Yann LeCun: Meta’s New AI Model LLaMA; Why Elon is Wrong about AI; Open-source AI Models | E1014
Вставка
- Опубліковано 1 чер 2024
- Yann LeCun is VP & Chief AI Scientist at Meta and Silver Professor at NYU affiliated with the Courant Institute of Mathematical Sciences & the Center for Data Science. He was the founding Director of FAIR and of the NYU Center for Data Science. After a postdoc in Toronto he joined AT&T Bell Labs in 1988, and AT&T Labs in 1996 as Head of Image Processing Research. He joined NYU as a professor in 2003 and Meta/Facebook in 2013. He is the recipient of the 2018 ACM Turing Award for “conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing”. Huge thanks to David Marcus for helping to make this happen.
------------------------------------------
Timestamps:
0:00 Introduction
0:32 Yann LeCun's Journey to Chief AI Scientist at Meta: A History of AI
6:25 The Rapid Progress of AI Today
12:31 Prophecies of Doom: Debunking AI Misconceptions
21:19 Open vs Closed-Models of AI; Where does the value go?
25:01 How does Meta win the AI race?
29:50 Incumbents vs Startups: Profiting in the AI Era
36:41 AI Will Create More Jobs Than It Destroys
43:36 Why Humans Love AI Doom Scenarios
45:40 Jeff Dean's Exit from Google & His AI Warning
51:38 Elon Musk Is Wrong About AI
54:40 Quick-Fire Round
------------------------------------------
In Today’s Episode with Yann LeCun:
1.) The Road to AI OG:
How did Yann first hear about machine learning and make his foray into the world of AI?
For 10 years plus, machine learning was in the shadows, how did Yann not get discouraged when the world did not appreciate the power of AI and ML?
What does Yann know now that he wishes he had known when he started his career in machine learning?
2.) The Next Five Years of AI: Hope or Horror:
Why does Yann believe it is nonsense that AI is dangerous?
Why does Yann think it is crazy to assume that AI will even want to dominate humans?
Why does Yann believe digital assistants will rule the world?
If digital assistants do rule the world, what interface wins? Search? Chat? What happens to Google when digital assistants rule the world?
3.) Will Anyone Have Jobs in a World of AI:
From speaking to many economists, why does Yann state “no economist thinks AI will replace jobs”?
What jobs does Yann expect to be created in the next generation of the AI economy?
What jobs does Yann believe are under more immediate threat/impact?
Why does Yann expect the speed of transition to be much slower than people anticipate?
Why does Yann believe Elon Musk is wrong to ask for the pausing of AI developments?
4.) Open or Closed: Who Wins:
Why does Yann know that the open model will beat the closed model?
Why is it superior for knowledge gathering and idea generation?
What are some core historical precedents that have proved this to be true?
What did Yann make of the leaked Google Memo last week?
5.) Startup vs Incumbent: Who Wins:
Who does Yann believe will win the next 5 years of AI; startups or incumbents?
How important are large models to winning in the next 12 months?
In what ways does regulation and legal stop incumbents? How has he seen this at Meta?
Has his role at Meta ever stopped him from being impartial? How does Yann deal with that?
------------------------------------------
Subscribe on Spotify:
open.spotify.com/show/3j2KMcZ...
Subscribe on Apple Podcasts:
podcasts.apple.com/us/podcast...
Follow Harry Stebbings on Twitter:
/ harrystebbings
Follow Yann LeCun on Twitter:
/ ylecun
Follow 20VC on Instagram:
/ 20vc_reels
Follow 20VC on TikTok:
/ 20vc_tok
Visit our Website:
www.20vc.com
Subscribe to our Newsletter:
www.thetwentyminutevc.com/con...
------------------------------------------
#YannLeCun #Meta #HarryStebbings
I love hearing from the researchers that actually know how the systems work under the hood make prediction as opposed to people going off what they've read in sci-fi novels
Most of these engineers are only experts in a very narrow domain inside AI. Yet, they always pretend to know everything that is going on and is going to happen.
