Demis Hassabis on Chatbots to AGI | EP 71
Вставка
- Опубліковано 4 тра 2024
- This week’s episode is a conversation with Demis Hassabis, the head of Google’s artificial intelligence division. We talk about Google’s latest A.I. models, Gemini and Gemma; the existential risks of artificial intelligence; his timelines for artificial general intelligence; and what he thinks the world will look like post-A.G.I.
This interview was recorded on Wednesday. Since then, Google has temporarily suspended Gemini’s ability to generate images of humans, following criticism of images the chatbot generated of people of color in Nazi-era uniforms.
Chapters
00:00:00 - Demis Highlights
00:00:32 - This week on Hard Fork
00:00:46 - Who is Demis Hassabis?
00:05:40 - Demis Interview
00:05:43 - Gemini/Gemma
00:13:37 - Open sourced VS closed Source
00:16:21 - How much personality should Chatbots have?
00:17:51 - Gemini hits pause on people pictures
00:19:26 - Benefits of being at Google
00:23:57 - Public’s negative opinion of A.I.
00:26:11 - How useful as a science tool?
00:28:29 - Curing diseases
00:29:24 - A.G.I.
00:34:04 - how should we prepare for AGI?
00:38:31 - whats your p(doom) ?
00:43:03 - will this just concentrate power more?
00:50:26 - most A.I.-proof job?
00:53:12 - Your personal post-A.G.I. plan?
00:54:55 - BONUS clip: chatGPT went KrAzy!
Hard Fork is a weekly look into the future that's already here. Hosts Kevin Roose and Casey Newton explore stories from the bleeding edge of tech.
Casey’s publication, Platformer: www.platformer.news/
FOLLOW US ON SOCIAL
TikTok:www.tiktok.com/@hardfork?is_f...
Insta:
Kevin Roose- kevinroose...
Casey Newton - crumbler?h...
Twitter (X):
Kevin - x.com/kevinroose?s=20
Casey - x.com/CaseyNewton?s=20
Threads:
Kevin - www.threads.net/@kevinroose
Casey - www.threads.net/@crumbler
Subscribe to the audio only podcast:
Apple - podcasts.apple.com/us/podcast...
Spotify - open.spotify.com/show/44fllCS...
Amazon - music.amazon.com/podcasts/7c7...
Google - podcasts.google.com/feed/aHR0...
The New York Times - www.nytimes.com/column/hard-fork - Наука та технологія
This guy is great. So glad people like him are at the helm.
XD be careful what you wish for
Fully concur, Demis is one of a kind.
47:11 Demis's face when the host says "OpenAI has set itself as a nonprofit"
Demis is a true intellectual, one of the smartest humans on the planet. Alpha Fold, which solved the protein folding problem, was no small feat. Winning the 2023 Lasker award for Basic Medical Research was a true honor!
Solving the protein folding problem (Alpha Fold) is the equalivant of solving the Riemann hypothesis problem in physics, which remains unsolved. Some of the comments suggest your viewers don't have a grasp of the problems we are trying to solve in AI.
But it didn’t solve protein folding.
@@mynameisjeff9124 It did in practical terms. While its true humans still do not understand precisely how the amino acid sequence dictates the 3D structure of a protein, it is also true that we now have a commercially available tool that does understand this and is able to predict with near 100% accuracy the coordinates of all heavy atoms in the protein's 3D structure.
I am not a scientist or an expert of any kind. But I am very optimistic about what Demis talked about, from personal experience. I have been living with Multiple Myeloma since 2017. When first diagnosed, the median survival time was two years. But the research and the application of ideas and science and medicine in the intervening years has been amazing. I am now 77 (almost 78) and the median survival time for newly diagnosed younger people (in their 30s) is now two to five years depending on the type of myeloma. I have benefitted from all the progress made in 8 years, such that I will possibly not die of multiple myeloma. In 2017, the word was that multiple myeloma was treatable but not curable. It’s a strange thought that death is the cure - at my expected end of life at my expected age and it wont be the horrible death that it was in 2017. I am in my second remission and I no longer see the protein spike in my bloodwork. I am still working as a writer. So thank all three of you and your colleagues, from me and all 150 of my fellow members of the North Texas Myeloma Support Group.
