Feeding a function to an LLM vs. feeding an entire codebase, unlocks new functionality that didn't exist before. So the quality may not have improved but it is able to do new things. I think that is why this 1M context window is huge. It also doesn't seem to be bounded. So with more compute you could feed it every book ever written and it could use all of that context for new outputs. And by feed I don't mean train but interact with.
This group can do better than this! The discussion around AI context, RAG and diffusion models will confuse more than educate (except maybe Friedberg's speculation at 16:39). On other topics, felt like Jason was pulling teeth to get them to converge towards a useful debate. Hope for better collaboration next week!
As someone who works as a 3d artist, and listen to everyone discuss this AI video vs unreal engine as if it's doing the same thing but better goes to show they don't understand what they are truly looking at with the Tokyo video and how it was not created with any 3d models or built off a self created physics system that the AI wrote itself, essentially it's scouring pre existing data to create the video, if the AI was building it's own internal game engine itself, it would allow you to move the camera to more precise angles and allow you do further edits to the final result ai created.
Bro you're partially saying what they also said. It's one of the current limitations of generative ai. It does however have some understanding or concept of physics, that's why it's able to show different elements together (so not only mimicking) but still acting in a way coherent with physics.
I really value when the pod talks about everything but politics.. everyone’s insight is so valuable .. the moment I hear them choosing a political side.. it makes me sceptical about their hedges… I know it increases engagement.. people like me debate and comment and their views increase..😂
This was one of the best podcasts I have every seen! Really rich conversation. You can tell these guys have put a lot of thought into AI and how it will affect the future of entertainment, work, human interaction, etc .
If Tucker interviewed Hitler after he invaded Austria these boys would be talking about how interesting it is that Adolf has such a keen view of history and his destiny to unite his dream of Germany. Friedburg would call out their nonsense while Sacks notes that Hitler doesn't seem to be too mad and all that's needed to achieve peace in our time is a brave Western leader to have a rational chat with him.
Putin's "offer" to join NATO was a very wise move on his part. If accepted, NATO might have become irrelevant (as Friedberg points out) and if declined he could have gained the high moral ground of saying "Well, I wanted to join but you wouldn't let me". He knew the most likely answer would be no. The point is he did this because it was a very smart strategic move that didn't cost him anything, not because of his "openness" to the West.
@chamath, I agree with what you said about how the new tech age will increase loneliness and decrease communication skills among our youths. Keen to learn more about how you are trying to make a difference with your kids and how you parent them to survive in this new world. Perhaps in a blog or another video? 😊
I heard Chamath’s comment about social media isolating our younger folk and I too am concerned about how social media is influencing us all. But, without social media, I would not be nearly as close to our four children (they are all grown up now). It’s a case of the ‘double edged sword’…a good servant but a bad master. At the end of the day, it comes down to education. We have always taught our children about ‘everything in moderation’…including social media and ensuring we control it, it doesn’t control us. But, we must be ever vigilant!
Love your podcast. Nothing better than to spend time learning from really smart people. This episode was spot on, Sacks and Chamath nailed the Putin position to become a productive part of the western economy. Putin is a businessman, Clinton saw that. Oh by the way, my parents were forced out of Hungry by Russia and Khrushchev in 1957. They left everything behind when Russian tanks showed up on the front lawn. So yes, I do have an opinion on Russia's past leadership. However, if you have at least an average IQ, you should be able to see the difference between Putin and Stalin leadership agenda's. Thank you for elaborating on the NATO conversation, I believe more people will understand why Putin is upset about a Ukraine Nato inclusion. Discrimination comes to mind...
Sacks’ straightforwardness is what we need more of in this country. No more sugarcoating. The problem is all of the educated are institutionalized and fail to see it. People need to start thinking for themselves.
Hey guys, great show again. Just wanted to share a quick insight. So, at around 22:45 mins Sacks talks about the DALLE model giving a 100 different answers if you give the same prompt a 100 times. And he is right. This happens coz AI models are by construct probabilistic and the initial config is randomly chosen and then it iterates towards the optimized result. But in regular ML we have solved this problem some time back with a technique called “setting seed”, essentially meaning that you set the same initial configuration. OpenAI introduced that in their latest GPT 4 Turbo LLMs and I’m sure very soon this feature of repeatable results would be available in DALLE as well.
My reaction to Putin's interview with Tucker was very similar to Jason's. It's tough to pin down a guy with a quarter century of statecraft experience plus however many years he spent in the KGB. Another great and thought-provoking episode, it's why I keep coming back for more. Thanks guys!