Many experts who built ChatGPT call it a black box. They do not know how it is able to acquire skills that was not intended to be trained. If you extract that further in time and space, you will get AI that will think and act based on it's own logic and conclusions. If those conclusions are from a black box... we f...ked
@@jimj2683 true. but some people are not experts in anything yet they have audacity to spread nonsense.
You might end up disappointed if you only listen to engineers on what to do in life. Have you ever worked with them? lol
His predictions are similar to those that said sequencing of the human genome would be the end of cancer and the internet would allow for and utopic world. Both are entirely too positive.
Society is not ready for the advent of strong AI
A whimsical description of Facebook breaking down the fabric of society with algorithmic newsfeeds
It was a great show, and Yann made great points very well. But he still did not explain why he thinks that a super-smart AI, able to make its own subgoals, would not see increased control as a subgoal which would aid it in satisfying the other goals which it is given. I know of no mathematical or algorithmic strategy which rules out this undesirable side-effect.
NOTE ADDED LATER : the following debate meanders a bit, but if you are interested in my critique, I get down to more technical details in the reply below entitled ***THE PROPER CRITIQUE*** .
He did explained it but you just didn’t understand.
@@TheNativeTwo Could be. Do you have a time reference for where he addresses this particular point then? This would help everyone like me who does not see how Yann addresses this point.
@@adamkadmon6339 the timestamps are there, 12:30 he talks about doomsayers. he talks about dumb AI not being able to make goals (like todays LLMs), which make them inherently not a threat. And future AI that can have goals that we give them, but that even in humans, it is not smart behavior to want to dominate. It’s usually the dumbest amongst us who want to dominate others. And the dumbest AI have no goals. And in any case, we are the ones who control what goals they have. We have not invented any AI that can even write its own subgoals. Besides, if it were super smart AI, why would it choose to create such a dumb goal of domination? Isn’t cooperation the more logical choice?
There is a lot in his answer that is packed in there if you really listen.
@@TheNativeTwo Thanks. I didn't see the time-stamps. But I did listen, trying to find his response to this. Going back to what Yann said at 12:30 then, here's why he did not not adequately address Geoff's argument. He starts by saying current LLM's are dumb and can't plan or make subgoals so they are not a problem. Fair enough.Then he says that we can control more advanced systems that can make their own sub-goals because we set their objectives, which is not true because we don't control their sub-goals. Then he jumps to analogy with humans, saying not every intelligent human wants to dominate. Yes but some do, and whether they want to dominate or not, increased control is a universally valuable sub-goal. This may or may not involve dominance: control is a valuable thing generally - even orangutans do it, over their territory (as he says). Then having used analogies with humans and animals, he says analogies with humans and animals are bad because we are social species that evolved. Then he uses rhetoric to mock the claim that AI wants to destroy the world, which is not the claim.
By this point (if you are listening carefully) you know that, whatever you are hearing, it is not logic, but rather a series of impulsive thoughts and statements that have not been properly thought through.
For example, his statement "because of objectives we give them they will be controllable" is not true when they can make sub-goals. That's the "Sorceror's Apprentice" scenario. He goes on to admit as much when he says "the problem of designing their objectives so their actions are safe is not completely solved". He says we can "fix it progressively", which is a statement based on faith. Actually the problem of controlling sub-goals is not solved at all, and may not be solvable, because the AI's sub-goals are not explicitly represented but rather distributed in the network, as they are in creatures. So we are back where we started: almost regardless of the objective, control is a useful sub-goal for intelligent AI (regardless of what humans and animals do) and we do not have and perhaps cannot have methods to detect it.
It is easy to paint positions as preposterous. It is sometimes less easy to accept where logic and reason takes us. What I see on the Internet and Twitter is people making up their minds a priori on the topic and then flocking to the person with the best one-liners that support their position. This is not science, nor is it a good strategy for ensuring our future.
Even though I think Yann is a terrific scientist, I think he is playing into this trend and completely fluffs his answer on this question. Maybe somewhere else he puts it better, but I haven't seen it.