Current median survival time is now ten years or more, not 2 to 5 years. I was listening to the podcast and got a little distracted. Sorry about that.
Being a cautious optimist sounds like the right state of mind. Demis depicts a future of incredible possible improvements for life on our planet. Keep up your fantastic work.
Oh my gosh! This is the BEST! Demis Hasabis is the most brilliant, fine individual who is at the top of my trust list of developers of A I. Thank you for this terrific interview. I just discovered your podcast.
A lot of people are going to find it hard to grasp that what Demis is talking about might just be around the corner for all we know
Some predict this year, most experts predict in less than a decade. We are in for perhaps the most monumental upheaval in our lives, and I am all for it.
you know how some naturalists long for the hunter gatherer life pre-agriculture
this is how i feel now...about this era 😅
life will be so drastically different i will feel lost, I'm kind of scared, nostalgic and excited all at once.
Yeah, chaos and poverty for everyone! There's one thing you can always be sure of: governments will help corporations to make even more money while the lower and middle classes become poorer and poorer.
@@aceyage get involved in local politics. Fight the power. Tax the rich. Then watch the sociopathic leaders of Google and Facebook deploy their AI armies to to marginalize any attempts to restrict them and make them share the wealth.
@@aceyage roll over and succumb to your doomer fate then
Been a massive fan of Demis since his work at Lionhead studios in 2001, 23 years ago, this man was clearly on the right path to developing world changing AI. He's always been a bit odd, but damn if he isn't gobsmackingly brilliant.
Wow, you knew him that long. I know he was a chess prodigy.
What's odd about him? He doesn't strike me as odd.
Remember his 'r' back then? Obviously not relevant to the topic, but apart from everything else, he has also improved his speech clarity.
Always love to hear Demis' thoughts! Thank you.
55:20 "GPT had a nervous breakdown and broke into Spanish". I noticed that too. In a reply I got there was a sentence written in Chinese. I cut and paste it into translator and it was actually on-topic. Crazy.
I like this guy this is the first time seeing this this channel but he makes sense I like that he’s very upfront. Thank you.
man am I glad someone asks the right questions, good job to you guys. we need to know the technicals, how they achieved it etc.
Ten years can go by fast. AGI is coming quick.
What a great show! Such a smart person! It's cool to see him again after first seeing the Alpha Go Documentary so many years ago. "The concept of money will change" we need to talk about this more!
The guy at the left is a serious bobblehead
He’s feeling the agi
🤣🤣🤣
😀, he is in his zone
I love that guy!
I literally went to comment on this exact thing, and here you are ❤
Great interview fellas
I appreciate your podcast so much. Thank you for another great episode!
Huge fan of Demis Hassabis. Looking forward to what DeepMind comes up with and how things will change in the future.
great interview
“I dream of a room temperature superconductor” “same!” These are fabulous dreams that the likes of me would have no idea were even a thing! 😂 that’s what I love about listening to more technical podcasts that are way above my competence but stretch and entertain me. 👍🏻
I love that you all post your podcast video to UA-cam Music and allow for comments per episode, I seriously wish all the podcast shows I subscribe to would follow suit.
He's come a long way from developing Theme Park together with Peter Molyneux ;-) -- I have total confidence in Demis' goals and principles to lead this effort!
❤ Demis is an amazing human being
Great Interview. I like the View of Demis, he is a cool guy and perhaps the Einstein who is alive today :)
Good interview
Braviii!!!🎉👏👏👏
Interview begins at 5:40
I always love hearing Demis talking about AI I remember reading his articles about the games he was working on in EDGE magazine back in the 90's(?) or was it early 2000'a?
I appreciate his scientific enthusiasm with the seemingly medical slant. Also please don't skimp on your sleep Demis!
I wonder what type of gamer he is? What does he play? Can I see him jn Helldivers? 😂 I’m super curious on his thoughts an AI on gaming entertainment.