@@BBStyles777 take an history book and check everything he said, you will find out that there are multiple factually wrong statements he said during his "history teacher" section... I like listening to Putin, Chavez, Castro and many more dictators cuz we can learn a lot about strategy and game theory from them. But the issue is that they lie a lot, so is hard to trust their stories
Remember Tucker was at risk somewhat. When the NYT guys is in prison and Putin knows the US govt would just assume Tucker was locked up or killed …..if I’m Tucker, I DO NOT get too carried away with questions that shed him in a bad light. This interview I think was more to just allow to world to hear and see the bad man and see the Russian economy as well
Chamath point on the interview with Putin was one not to miss. Basically after the Cold War ended the most logical steps in the eyes of everyone was to; 1. Dismantling NATO If not the next acceptable option is 2. Accepting Russia in NATO by the logic that we are not enemies anymore and if we are going to have a global security force Russia should be included. If not, the last acceptable option is 3. Not NATO membership but very close cooperation between Russia and NATO - Unfortunately non of these happened. We could have different world today.
If America is indispensable nation, Russia is indispensable enemy, if Russia is a NATO member, who America will fight, how they can sell weapons? If they fight Cuba, how they will sell weapons?
All of this posturing by the government and it’s basically off of emotions and hunches, and what they may think is possible. How are you ever going to come to any type of resolve if you don’t sit down and negotiate I mean everything can’t be solved via war.
I heard Christopher Nolan liked SORA so much, he is willing to stop logging around that heavy Imax cam on his DP's back. It's also been said Emmanuel Lubezki is going to stop camping in remote locations to find out the exact precise moment something magical happens with the sun for a scene in a movie like The Revenant and he is also switching to SORA!
Nick, can you pull up this chart? And while you are doing that run this model to add a ponytail to Chamath. Also, find the clip of me saying xyz a few weeks ago... Love Nick 😂
One simple improvement to video/media that I would appreciate would simply be to take a great movie and set the Parental Guidance level, or make it more/less suspenseful or whatever.
24:33 to create an understanding of the output, a potential solution could be to self label the output and index it, so that it can retrieved to provide context to make small tweaks. It makes sense that the foundation model alone cannot understand the output, most likely because it is not set up to do so.
The Trump rant about Nikki Haley causing Jan 6th in my mind is the scariest cognitive moment of our two presidential candidates. It wasn’t a simple mistake, it was a wholesale cognitive breakdown. Just yesterday, Trump told Michigan voters they need to vote on November 27th. Yes I think both are too old, but ignoring this in the discussion is dishonest. Sacks is a full blown Cult45 member. I'm surprised he didn't quote Trumps doctor who said he would live to be 200.
Thanks for the Putin discussion. While I’m not a fan of Russian or Putin today, having little historical context does help to understand the other side. We have to deal with Today’s Putin, despite possible historical policy mistakes, but perhaps understanding history can help find a way out. Perhaps the near term there is hope for peace through negotiation Ukraine as a neutral zone (hands off for both Russian and NATO) (EU/Russian DMZ).
The sample A.I. movie video had teleporting background objects, and objects that transformed while in scene. Even the initial position of the walking people into the sceen was off scale.
Masa is not just a gambler. He has been operator in most of his career. His first business was MS software distribution and then he set up Yahoo! Japan(No.1 search engine in Japan) with Yahoo and acquired Vodafone and made it Japan's 2nd largest Telco operator. He is currently using most of his time in operating ARM. Also, even though Vision fund's big failure, his Telco business in Japan generates multi billion dollar of cashflow every year so he is always ready to make a big bet.
Seems like Palantir is one of the few that actually has a functional closed LLM model for business that is working .Ontology is required to be a winner in the business space
100% Chamath and Jason haven’t read the latest literature. I half expected DF to chime in and educate them. This vid explains perfectly why the mega context window does indeed help: ua-cam.com/video/Cs6pe8o7XY8/v-deo.htmlsi=p_tfbB6wxf6Of9tS
What you all missed at the end is that NATO is not just some defense pact, it's a political alliance. Russian admittance to NATO was very much possible. All that would have meant is that the nature of the alliance would have evolved from its original founding purposes, as alliances and pacts have done throughout history.
Chamath is actually very wrong on the context window point. Jason also misses why the architectural improvements to Transformers (whether through Mamba or Mistral) are significant. Excellent vid explaining it here: ua-cam.com/video/Cs6pe8o7XY8/v-deo.htmlsi=p_tfbB6wxf6Of9tS
Putin is career KGB officer guys. You’re way too smart to not understand what that means. Thus his current dictatorship and how he deals with any opposition. This is so tough to listen to 4 guys I really enjoy, kissing up to Putin
The difference between your total tax liability and your regular income tax, you get back as a credit in future tax years. This means you still get the money back even if you pay taxes up front for AMT, so long as you make income such that your tax liability eventually adds up to the credit
Sora takes one prompt and does everything to make the video. Once it is done it doesn't really know anything about what it just spit out. What it needs to be made to do is create a skeleton of prompts, and then recursively create each bone individually, before rendering the output using the pieces it created. I think this is also the key to eliminating hallucinations, such as dogs multiplying out of a single dog. This is sort of the merging of models and deterministic programming the boys talk about.