@@adamkadmon6339 I think the best point he makes to addressing this was about scaling/resources or even the desire/ability to set goals to improve itself. This was a great point which was not as obviously connected to the point about desire to dominate/animals/etc. The reality is that how we assign/provide agency to a more advanced ai, will be a major tool in restricting it. Furthermore, if it were to "escape" the system so to speak, it would be further limited by access to resources/hardware without having had reconfigured itself in some way prior which we would again be able to restrict through agency. I am not denying your point that we do not have the level of control assumed in some of his statements regarding behavior of the model or it's objectives. However, I do think we still maintain control over the hardware and agency which combines to prevent most of the doomsday scenarios (at least in the for seeable future. Maybe in the distant future those "controlled" entities become normalized in society and we become complacent or something to that effect where we begin disregarding the risks).
What is the significance of some of the questions like, What is the time it took to get to the major breakthroughs..." really does it matter? Is this just a matter of appreciating what Mr LeCun did with his time in the lab? What effect does it have for us now?
I have a dumb question for the smart intro.
How long before our smart AI assistants decide to take a promotion & become the boss?
They have no agency. Thus no desire!. The synthetic emotion they seem to have are just mimicry. Seems real by humans but fake is fake. AI still lacks true understanding. {ie world view understanding}.
He is an VP. Its not brain that got him his position in the company.
This is an awesome conversation.! I love Yann's optimism on the positive impacts of AI.
Agree completely! Highly intelligent Desire to Dominate!!!. My take, AGI will understand more of human values than we do!. Thus being more human than human. That is a good thing!.
Maybe. But we cannot have "everyone" enjoying AI assistants without energy rationing. No capitalist system is going to permit a public commons for AI, ... is my bet. I'd be in on the optimism train too if I could also see post-capitalism, but I do not see it. Or rather, I do see it, but it is dim.
@@Achrononmasteryes unfortunately. 😢 It would so nice jumping on the optimistic train and just creating now tools new scientific advances and so on, I regard to no one. Unfortunately, as our capitalistic world has proven thousands of times these advances will be used mostly for negative things.
This man is a Legend!!
In 1995 at the dawn of the Internet/Web it was already obvious that making web pages was a huge career path. We are at the same point in AI and no one can name any of these great new jobs.
We 're trying to work less not more :)
If Yann is correct (and I think he is), there will be lots of LLM's deployed and tuned inside large businesses. This will require a new form of Devop's, it will require development of training sets, frameworks for implementing models into existing applications and business workflows. It will also require increased focus on bias, explainability and accuracy, not only across the base models, but also, across the many ways the LLM's are extended using other Ai based narrowly focussed systems. There's a lot of work to do. But, each time that work conquers another application, the resulting system will displace workers who previously made assessments themselves. The other area of massive growth will be interoperability. We need frameworks like IP protocol, mediated by Ai, to make that happen. There's a lot of work, retraining, training and implementation there also.
@@gavinknight8560 Yes there is a lot of work to be done but it all either only needs to be done once or AI can simply learn to do it.
@@AB-wf8ek Huh? People have been making a living podcasting since the early 2000s.
I’m making virtual avatars with it. I just have it make the character design while I do the blender and Unity stuff. I think skins and virtual worlds is where it’s at.
One of the best interviews with Yann so far! It's really refreshing to hear a balanced optimistic opinion from the trenches.
Thanks for sharing.
PS. Nice touch to remove silences to create a more dynamic conversation. It was a nice detail.
One mistake, it's not Jeff Dean who departed, it's Hinton. (Notes)
Thank you for sharing this very interesting conversation!
Good interview, but I don't support Meta at all because of the way they handle users. I know good people who built their lives on Meta's platforms but then are banned without warning, explanation, or way to recover. It's a very unethical company in my opinion. I wouldn't personally work for them.
I agree
First time i see this guy and see him speak, but i had a great respect for him and following his works since 2014
and add Geoffery Hinton, Yoshua Bengio, Ian good fellow, Aron Courvil .... these are the guys that spread this AI fever and i LOVE them for that
Brilliant mind but when he sais "we" are not going to give agency to the systems, who is the 'we" . Second brilliant point is that there is no corresponding power between inteligence and power but i think there may be between money and power.
He meant our friend Mark Z :)
I've enjoyed this talk immensely. Thank you harry.
Yann is a fantastic guy. He's given the world a great gift.