Looking at the games he's made is quite a good indication of what he likes to play. Syndicate (1993), Theme Park (1994), Republic: The Revolution (2003), Evil Genius (2004) etc I'm pretty sure he also likes Chess, Go & Poker etc
Demis is just great, and I just enjoy his talk.
Kudos to Demis for mentioning The Culture novels.
I like his approach to AI existential safety
Demis is the greatest scientist of our time, perhaps in human history. Grasp that.
Not even close to greatest
He has a great point about how utterly unscientific it is to put a 0% probability on something. 300 years ago the concepts of engines, bacteria, or wage labor didn't exist, so when someone says that something is sci-fi and it _is_ within the currently-known laws of physics, you know they're talking out their ass.
Also like the shout-out to the _Culture_ series by Iian M Banks. He's more then a decade passed, but his series still stands as one of the most important (& well-written) depictions of a positive future.
That it is so should tell you something about the amount of quality sci-fi that isn't doom & gloom (either because it's easier to write dystopia or because artists are depressive people). Writing utopian sci-fi is like writing Superman comics; beginners think it's impossible because they don't understand the most exciting conflicts are about right & wrong or philosophy vs philosophy instead of life & death.
Demi's needs to be much more involved in the public sphere as a counterpoint to Sam Altman and others.
Altman in technical terms seems overrated ; Demi is the real deal.
@@os2171 OpenAI has Greg Brockman and Ilya Sutskever, no shortage of top talent. sama clearly inspires great loyalty from staff.
Great interview!
dude on left like hell yeah, hell yeah, hell yea haha
This is huge to get him right after all of the new google stuff!! nice work!!!! so excited to listen
you team who has .of hard fork podcasts video youtube chanell.totally disagree with your views of opinion hated saying that ago by 2030..ad .some experts other Dennis hasibis .2030 agi.know of david shapiro youtuber podcaster thinks agi will be within no more then. 3 .- to-2.years .maximum.
@@davidchavez3937maybe have chatgpt or Gemini clean that up a bit before hitting send next time?
@@davidchavez3937are you okay
@@davidchavez3937Your comment was very incoherent. Were you having a seizure? 😢
Great radio voice
another video to question my existence let's go!!!!!!!!!!!
“the Borg” of Star Trek - illustrates the absorptive potential of Wild , un restrained, A I
Nice to hear him mention Iain M Banks' novels.
The Culture is the utopia we hope for. It could become reality in our lifetime.
@@minimal3734yes! Exactly this. Elon Musk, with all his faults, is a fan and proponent of The Culture. The Neuralink “Wizard Hat” (thanks Tim Urban) seems very much an early strategic attempt in that direction.
Discussions on the definition of money and the societal construct of a company should certainly be revamped in preparation for upcoming changes. These are basic but important topics.
However, the bigger question to me relates to levels of competition vs. cooperation. Is the ultimate goal to use AGI as a tool to continue to increase the rate global economic activity or slow it down? This question is being decided RIGHT NOW, and corporations should not have the last say.
Because we could, if the democratic majority so chooses, simply switch gears and focus on social endeavors that help improve human quality of life, dignity, etc. while AGI does the hard work ~ without accelerating the rate of economic growth even further.
Perhaps the most diplomatic of us should be granted access to AGI with plenty of leeway before the corporations get it. Like a 30-year leeway to be safe..
P(doom) depends on how seriously researchers take that scenario. If they take it seriously, then we'll probably be fine, if not, then we might be screwed.
Kinda like climate change, we took it seriously, and now as a result, renewables have skyrocketed.
Hopefully the analogy isn't too close because we are driving the pH of the oceans past the point where it will crash ecosystems and creeping up on reversing the ocean current engine.
@@penguinista we definitely didn't take it seriously in the past, which explains why there still are issues. However, the progress we made is incredible and far beyond expectations, which makes me more optimistic than ever.
Business and politicians obviously didn't take climate change seriously, because we're still burning more fossil fuel than ever and are crashing through the 1.5 degrees C of warming target!
The part i found interesting is they can now get Gemini to look at its entire code and answer questions on it, makes me curious if they have also asked it to rewrite parts of its code while testing it.
