AI Threats 'Complete BS' Says Meta Senior Research, Who Thinks AI is Dumber Than a Cat (msn.com)
- Reference: 0175245977
- News link: https://tech.slashdot.org/story/24/10/13/2220258/ai-threats-complete-bs-says-meta-senior-research-who-thinks-ai-is-dumber-than-a-cat
- Source link: https://www.msn.com/en-us/money/other/this-ai-pioneer-thinks-ai-is-dumber-than-a-cat/ar-AA1s7O1F
> When a departing OpenAI researcher in May talked up the need to learn how to control ultra-intelligent AI, LeCun pounced. "It seems to me that before 'urgently figuring out how to control AI systems much smarter than us' we need to have the beginning of a hint of a design for a system smarter than a house cat," he [2]replied on X . He likes the cat metaphor. Felines, after all, have a mental model of the physical world, persistent memory, some reasoning ability and a capacity for planning, he says. None of these qualities are present in today's "frontier" AIs, including those made by Meta itself.
LeCun shared a Turing Award with [3]Geoffrey Hinton and Hoshua Bengio (who hopes LeCun is right, but adds "I don't think we should leave it to the competition between companies and the profit motive alone to protect the public and democracy. That is why I think we need governments involved.")
But LeCun still believes AI is a very powerful tool — even as Meta joins the quest for artificial general intelligence:
> Throughout our interview, he cites many examples of how AI has become enormously important at Meta, and has driven its scale and revenue to the point that it's now valued at around $1.5 trillion. AI is integral to everything from real-time translation to content moderation at Meta, which in addition to its Fundamental AI Research team, known as FAIR, has a product-focused AI group called GenAI that is pursuing ever-better versions of its large language models. "The impact on Meta has been really enormous," he says.
>
> At the same time, he is convinced that today's AIs aren't, in any meaningful sense, intelligent — and that many others in the field, especially at AI startups, are ready to extrapolate its recent development in ways that he finds ridiculous... OpenAI's Sam Altman last month said we could have Artificial General Intelligence within "a few thousand days...." But creating an AI this capable could easily take decades, [LeCun] says — and today's dominant approach won't get us there.... His bet is that research on AIs that work in a fundamentally different way will set us on a path to human-level intelligence. These hypothetical future AIs could take many forms, but work being done at FAIR to digest video from the real world is among the projects that currently excite LeCun. The idea is to create models that learn in a way that's analogous to how a baby animal does, by building a world model from the visual information it takes in.
In contrast, today's AI models "are really just predicting the next word in a text, he says... And because of their enormous memory capacity, they can seem to be reasoning, when in fact they're merely regurgitating information they've already been trained on."
[1] https://www.msn.com/en-us/money/other/this-ai-pioneer-thinks-ai-is-dumber-than-a-cat/ar-AA1s7O1F
[2] https://x.com/ylecun/status/1791890883425570823
[3] https://slashdot.org/story/24/10/09/1949256/after-winning-nobel-prize-geoffrey-hinton-said-he-was-proud-ilya-sutskever-fired-sam-altman
power (Score:4, Interesting)
A for loop is an extremely powerful computational tool, it has made $ trillions $ for humanity.
Re: (Score:2)
Better yet, the for loop with unsigned loop variable has made $infinity$ for humanity
Re: (Score:3)
Found the programmer easily replaced by AI.
The Superconducting Super Collider problem (Score:4, Interesting)
The late Freeman Dyson remarked at the diminishing returns from throwing resources at scaling up a known type of accelerator and hoping meaningful scientific discovery comes out of it. Maybe he was expressing a contrarian opinion about devoting a big chunk of the NSF budget to the eventually cancelled Superconducting Super Collider?
AI has always been about solving problems with exponential complexity with hardware that only grows at a polynomial rate when you build a bigger machine at a current level of technology. Yes, Moore's Law and all that about how hardware has grown exponentially in capability, and this is behind the current AI renaissance, the current hardware is on the cusp of finally doing something useful with AI. But this idea of restarting Three Mile Island and dedicating its electric output to a server farm, can we think a little bit more critically about this?
Think of Eric Schmidt's "We need to destroy the Earth's climate in order to save it" about going Hell-for-leather consuming hydrocarbon fuel to meet an exponential growth curve in computing power consumption so the AI will come up with the solution for Climate Change.