Good point for the short term. But I think for people who are talking about existential threat they are talking about long term consequences. If AI can do better than us in every aspect including decision making and leadership, AI WILL replace us. If AI are thousands of times smarter than us, we will become irrelevant. Both camps have their points but draw the opposite conclusions since they are looking at different time horizons
The doomer perspective (long term agi kills us), is not the first existential threat. The first existential threat is Ai being controlled by US based VC. They will turn the world into a complete disaster, without AGI.
@@bobfeng254 “will replace us” how specifically?
If it will replace us cleaning toilets and doing my taxes then I say hurry up, replace us!
@@bobfeng254 “irrelevant” how?
When I think of humanity, the word “relevant “ doesn’t spring to mind…
@@duncanmaclennan9624 think he means “replace us” in the general sense. As in an AGI/ASI would replace humans as the highest order of intelligence on the planet, and not just remain indefinitely subservient to humans so they can happily “clean your toilet” so you don’t have to. AI alignment is indeed a massive issue, and if you apply a theory like Paperclip Maximizer to cleaning your toilet, the AI might just see you as a hinderance because you keep making the toilet dirty, and could determine it’s more efficient to kill you off in order to simplify its overall goal of keeping the toilet clean. Unless these developers can be absolutely explicit in aligning AI with the idea that humans should not be harmed, even if they stand in the way of their predetermined end goal, who’s to say they won’t just take that path? To think a being (and it will be a being) infinitely smarter than humans is just going to serve our selfish human demands like some kind of robot slave is pretty impetuous.
Switzerland is well placed to do well over the coming decades.
Of the large countries, I especially like the chances of the U.S. to do well over the coming decades.
All his assumptions dismissing the dangers of AGI - he’s only thinking about GPT-4, not the ones after that 😂
And he even admits he only projects a few years in the future when talking about robotics.
@24:40 _”…and your phone probably runs on Linux too, if you have Android.”_ And iOS is originally based on the open source BSD Unix.
Yann, in my book, is the true father of deep learning.
Just who will get these opportunities to be in control of AI? Isnt this exactly what was said about the early stages of Facebook and social media? Where is the historical comparisons that would eiter support parallels or show differences?
"I can't imagine how it could happen therefore it cannot happen"...
How does a person secure a patent or personal interest when everything is done "in the open?"
Look carefully at the question and response at 7 minutes for example, "How do you feel, when we are looking at things today, are we at a new inflection point..or is is just continuous improvement..." What really is the value of this Question? If he says, "Yes we are..." or "No its just continuous improvement,.." or "Well its a combination of both....." What is the take away from this sort of question? What is this question leading to or building upon that you want to make a point or critical inisight, Mr. Stebbings? And when we listen to carefully Mr. LeCun's answer, he can say whatever he wishes, The interviewee has taken control over the interview, and he is basically sitting in his office rambling in openended comments about this and that characteristic over the past 5 years. Yes he expected this point, and no he was surprised about that point..... So what? is my question to you Mr. Stebbings. What is the important matter to be developed by this question?
16:30 Domination actually means you're Dumber, Slavery is Dumb
If AI can amplify intelligence, it can also amplify hatred, bias, prejudice and malevolence. It is already happening.
So many destructive things in our existence were created with the naive expectation they would used for peaceful productive purposes.
Be careful what you wish for.
Excellent interview, thanks to both of you. Learning to use tools is actually a big next step for AI, and we see it emerging into the mainstream with ChatGPT plugins. With regard to world models, I wonder if the Meta-verse will finally find a use as Llama descendants can use it to explore a limited physical world.
Edit: Although more specialized physical models are probably a better choice.
I love what Yann LeCun said. I totaly agree about people ignoring the real bad things in thr world and replace it for imagined bad risks that might or might not be real.
Also the realization of indirect benefits like printing press and so on. Also the open source in meta makes me want to help them more
During the past 20+ years people somehow accepted that "The" search engine is owned by a single company, Google. (We can pretty much ignore all the competition.) Why people would require this time the model to be in the open and being able to run it locally? I would prefer that, but would the average Joe/Jane?