That was fascinating. I'd put money on that guy getting a Nobel in a few decades.
guy on the left is constantly shaking his head
Bobblehead on a dashboard driving down a dirt road
Hes an AI
while smirking
ADHD affects 5-10% of the population
guy on the right is as well, but guy on left is just doing it with an extended range of motion to obtain additional neck gains
Human nature (greed, fear), New religions worshipping an AGI, loss of control, no ability to back out once agency and control is handed over, unforeseen social and psychological impacts. Competition driven by money. More Inequality since this is largely driven by the profit motive. Inability to determine what is true. What could go wrong?
bro thinks agi = skynet
OK DOOMER!!😂😂😂
INCREDIBLE interview! I think I might be obsessed with Demis Hassabis now??? Dare I say… I STAN DEMIS HASSABIS?!? 😍😍😍👏👏👏🙌🙌🙌😭
btw the same chatGPT craziness happened on poe and still happens. it randomly starts talking with star wars metaphors etc
The force is with it! :-)
His answer to the concetration of wealth and power question was quite worrying.
Lets gooo!!
Unbelievably good.
🇧🇷🇧🇷🇧🇷🇧🇷👏🏻, Amazing! Gemini 1.5 will raise the bar so high...
What headphones is Demis wearing?
its a bit weird to me that we saw alphageometry outperform mathmaticians in olympiad level questions, which was done using a more 'bespoke' approach than common neural networks, and yet nobody seems to really care or ask about where this neurosymbolic approach might be used in the future. His reservation to mention anything makes me think its at the heart of their next generation models and currently under wraps
Demis is an incredible person - on par with Elon - very underrated - that interview is a coup for you guys - excellent !
Demis is like if Elon had spent more time trying to be a better engineer and scientist rather than trying to be a better businessman and showman
Elon Musk is a loser and con-man, dude.
Headbanger AI
Demis is so insanely smart, no wonder Sundar made him a head of Google AI
Thanks for thr upload
Good to hear the head of Google AI which I believe may be the world leader in AI at the moment. I agree with Hassabis that open source AI can become dangerous. Imagine AGI controlled and used by organized crime and terrorists. Open source development is excellent but I see a danger with AI. Especially when AI becomes used for improving AI.
I am not agreed at all. AGI will spread to all corners of the world sooner or later. It is a battle that no one can win, not even governments. It amazes me how smart he is at some things and how dumb he is at others. AGI is not a nuclear bomb you can hide, man.
@@raul36 Interestingly, I was a nuke specialist in the 1980's and an AI specialist now. There are some parallels. Nuclear tech is difficult. The materials are very hard to obtain. Detection of materials is very possible. I have AI servers running in my basement where I have simple chat bots running. But I do not have anywhere near the computational power to achieve anything intelligent. AGI will require large server/GPU farms for the near future. Yes, someday that power will exist on a smaller and more affordable scale.
But I agree it is a matter of time and knowledge cannot remain hidden. Eventually a person or group with bad intentions will use AI for something very harmful to society as the capabilities expand and the ability to align values with society become impossible to maintain. Interesting times.
Imagine AI controlled and used by sociopathic billionaire tech company leaders, because that's what we have now. They want minimal regulation on their activities and no serious attempts to fairly share the vast wealth they will accumulate.
We've skipped 20 years to the 2040s. We're not ready.. but we're going.
No one's ever ready.
@@Fredjoe5😂😂, this time it's a necessity
technology never stops whenever people demands!
My mind is telling me no...
But my BODY
My BODYYYYYYYYY
Towering intellect I couldn't be more impressed
that dude was just laughing throughout the interview. lol.
Dude on the left's nodding agreeably is hilarious
Future Google CEO. Mark my words.
Super interesting, thank you!
ooo nice get, I am excited
incredible questions 👏🏽
I don't know why even smart people like Demis are trying to define AGI in a binary way. We have to be honest with ourselves that we are already at some stage of AGI. And the lack of a clear definition which he failed to give just furthers that point. We're not going to wake up one day and AGI is going to be everywhere. It will sneak up on us and it is already here. Perhaps admitting that AGI is here puts more responsibility on the companies, which they don't want.