For the faction of Slashdotters who regard Climate Change as overhyped as AI, current levels of CO2 emissions may not be the problem many say they are, but certainly we don't want to greatly increase the rate of CO2 emissions and do we want to write a blank check to build out AI? Even the CEO of TSMC was rolling his eyes at the AI people wanting to spend trillions with a capital T on more fabs to keep up with projected AI demand.
Re: (Score:2)
I think you are on the right line of thought; the AI bubble can only go so far before investors demand some tangible results from the billions/trillions they have been throwing at it. I’m really hoping the AI craze does not tank the world economy when it pops like the .com bubble.
Re: The Superconducting Super Collider problem (Score:5, Interesting)
I think the investments are not large enough to massively damage world economy, but since none of the big investors will recover their investments, some of them will suffer and some will die. Wouldn't it be funny if AI is what finally kills Microsoft?
Re: (Score:2)
"...certainly we don't want to greatly increase the rate of CO2 emissions and do we want to write a blank check to build out AI? "
We don't want to do either of these things. Frankly, LLMs are a proof of concept, I don't want them "developed" further at the current time. Efforts need to shift to simulating intelligence, not increasing the amount of knowledge even larger LLMs can integrate. But Altman and Musk are more interested in money.
"...AI people wanting to spend trillions with a capital T on more fa
Re: The Superconducting Super Collider problem (Score:2)
Plants are still CO2 restricted, but I have a few ferns in my house, so I am ready to make new coal once it ticked up high enough
AI, again (Score:1)
Is there any way to filter out those annoying AI news? We all know it's the hype until that bubble finally bursts, but it's getting very old very quickly.
Re: (Score:2)
> Is there any way to filter out those annoying AI news?
Simply scroll past AI stories.
> We all know it's the hype until that bubble finally bursts, but it's getting very old very quickly.
It's not all hype. LLMs are extremely useful for a lot of use cases.
Since when are cats.. (Score:1)
..able to impersonate a fellow human being of average intelligence? He/she might be a professor, but logically this reasoning does niet seem to make much sense.
Re: (Score:2)
ELIZA was said to impersonate humans too.
Re: (Score:1)
[1]This piece of cardboard can impersonate a human being [bbc.com].
It's also kind of strange that you refer to cats and mention "fellow human being."
Are you a mouse? You do a good job impersonating a human, nice job typing.
[1] https://www.bbc.com/news/blogs-trending-44367371
One threat is those that use it (Score:2)
I think a significant threat is going to be people that use LLMs to generate code without understanding it well enough to check for errors. We have enough trouble with people that don't understand memory management, proper input validation, error handling, etc. I'm fairly certain that AI generated code will lead to a whole new wave of insecure code.
Re: One threat is those that use it (Score:2)
I completely agree.
Re: (Score:2)
> I think a significant threat is going to be people that use LLMs to generate code without understanding it well enough to check for errors. We have enough trouble with people that don't understand memory management, proper input validation, error handling, etc. I'm fairly certain that AI generated code will lead to a whole new wave of insecure code.
This here is the real threat, yes. We desperately need to stop with the idea that "everyone should learn to code" or sending people to "code camps" and related nonsense. Software engineering and programming are very hard disciplines. Any idiot can string together enough scripting language nubbins to be dangerous, but writing reliable, deterministic software is difficult and complex. Software already faces problems in that it is not as serious as a discipline as say, electrical engineering or mechanical engi
The singularity (Score:1)
"One day, machines will exceed human intelligence." - Ray Kurzweil
"Only if we meet them half-way." - Dave Snowden
Sounds like Meta, et al., are desperate to become the dumbest guys in the room.
He is not wrong (Score:2)
Even for a really dumb cat.
The problem is people thinking it's intelligent (Score:3)
So they start using it to make reports and reply emails. Then they make it participate en business strategy and take risky financial decision. And that's the real threat.
Re: (Score:3)
They think that because it's obviously more intelligent than them.
They just don't have or can't keep up with a good role model.
Current threat from AI (Score:2)
I think AI is already a tool that is being used by threat actors
Imagine a phone call from your child. Phone number is spoofed and gets right through because it's the correct number, the voice sounds enough like your child that the distress in the voice bypasses all reasoning in your mind. Actual photos of your child are altered and sent showing them in a distressing situation. I think this type of threat would fool a lot of people if it was timed right.
Cats arn't that dumb (Score:2)
And with the intelligence they've got in the right enviroment they can be lethal. If you don't believe feel free to wander the africa savanna unarmed.