Chamath is correct. Putin was serious about joining NATO and the purpose of joining was to potentially fight terrorism (especially islamic) that both sides were dealing with.
Chamath wrong about the importance of context window size in regard to Gemini Ultra. The context window is not like parameter size, it allows you put a larger code base in there to ask questions, it allows to summarize larger texts, it allows for more memory in conversations. Annoying how overconfident and wrong he is.
He's insufferable arrogant and things he's a genius because he rode Zuckerbergs coattails to fame and glory. He was VP of growth at Facebook, a company that was already growing like wildfire. A little like being VP of happiness at Disney.
Leonardo ai allows consistent characters through images. Value will confer to application builders on top of these models that have unique product insight imo
25:28 - I think so as well, that way they might for example maybe solve the clipping of hands/legs issue using an animation rig and a 3d space environment for it (so it would allucinate coherent images on top of the rig and according to it to use it as helping guide for the AI model)
Where Friedberg says "fusion" of 3D rendering & LLM training, I say "feed-forward axioms to constrain how a model learns parameters". Axioms are wisdom that guide perception of new data. It's going to make them smarter per parameter. The axioms can be the parameters of domain-specific models that are the most efficient (smartest with the fewest parameters) via competitive training on ideal data.
@Chamath, the Boeing 737 MAX 8 issues are actually believed to be the cause of the issues that resulted in hundreds of deaths and the mass ground a few years ago. Boeing sought to save costs by moving from manual testing to automated and they overreached - automated testing tests exclusively for what it was told to test for. I believe in automated testing, but I also believe it’s situational and humans are still required (for the foreseeable future, at least).
40:00 yes but the bigger and more immediate threat would be to video producers who worked on producing short ads. we can upload our own video and make changes to it which gives massive power to us in producing short high level content within minutes and 1/100th of the cost if not even less.
Chamath’s Putin interview analysis lacked so much depth and was so superficial. “Putin, he is just misunderstood” on the day that Navalny dies in a remote penal colony. Really frustrating! Sacks is very one sided as well, though he has some valid points.
1:42 wrong. we tried to negotiate with Putin but we failed. Macron was giving him calls without success. Lets understand one thing - Ukraine joining NATO was not realistic. They would not have joined even if Russia were ok with that. Putin is lunatic. He justifies his atack by saying that those teritories belonged to Rus'' some time ago. Then Italy could claim all teritories that belonged to Roman empire. Mongols could think of rebuilding back of what they had during times of Golden Horde. Total lunatic. The ones who were thining to "negotiate" with him are simply delusional
This is the start of off-shoring or AI-shoring of actors. If i was an actor, i would invest time in understanding these technologies and focus on developing/producing movies instead of acting...I give that labor force 10 years to start receding and feeling the impact..and the major studios rejoicing.
It's my moral imperative to vote for the candidate I think is best, irrespective of their political party. Regardless of whether it is 'throwing away' a vote, I'll be voting for a 3rd party candidate (Robert Kennedy in my case) because I can't conscionably vote for either Biden or Trump. I would like to believe that veterans that fought for my rights to vote, would approve of this choice.
Open AI has Text to Image, Text to Video, and LLM, If they combine these three, they already have a phenomenal way to create artificial intelligence... OpenAI congrats on creating the first AGI
Have to agree with Sachs with curation. I'd rather see something thoughtful someone else has put together than just give a half assed prompt to AI myself. Conversely, if I'm doing the prompting to AI, I'm probably going to put a decent amount of effort into it and that's the product that will be more likely to have legs.
1:42:10 how come the world's greatest moderator decides to shut down what may be the most important discussion of the week? J-Cal, Are you actually afraid of the potentially uncomfortable truth, that we as Americans are responsible in part or whole for the death of all of the Ukrainian soldiers and civilians? Or worse, that our democracy is actually not a democracy at all? That in fact Americans are not the good people anymore? That perhaps we as Americans owe amends to the world?
Putins ask to join nato was not to join the anti russian group, but it would have started the process of reducing arms race, & move towards resolving socio economic issues of poverty or infrastructure. Pure theoretical question that tests the desire for peace.
Filmmakers will be vertically integrated if they utilize the blockchain to tokenize their films and distribute the via the blockchain. The opportunity to buiild your own global audience is the future
Hell yea, screw the plebs... bots ftw. Can't believe the most profitable companies in human history had the audacity to ever give their employees equity and foster an environment of individual ownership and commitment. Luckily those days are coming to an end. Scary indeed
You need to take tests for a California drivers license (vision, driving test, paper test) Seems like a no brainer to at least be aware of cognitive abilities of candidates for president, along with current president. Why are we allowing leaders, who can’t see 3 ft in front of them, license to drive the car off a cliff?