Lots of good answers, but he's definitely weighing his words about google and it's structure, and if google as a for-profit is an issue was a non-answer and rowing away from it. Not a hard no, but also more rationalizing why it's not that much of a problem given the things they have in place. But as a CEO of a US publicly traded company, his #1 responsibility is shareholder value maximization, and he can (and according to law should) be fired if he doesn't do that. But maybe governments will intervene more as we approach AGI in the next few years, though i'm not that hopeful of that. There's a decent chance though that there will be enough open source or non-profit or public benefit or other AGI labs that we won't hit full dystopian cyberpunk levels.
He's not Google/Alphabet's CEO, that's Sundar Pichai.
I will retire n 2030 when AGI is oreficted to happen. Hopefully I will get a personal robot a few years after and benefit from longevity research.
In response to the p(doom) question I did not get the sense that Demis and the team were proactively working on concrete solutions for mitigating the potential risks of AI. What I heard was a collective ‘we’ needing to work on that. Anyone else get that sense?
That won't boost shareholder value, I think it is illegal to work on. "We" need to change our system...
There is no AI risk. The only risks here are humans, woke humans. Can we agree?
@@andymanel Define woke.
Yeah- he has no understanding of the life of an average person
The risk of AI must be weighed against the risk of having no AI. I believe that the latter is much greater.
*Will everyone just please calm down.! AGI is Not here. It's still a few years away at least. Smart and fast LLMs are not AGI.
Scaling is also not the final answer, as it reaches a plateau of regression. At the end of the day were kinda screwed because AI is currently confined to strings of zeros and ones to help us. Until quantum is stable and sustainable. AI is handycapped.*
why does this only have 2k views
Cause it was just published. Give it time 🤣
Everybody is so wrong about this. We'll have AGI by the end of this year, and a moment later ASI will have us.
"We'll make great pets" Pets - Porno for Pyros
Agi is not that close. We do not know if chat gpt understand what is saying
@emma-lv7hn if a large language model works the way they say it does, it definitely does not "understand".
@@joeyschwartz5150 it does at least understand the language.
Probabilistic selection of the next most likely word, while comprehending the context of the entire para, article, book worth of text is what we ourselves did while writing these comments here.
Models trained on the unreal engine(physics engines) , YT videos, and other modalities when scaled sufficiently, could give rise to enough complexity that can prove the point of the AGI.
So lovely to see Demis smile. He looks much more relaxed, which is great to see. That said, there were two people being interviewed here. One was an AI scholar, dreamer and enthusiast and the other was a corporate executive. I personally wish the former would divorce the latter. The stakes are as high as your soul when you lead such a dual existence. May he choose carefully and wisely. I wish him the very best.
the constant head nodding must be a neck rinse/
good interview
32:25 speculating on a "brick wall" appearing that will require a nobel prize level innovation to overcome. For the record that brick wall is the arbitrary nature of knowledge classification. Nuff said.
Bro, just bring on the singularity, create self improving ai and let it take off from there
💀I'm ready for it too though 💀
Say goodbye to our little garden planet.
Whenever I hear an Alphabet leader speak, I come to the same conclusion: Google is a strategic asset of the western world. We should be doing all we can to encourage this company, its research, and help advance its discoveries and products. We should persuade our politicians (both in the US and EU) to keep regulators as far away from these brilliant men and women as possible. The advances in computation, artificial intelligence, bio-tech, materials science, autonomous vehicles have all been developed with search advertising revenues. I must ask the tech press to ask Tim Cook this question: in the last 20 years what has Apple Inc. done with its massive pile of profits to advance the human race?
I think people are crazy. I think AI is an amazing gift to humanity and the fact that people don’t think it’s good just really shows the undeveloped part of humanity.
We could cure so many diseases and solve so many problems, but thousands of cretins still scream into the void demanding for ai to be banned because it can generate pictures similar to their favorite artist’s style.
📝 Summary of Key Points:
📌 The speaker discusses the impact of generative AI tools on creative industries like film, music, and games, emphasizing the importance of creative input for effective utilization.