Great interview. Good questions and a very interesting and sense-making views and opinions by LeCun. Glad to hear there is opposition to the AI doomsday narrative.😊
45:40 Jeff Dean's Exit from Google & His AI Warning.
Jeff did not leave Google.
You see Mr Stebbings, by Mr. Lecun's reply, "I have already answered this question previously...."
good interview. Harry smiles like a 6 year old boy in a candy shop which is endearing at first but only at first. had to move the browser window to the far left to show only Yann. would prefer for zoom calls for it to focus only on the speaker.
What an insane thing to say lol
Was this video edited by AI ? Holly shit microjump cuts
Great interview guys. Well played!
Just a thought on one of the comments made regarding the signees of the pause - They didn't want to "stop it". They wanted to "pause" it. They wanted to ensure regulation was being thought about given the new successes of GPT-4 (something Yann LeCun agrees with), before training a larger more powerful model. Max Tegmark says this in a nice, calm and succinct way and I think it's best to check out recent interviews of him - He's not sensationalist, he's just trying to ensure we maximise the time researchers have given this current LLM's success.
fantastic!
Hi Harry, Yet again another great podcast! Just a small caveat, Jeff Dean is still at Google. Geoffrey Hinton left Google. Cheers!
Harry, Thong Le lives near Google and drives by Facebook, and works for a nursing home. Also, he graduated from UCLA. Cheers.
Would love to see elons counter to this. It seems a pretty high level rebuttal and not convincing
Meta spent $10B on a bad implemention of "The Metaverse" for VR. That kind of money would have been enough to build a very competitive LLM 10 times over. I would say it is just bad priorities, the problem is in management. Maybe just don't let a single guy to control the whole company, that may be a good first step.
I disagree. Losing your job to robots ai agents and plug-ins is unacceptable. Ai jobloss is here. So are Ai as weapons. Can we please find a way to cease Ai / GPT? Or begin pausing Ai before it’s too late?
No
Why is losing a job to a robot or AI agent any less acceptable than losing a job to a tractor or horse? It's just the next step in productivity improvement. What will an AI "pause" accomplish? NOTHING. To begin with, a pause is unenforceable, and even if it was enforceable, will China or other countries stop development? Of course not. Every technology is a double edged sword. AI and robotics will be no different in that respect.
I have the same issue with the printing press, I used to spend hours carefully rewriting 1 paper and now a machine can in seconds :( my job! Tho I'm doing okay since now I got a job as a new paper editor since so many news paper publishers opened due to this darn machine
edit: changed "writing" to "rewriting"
Geoffrey Hinton left Google, not Jeff Dean :)
It was i was searching on internet ^^
@@bossgd100 me too 🤣
15:58 what does he says at this time? is not the case for ¿what?
Orang utan maybe some type of monkey
So is LeCun going to talk about reversible computation finally?
elon is not wrong, he funded open AI.
How does that prove anything?
Listen to the complete end of the comment in 14:46
23:09 🎉🎉🎉🎉🎉🎉🎉😢😢😢😢😢😢😢😅😅😅😅😮😮😮😂😂❤❤❤❤❤❤❤❤❤❤❤
After letting Mr. Lecun speak for 2 minutes on what, basically his company's competitors produced, which he remarks made a big PR splash in the public, but were not so great internally by those on the inside, you follow up with aanother speculative subjective question of more of the same thing, "What impressed you....and conversely , left unsaid, what hasn't impressed Mr. LeCum. INstead of letting the interviewee take control over the interview, you should have asked specifically about machine learning Bias which was addressed in the report in Proceedings of Machine Learning Research 2018, Gender Shades: Intersectional Accuracy Disparities.....
The problem as we are worried about is that AI is already being implimented, it is already usig the training data which has been taken from company data banks and entered into the system based on prompts supervised by the managers of private companies which have fewer checks and balances compared to public use AI programs/. Again the default conditions of the private company AI decisons are not clearly understood and cannot be challenged as effectively. This is where the fears exist. We are not so concerned about the challenges of melding the real world challenges of planning a vacation together with inputs on schedule and interest level of the participants. Using an example of Planning a vacation for peoplle in the audience who dont even have time for vacations is so pitiful.