LeCun has been saying this for a while. (Score:3)
LeCun has been saying things like this for a while, and it is striking that he has not grappled much with the people who disagree with him, and has mostly responded with a mix of condescension and incivility, even when they are often other highly accomplished people. Meanwhile, Hinton who shared the prize with LeCun has become deeply concerned about AI risk. See [1]https://www.cbsnews.com/news/geoffrey-hinton-ai-dangers-60-minutes-transcript/ [cbsnews.com] and [2]https://mitsloan.mit.edu/ideas-made-to-matter/why-neural-net-pioneer-geoffrey-hinton-sounding-alarm-ai [mit.edu]. And it is worth noting that Hinton became convinced of these concerns, by actually going and looking at the arguments people have made. At this point, I'm not sure what would possibly convince LeCun that AI is a threat that wasn't immediately fatal to most of humanity.
[1] https://www.cbsnews.com/news/geoffrey-hinton-ai-dangers-60-minutes-transcript/
[2] https://mitsloan.mit.edu/ideas-made-to-matter/why-neural-net-pioneer-geoffrey-hinton-sounding-alarm-ai
Re: (Score:3)
It is common courtesy to politly ignore old professors when they go senile. Too bad the news aren't that polite.
Theoretical AI is deadly ChatGPT Is autocomplete (Score:2)
LeCun is on the money in regards to today's AI. Generative AI is fancy autocomplete. It's not the Matrix or Terminator. There are dangers with Generative AI, but not in the ability to outsmart us, but the ability to scam us. It helps scammers write better spam....it's too inaccurate to replace real workers, but if it gets a little better it can generate content "close enough" to reality to fool your grandma into investing into fake companies, for example. It will enable a new era of grift. It will gen
Cats are evil! (Score:2)
If AI ever gets as smart as a cat we are all doomed. They've been trying to kill us for centuries!
AI as a human-replacement intelligence (Score:2)
AI, currently, is dumb as a box of hammers. It is amazing in its ability to mimic some of the output of a talented human being while completely lacking any kind of intelligence.
However, human brains aren't made of magic, and there's no reason to believe that our intelligence is anything other than the results of an incredibly complex web of patterns resulting from fairly basic stimulus and response chains.
Eventually, we're going to make a true AGI. I suspect there will be some hardware development require
The true danger are evil people with AI... (Score:2)
Someone faking your election...
Someone using it to scam you...
Someone putting it on flying bombs to bomb cities and hospitals...
AI isn't smart but people are stupid (Score:1)
The problem at the moment isn't so much that AI is "smart."
The problem is all the people who think it is. It's being moved from a decision-support role into into a decision-making one. Whoops.
AI is bad for Palestinians (Score:2)
[1]‘Lavender’: The AI machine directing Israel’s bombing spree in Gaza [972mag.com]
[1] https://www.972mag.com/lavender-ai-israeli-army-gaza/
This person has never watched a cat video (Score:2)
It doesn't take intelligence to be destructive.
And given that AI is taught on media full of cats and people being destructive for "Lulz", its almost cetain LLMs will do horrible things.
AI is not the problem (Score:2)
The problem is people who use AI
Today's AI is mostly harmless and useless. Tomorrow's AI will be a powerful weapon in the hands of bad people
We need effective defenses
Rogue AI isn't the issue (Score:2)
it's AI that works. Specifically AI that works and suddenely replaces 20-30% of all workers.
Folks are focused on chatbots and call centers, maybe programmers, but we're seeing lots and lots of other applications, like advanced manufacturing robots, self driving cars, etc.
Part of that is we got so excited with LLMs that we forgot about general purpose ML.
We're nowhere's near ready for what's coming. It's another Industrial Revolution.
Everyone remembers the luddites as just loons that impeded
So what he's saying (Score:2)
Is a cat is smarter than something created by humans. Good to know.
I wholly accept our fuzzy feline overlords.
And yet its destructiveness is already on display. (Score:5, Insightful)
The algorithms that now dictate how most people get information already behave like Orwell's Ministry of Truth, conjuring pure fiction out of the ether while banning and erasing history (even recent history) as taboo subjects. This kind of reality-neutral and consequence-neutral "amplification" product is exactly what AI systems seek to perfect, and it's basically just a form of bomb.
Re:And yet its destructiveness is already on displ (Score:5, Insightful)
Isn't that basically what search engines have been doing for the past 15 or more years?
Re: And yet its destructiveness is already on disp (Score:4, Insightful)
Yes.
And they have already given us an accelerated blue/red division of politics and gigantic echo chambers for antivax, flatearth, and other nonsense.