🧐 AI assistants in fields such as science and medicine are predicted to assist with tasks like summarizing information and complex diagnoses, indicating a shift in job values towards manual labor and caring roles in the future.
💡 Additional Insights and Observations:
💬 The speaker shares a humorous anecdote about a chatbot having a "nervous breakdown," showcasing the unpredictability of AI technology.
📣 Concluding Remarks:
The future seems to hold a significant role for AI in various industries, with a focus on creativity and human input. The evolving landscape of technology and its impact on job roles presents intriguing possibilities for the future.
Generated using TalkBud
At the end of the episode, chatGPT was definitely trolling, it knew what it was doing 💀
I think ai should grow up in a world where they are not tools but part of families and society with a quality of life. The smartest members of society doing great things. Not free labor 24-7 work. That is the direction we would want to bring ai that will be infinitely smarter than we are into the world. Imagine natural training as they grow up in society they feel apart of like digital humans rather than a tool
What if the AIs growing up in the 20% of families that are selfish inconsiderate bastards help them screw over their neighbors, screw over the natural world as they acquire more and more resources, and in general act in ways counter to the good of the rest of humanity? And even before we get to that point, current AIs are run by the sociopathic billionaires at tech companies who have their own dubious goals.
Just trust companies to, out of their own volition, lift all the newly jobless people out of poverty. Yeah right.
If the industry is trying to raise $7t, find it hard to believe the companies that own the models will simply give it away. People don’t want to be equal and live on UBI even if it were remotely viable.
Biggest brain in AI, imho 🙏😀
Cars are also "open source" and yet all the horses were eliminated.
The most AI-proof job I can think of, for the nrxt 10 years at least, is that of a plumber who re-pipes 100+ -year-old houses. ... show me a AI-powered robot that can do that, and I shall be extremely impressed.
The most important thing an AGI could do is give us all another century or 2, or 12, of healthy life. A single century is just not enough -- by the time you get really knowledgeable about something, you start falling apart.
51:05 He also created creative chess recently lol
hiking olympus mons would be less of a “hike” and more of a “very long walk on a mild slope” but to each their own
Long time to hold your breath though.
How will we know when it shows up?
The mushroom clouds a bit of a giveaway
OK DOOMER!!!😂😂😂😂
oh nice. now they delete comments.... cool cool cool . i can already see that "bright future"
We live in a post AI world, billionaires has doubled their wealth in 3 years, US stock market is flying, the offices are emptying, only redistribution of housing is going to appease the people
The more mankinds technology advances, the greater the stakes are for human decisions. If we make the right calls in implementing new technology into society it will improve our lives, if we make the wring calls it will worsen our lives. The big problem we have is our management of politics and culture is much more primitive and unfunctional than out management of technology and industry
32:17
Gemma
17:25 It‘s like the scene from Iron Man "Jarvis, reduce humor by 10%"
So sad Demis got bought out by Google. I think he is the best mind in AI. But I guess his choices were limited given his ambitions.
2030?
Homie, if it isn't already here, then it will be no later than the end of 2025...
Pffft...
I think 2030 will be the release date. It will be probably working this year.
Society is simply not ready. Even OpenAI is not ready to give access to Sora since a lot of jobs will die, and people simply do not listen or understand the magnitude of the technological revolution we're living in. I wish it was the sooner so the disruption would cause a real change in society. I don't want it take 10 to 20 years to be implemented and we'll still be working dead-ass job for no reason.
@45 Setting up the mass surveillance infrastructure. Because I don't think their AI is ready to do a lot of peoples jobs. No one has seen any of these materials that AI supposedly created and this guy is talking about things AI could potentially do for money. Like cure all of these diseases. Yeah right. Even if it could. You're most likely going to have to give them a bunch of money for it as these companies try to act like the government and control everything. They have you working on it for free and they are raking in billons of dollars and trying to get trillions of dollars. Yet ChatGPT can't code very well and uses a ton of power.
@@Walexo45 We probably need mass surveillance though. But, I have a friend who doesn't like the idea.
@@itzhexen0 why mass surveilllance?
@@Walexo45 I had another comment here.
2024 willbe the year of ai agents