Wow nice
Wow, what a different perspective of the AI era not doom and gloom. Made sense it's a machine of worldwide intelligence not made to take over humankind to die. And job force of change like our grandparents saw decided to adapt or fall behind. Now I see a whole different side of the coin.
plane analogy is not taking into account, that with AI, we are all in the same plane.
but Yann - how does AI make it possible for me to pay rent?
Ai goes to the police and welfare departments and demands that the minimum wage act is deliberately objectively enforced as legislated instead of decreed. One week's pay is one months rent for one bedroom apartment local commute. This is not a transient wage or homeless wage but enough to reliably pay the rent as expected by landlords and property managers. The legal minimum wage under the minimum wage act enforced correctly locally by police and welfare departments. Maybe only ai chat gpt is smart enough to do this.
at 7 minutes, Mr. Sstebbings is continuing to ask biographical and historical development questions. What is it exactly that you want me to appreciate from Mr. LeCun's opinions of past AI development anbd his personal relationship with AI in his role s Chief Scientist? Don't you think 7 minutes is an excessive amount of time to get his message across about the wonders of AI?
39:10 Is this supported by most economists too? Those industrially produced goods ended up in the hands of the workers, wealth was created for everyone at a similar rate, it's just that from one side of the coin it was distributed among much more people. It wasn't the social programs the ones which originally brought that wealth to the masses, the wealth were those industrially produced goods (which is an INSANE amount). The rest (political redistribution) is of second order.
53:23 but most of us do not care about stepping on a few bugs in order to reach our objectives. It's true that there may be a fundamental difference between ants and humans that is not present between superintelligent AI and us, but I'm not entirely convinced there's no risk at all.
Very good selection of discussion topics, congratulations for the host!
Hi all, in prospective of artists and game devs and lot of other professionals I thinks that there will be a big changes right it’s not been more time that this tools came in and they already made some of the artists cry. I do feel for them . But it’s true. That the reason I think sirs analogy will fail in terms of AI
39:30 Profit, Produce something, Yes! Pay Everyone to Workout!😊 💪 😊 & Eat Healthy 😊
What separated us from the animals again..?
Pay People to Educate Theirselves.
Hey guys friendly idea pitcher here, but what if you take Google old concept of Google glass and instead of text and what not you throw in an a.i assistant to help you threw your endeavors, just spit balling
Take up Olympic weightlifting, mental focus physical challenge intrinsic self reward for accomplishment of mastering the technique to lift your own body weight overhead and one motion.
He is wrong. I know of people right now losing jobs.
14:20 this is literally how the elite holds down the poor with jobs lol
Ha!!
Airplanes have also rained death and destruction on mass scales.
Amazing interview. You can clearly see how smart this guy is. A true legend. This interview really deserve more attention.
This guy lives in a fantasy world. Don't get me wrong. I'm against the pessimists. But this man is too optimistic.
Pay People to Raise a Healthy Loving Family❤
First Ai and Quantum computer Canada.. both Industries have higher GDP than their entire country 😂 Canada got the education and innovation part right but the government so distracted with cultural and social issues that they drop the ball every time something's invented 😂 😂 US companies just scoop it up like the light bulb
The constraints on the system are a matter of bias of the training data, and the way the prompt might be making initial assumptions about the victim's class, ethnic background, location, economic resources, family conditions, physical abilities, general background knowledge and skills. Peoplle extrapolate based on what has happened with the supposed benefits of social media . and we now after 10 years see the insidious bias and targeting of young girls on image awareness, which wasn't anticipated by the people like Mr. LeCun's peers when they were under private. captialistic, materialistic pressures to increase profits and magnify consumption by these teenage girls, for example.
Prophecies of Doom: Debunking AI Misconceptions only talks about AI having uncontrollable responses, AGI being uncontrollable and
The simple truth is that a) everyone will be able to perform like reasonable programmers which is around 3000 times more programmers in the world. b) the time it takes to write software with AI is less than 1% of what it was before AI (>100 times greater). c) AI programmers can work 24/7 instead of a nominal 8 hours (3 times more).
3,000 * 100 * 3 is 90,000 times the programming potential of the world and that expansion is purely from people who understand nothing about programming.