Re: And yet its destructiveness is already on disp (Score:5, Interesting)
Honestly, I think the (major) search engines and algorithms had tried* to quash antivax and flat earth quite purposefully but can't override the power of social media to spread brain rot. They even managed to amplify the nutters' persecution complexes along the way.
In pre-communication-tech history, people were very used to just adapting to whatever dumb ideas dominated their local community, in early comm-tech(radio, television, newspapers) we had a brief "golden age" where for good and evil, there were gatekeepers who decided what to spread, and then social media allowed people to create communities that spread every dumb idea to every person susceptible to it.
*Admittedly not very hard
Re: And yet its destructiveness is already on disp (Score:5, Insightful)
Also as i heard it crudely described to me is if you were say into, having sex with toasters, that would be somethign you kept to yourself lest if you let the people you know in your community know they would likely say "hey man, that's not cool, you shouldn't do that" but today while those people would stay the same you can just hop online and find a forum of people discussing Emerson vs Sunbeam and updates which one they decided to copulate with this morning.
The internet normalizes behavior that would in the past be discouraged. Like all things internet this is good (folks who need connection that were not able to find it before can) but can also be quite bad (reinfornces behavior and opinions that should not have been).
Re: And yet its destructiveness is already on disp (Score:5, Funny)
What can I say? Toasters are hot.
Re: (Score:2)
The internet enables behavior that would in the past be discouraged
FTFY. The internet, without the people using it, would be nothing but wires. The perverted behavior and normalizing is just people . We do this; just faster, more widely, and easier than it would've been, previously.
Re: (Score:1)
> blue/red division of politics and gigantic echo chambers for antivax ... and other nonsense
Unfortunately, lab leak theory and a number of other reasonable topics or questions got lumped with "antivax".
Re: (Score:2)
Not unless there was a search engine that dynamically generated counterfeit content in top rankings to deceive people, the way news feeds have become. The single overwhelming application of AI has been, and will be, sabotage of communication. And since communication isn't optional, AI is certainly a threat to humanity. The only bullshit part is that it's something new: It's just a scaling of the same garden variety con games used in business since forever.
Re:And yet its destructiveness is already on displ (Score:5, Interesting)
"The single overwhelming application of AI has been, and will be, sabotage of communication. And since communication isn't optional, AI is certainly a threat to humanity. "
No, the "single overwhelming application" is a threat to humanity, not AI.
"The only bullshit part is that it's something new: It's just a scaling of the same garden variety con games used in business since forever."
Right, because the "single overwhelming application" is not new, all that is new is its use of AI.
Modern AI is merely a new, easy, efficient way to access massive amounts of information that, until recently, required skill and access to massive databases. It is not intelligent. It provides access to unlimited trivia easily, it is a revolutionary enabler of the worst types of humans in society.
Re: (Score:2)
No, search engines are entirely driven by user queries, they are passive. "Algorithms", in this context, as active query generators that push narratives.
Re: (Score:2)
And what titans of industry do we have to thank for these life-changing, world-bettering, productivity-amplifying innovations? Nothing will ever change until there is some mechanism for holding shareholders accountable for the societal harms caused by decisions of companies they invest in.
Re:And yet its destructiveness is already on displ (Score:5, Insightful)
We arrive at the very simple solution to pretty much all the great harms society endures today: Ending the corporation as a legal fiction (or, really, fictional law) and operating strictly by personal accountability. No more shell games, dodging fines and lawsuit damages by shuffling paper. No more double rights for the corporate owner class.
Re: (Score:2)
Problem is, most humans are dumber than a cat. And 100% of those, and 99.9% of those that are smarter than a cat take their brain off the hook and are determined never to use them again after high school. (Assuming they even used the grey matter then).
Server them up results that say "Yes, but" or "No, but" and they never get past the first word.
So, AI or no AI, "whale oil beef hooked".
Re: (Score:2)
It's not that people are dumber. Many are just suckered by lies and don't ask questions. But many are also scared of the truth because they cower from the bullies of the community. They choose instilled "beliefs" over reason even while knowing its bullshit.
Re: (Score:2)
Public stupidity is exploited, but it isn't The Problem. When I block false and malicious content from feeds, the algorithm just "retaliates" by amplifying it even further: Like it was designed to actively punish critical awareness, not merely reward stupidity.
That's where it crosses the line into dangerous. These tools are no longer just being misused by fools, they're being weaponized by malevolent minds against everyone who can remember more than 5 seconds into the past.
Please dont burst the bubble (Score:3)
I have not sold my AI and Nvidia stock yet