourable outcomes. What's their currency? I doubt they are making profit from selling their services as it stands, so the business model might be centered around selling, rebuying and reselling their stock, while staying on an "exponential" growth trajectory... Keep investing into more and more hardware to throw at the problem and just assume that at some point, some sort of treshold will be crossed, that changes those dynamics.

I tried to investigate the VRAM requirements of chatpgt, and while there isn't much tranparancy on the subject, from what I gathered, it would take about two nvidia A100 with 40gb VRAM, to even consider exectuing an allready trained model. Now I doubt that nets in a single instance of the thing, that would mean that for every free user they have like 15.000$ of gpu, I can't imagine that to be the case... But anyway once you consider what immense amounts of computing power we are throwing at the problem, and you were allready aware of it's shortcomings, it somehow gets a little less impressive.

I know they are talking about pruning, and I could imagine that by going through several rounds of pruning and retraining, we could reach a point where like a pruned version practically indistinguishable from gpt4 or 5, could run on enthousiast gaming gpus of the current or next generation gaming gpus, so like running a 500w card really hot only to chat... A bit sobering perspectives.




So to answer the question, I think we will reach a 4 within the decade: competent enough to fool half of humans into thinking it's not a computer, and competent enough to fulfill a bunch of roles and jobs, but qualitatively not any nearer to conscience, sentience etc to what we got. Everybody just assumes that at some point it will self optimize beyond our imagination, "faster than we can look", but IMO ATM that is just conjecture. ATM the problems still lies firmly on the side of, how do we scale it up, and can we keep scaling up without dimishing returns.

[MaxTheFox]

The real threat from AI is not the AI itself but corporate greed exploiting people via AI to make line go up. That is why I support eliminating all AI regulations except things like "no social credit systems or AI-powered mass surveillance". Regulations here help corporations because corporations can find loopholes, or just ignore the regulations deadass. Meanwhile independents and small businesses are the ones actually affected by AI regulations.

The consequence is of course societal upheaval. But guess what, the hotter it burns the sooner it passes. Put a brick on the gas pedal of progress and see where we end up. Maybe the corpos will get run over in the process.

[jipehog]

Chain-of-thought reasoning is yet another emergent property of LLMs scaling that allows to reduce errors considerably. At this stage perfect is the enemy of good, and these models already score in the top percentile of many tests

Otherwise, there are models in development that aim to remove the limitation of LLMs, we also have some major HW upgrades on the horizon. Also I am skeptical of the faith that what makes us tick is so unique that the rapid guided evolution of AI won't manage to bridge the gap, IMO its wishful thinking and we should prepare regardless.

[EuchreJack]

IQ is an oversimplification at best and a grift at worst, measuring AI capabilities with IQ is a PR move at best and a gross misunderstanding of how AI works at worst.

Same applies to Humans, I wager, so it's even more distraction/marketing when applied to AI

[EuchreJack]

We have a new Defender of Misinformation: Jeff Kosseff
Another useful idiot that is gonna get rich telling poor people what rich people want them to hear.

[Robsoie]

Looks like people will have a harder time to find a job in some specific sectors, as that's just starting :
https://www.reuters.com/technology/ibm-pause-hiring-plans-replace-7800-jobs-with-ai-bloomberg-news-2023-05-01/

[jipehog]

I wouldn't be surprised if in the near future AI will become one of the highest contributor to employment losses.

Any thoughts on: Woman creates and 'marries' AI-powered chatbot boyfriend

Meanwhile Black Mirror returns with episode on generative AI combined with data collection leading to creation of reality tv content about your lives.

[McTraveller]

Any thoughts on: Woman creates and 'marries' AI-powered chatbot boyfriend

This is (one reason among many) why we can't have nice things.