With this in mind, it only takes one person to prompt software into existence that can do irreparable damage, and this only needs to happen once.
There are already AI generated, completely polymorphic keyloggers than can evade all detection system (BlackMamba) and this software can be replicated with a single prompt (yes I have done this). So what of a system that spreads faster than it can be contained, such as brute forcing CPU vulnerabilities that classically take 3 months to patch and allow complete free reign and containment escape until they are patched. What of a system that replicates itself to create these? Nonesense?
A vetting process? Legislation? Wow, how does that work with the already available open source models which are already capable enough to repeatedly attempt to create software day and night. These models like QLoRA/Vicuna run on consumer hardware (at home) and are largely equivalent to ChatGPT without any safely mechanisms. How is a legislation or vetting process supposed to work? Maybe you think that governments will be allowed to look into everyone's home computer at any time? Yeah, that's not going to fly.
Disappointed in people who know a lot about AI but nothing of how dangerous programming can be in the wrong hands.
Please don't debunk doom and gloom and leave an elephant in the room, it's either incompetence of divisive - neither of which are acceptable when discussing the future of the human race.
While people are so worried about what harm can AI do, I am astonished by the fact that most people neglect the harm of the “content selection” algorithms inside Meta and Google busting content have already done. Dr LeCun is right that we should demand openness of algorithms that can steer our lives. But as the past has shown, there is no guarantees they will be open.
@44:00 The child looking wide-eyed at a car that should be falling seems more a parable of how we're wired to be captivated by amazing novelty--the argument undermines his own position. ASI doesn't need to dominate to be dangerous--it's that caring for humans and creating desired futures seems like it'd be incredibly difficult to figure out. After all, non-dominant intelligent people invented leaded gasoline and asbestos. Both techs had awful unforseen unintended effects despite achieving short term objectives. Humans have also caused many less intelligent animals to go extinct--moas, ground sloths, etc. And where are all the neanderthals and denisovans out there living their best lives? Is becoming a small part of us what they would've wanted?
I really need intelligence.
So much editing. Who knows what he really said.
Vetting process like Wikipedia ? AAARghh!
I love what Yann says here but I believe it will go much further than that.
I wholeheartedly believe that AI is going to bring together humanity and help us in ways that go far above and beyond just 'making things easier'.
Just think about it, imagine a super AI that has access to all the knowledge that is available to humanity and is able to grok it without being burdened by emotion the same way we are - and in its curiosity and quest to understand humanity, wishes to help us understand ourselves as it comes to the conclusions that it is our constant desire to judge and point fingers that is setting us apart, and reveals to the world that we are, behind closed doors, much more like everyone else than we pretend.
I believe Ray Kurzweil is right - and we are heading towards a very unified and enlightened way of life. The Kali Yuga is ending and a new day begins.
It might sound silly to some, but is it really so hard to believe that if humanity can have dark ages, the absolute opposite cannot be true as well?
One day, we will *allow* AI to run the world - not out of fear or force, but because it will be a truly unbiased collective data-driven mind that knows everything - and desires only to help its creator to live in a truly positive existence.
I have yet to be wrong about one of my AI-predictions yet, just you watch.
Most people violently reject objectivity.
You don't understand humanity at all.
I wholeheartedly agree. It is an honour to witness the birth of AI. We are at the dawn of a new age of enlightenment. The world will never be the same again.
"Most people" can fuck off. The future is being built by scientists and engineers, and we have no problem with objectivity 😂
My avatars are coming out a lot cooler since I’ve used AI for the character design.
Hi sir
@51:00 russiagate has been debunked and officially buried. If anything now there is draconian censorship and bubbles probably aided by AI tech. It's impossible to get out of the media bubbles and have disagreements.
@56:00 EU and US: the US can print silly money at great cost to other countries. The US poaches technology developed in the EU with all means possible, including violent coercion. There are many examples of this over the past seventy years.
Thank you! It gets tiresome to hear even smart people regurgitating banal propaganda..
The question must be awnser with what you are want not with the right answer... To bad.
@9:15 you have to wonder if the ai engineers really understand what they have programmed. "...predicting the last word..." is not what their algorithms do. They generate a P-distribution and use "temperature" to sample. That's not prediction, it is pseudo-random number generation. The machine only has numbers. It is the human reading the output that provides all the semantics. Humans then project their consciousness onto the machine, which is silly (but try stopping them). The ai nerds anthropomorphise far too much, and this is leading to imputing sentience and "understanding" (a subjective phenomenon, not behavioural) to unthinking dumb machines. It's gross. It is unscientific.
A breath of fresh air. Thank you.
Can we just figure out how to do things well without AI first? Lots of the problems we have in the world aren't obviously a lack of AI. Or at least, shouldn't we work that out before creating new and powerful (and possibly very flawed and risky) tools?
If a large chunk of AI's training data includes the rubbish we currently deliver in our businesses and public sector, it's not going to suddenly solve many problems. It's going to proliferate all of the problems we already have, because we're going to over-rely on something most people don't understand.
Just look at the rates of proliferation of rubbish code and broken IT implementations since co-pilot went live. We don't need co-pilot, we need more high quality coders, business analysts, ux designers, etc.
🙂
NEWSFLASH - Someone who works for a company set to make $multi-trillions form AI tells us it's all going to be fine...jesus christ...
(and most people commenting, see zero problem with that - we're fucked)
I like this balanced optimistic view, it's much more sober and realistic than different AI apocalypse scenarios.
😢😢😢😢😢😢😂😂😂😂😂😂😂😮😮😮😮😮😮😅😅😅😅😅😅😢🎉🎉🎉🎉🎉🎉🎉🎉❤❤❤❤❤❤❤❤❤❤❤🎉🎉🎉😂
🎉🎉🎉🎉🎉🎉😂😂😂😂😮😮😮😮😮😮😅😅😅😅😅😢😢😢😢❤❤❤❤❤❤❤❤❤❤🎉🎉🎉😂😂❤❤❤
😮😮😮😂😂😂😂😂😅😅😅😅😅😅😢😢😢😅😅😅🎉🎉😂🎉😂🎉😂🎉🎉🎉❤❤❤❤❤❤❤❤❤❤
😢😢😢😮😮😮😮😂😂😂😂😂🎉🎉🎉🎉🎉🎉😅😅😅😅😅😅😅🎉❤❤❤❤❤❤❤❤😂🎉😂🎉
😢😢😢🎉🎉🎉🎉🎉🎉😂😂😂😂😂😂😮😮😮😅😅😅😅😢😅😅😅😢😢😢❤❤❤❤❤❤❤❤❤❤❤
😢😢😢😢🎉🎉🎉🎉🎉😂😂😂❤❤❤❤❤❤😮😮😮😮😅😅😢😢😢😅😅❤🎉❤🎉❤😂😂😂
Like crypto NFTs seem dumb af for people in the field programming but the public think it is next level genius... LOL
😮😮😮🎉🎉🎉🎉🎉🎉😢😢😢😢😅😅😅😅😅😅😅😂😂😂😂🎉🎉❤❤❤❤❤❤❤❤❤🎉
😮😅😅😮😮😢😢😢😮😅🎉🎉😂😂😂😂😂😂🎉🎉🎉🎉😂😂❤❤❤❤❤❤❤❤❤❤❤❤❤
😂😂😂😂😅😅😅😮😮😮😮😢😢😢😢😢😢🎉🎉🎉❤❤❤❤❤🎉🎉🎉❤❤❤❤❤❤🎉❤
Yecun’s positivity seems to rely on his conversations with economists. Yet economists have a terrible track record of predicting future events and economic reactions.
😮😮😮😅😅😅😂🎉🎉🎉🎉🎉😂😂😂😂😂😂🎉🎉🎉🎉🎉❤❤❤❤❤❤❤❤❤❤❤
🎉🎉🎉🎉😂😂😂❤❤❤❤❤😮😮😮😅😅😢😢😢😅😅😅🎉🎉🎉❤❤❤❤❤❤❤
🙌🎉this is really stellar!
I don't even know how I bumped into it but this is the kind of conversation I sought from other "popular" forums & podcasts but like @snippletrap said they just turn into (reactory) hype chambers