We have a minimum age for POTUS, but I'm sure there are people under the age of 35 that are cognitively able to take on the job. Why would we not do the same thing for a max age? I don't really care if they can or can't pass a cognitive test... it's just the rules. Should be the same age as when someone is eligible for social security.
1:35:31 this is probably why we need new leaders and term limits. If anyone is in power for decades, they’ll accumulate a lot of these personal grievances over time that might affect their judgement on present matters and realities.
Feeding a function to an LLM vs. feeding an entire codebase, unlocks new functionality that didn't exist before. So the quality may not have improved but it is able to do new things. I think that is why this 1M context window is huge. It also doesn't seem to be bounded. So with more compute you could feed it every book ever written and it could use all of that context for new outputs. And by feed I don't mean train but interact with.
Ai is perhaps the dumbest overhyped nonsense people will blindly worship. Thats the true concern I have is not what Ai, but what people think Ai is.
@@eggnogaddict6287lol clearly your understanding of the technology is not very deep
Imagine rewriting your own GOT ending 😍😍😍😍😍
Sounds wack
i fucking love you Friedburg. voice of sound reason
Friedberg for president !
He has a tremendous mind. Level headed, unemotional, great reasoning and learning ability.
Love David Sacks. Thanks for your knowledge of history.
There should be an AI that processes the entire All-In-Podcast in Nats Italian voice
This group can do better than this! The discussion around AI context, RAG and diffusion models will confuse more than educate (except maybe Friedberg's speculation at 16:39). On other topics, felt like Jason was pulling teeth to get them to converge towards a useful debate. Hope for better collaboration next week!
As someone who works as a 3d artist, and listen to everyone discuss this AI video vs unreal engine as if it's doing the same thing but better goes to show they don't understand what they are truly looking at with the Tokyo video and how it was not created with any 3d models or built off a self created physics system that the AI wrote itself, essentially it's scouring pre existing data to create the video, if the AI was building it's own internal game engine itself, it would allow you to move the camera to more precise angles and allow you do further edits to the final result ai created.
Do you understand ML and diffusion model or read the paper?
Right? Fireship explained it better on UA-cam, Friedberg doesn't know what he's talking about.
They said that
Bro you're partially saying what they also said. It's one of the current limitations of generative ai. It does however have some understanding or concept of physics, that's why it's able to show different elements together (so not only mimicking) but still acting in a way coherent with physics.
@@upstar21twho or what is fireside? Link?
1:15:01 agree with Dan!!! It’s the same in the UK. Voters don’t get to pick their leader- the parties do; and that’s not democratic.
“Rogue one, not optional “ J Cal is right!
It's the best movie in the modern series by a long way
Incredible deep dive into the latest AI developments - bravo!
I wish freidberg would want the moderator spot
Please discuss Navalny next episode. Don't beat around the bush, do you think Putin had him disposed or did he die of natural causes? Go!
I really value when the pod talks about everything but politics.. everyone’s insight is so valuable .. the moment I hear them choosing a political side.. it makes me sceptical about their hedges… I know it increases engagement.. people like me debate and comment and their views increase..😂
This was one of the best podcasts I have every seen!
Really rich conversation.
You can tell these guys have put a lot of thought into AI and how it will affect the future of entertainment, work, human interaction, etc .
If Tucker interviewed Hitler after he invaded Austria these boys would be talking about how interesting it is that Adolf has such a keen view of history and his destiny to unite his dream of Germany. Friedburg would call out their nonsense while Sacks notes that Hitler doesn't seem to be too mad and all that's needed to achieve peace in our time is a brave Western leader to have a rational chat with him.
Perfect analogy!
Putin's "offer" to join NATO was a very wise move on his part. If accepted, NATO might have become irrelevant (as Friedberg points out) and if declined he could have gained the high moral ground of saying "Well, I wanted to join but you wouldn't let me". He knew the most likely answer would be no.
The point is he did this because it was a very smart strategic move that didn't cost him anything, not because of his "openness" to the West.
@chamath, I agree with what you said about how the new tech age will increase loneliness and decrease communication skills among our youths. Keen to learn more about how you are trying to make a difference with your kids and how you parent them to survive in this new world. Perhaps in a blog or another video? 😊
I heard Chamath’s comment about social media isolating our younger folk and I too am concerned about how social media is influencing us all. But, without social media, I would not be nearly as close to our four children (they are all grown up now). It’s a case of the ‘double edged sword’…a good servant but a bad master. At the end of the day, it comes down to education. We have always taught our children about ‘everything in moderation’…including social media and ensuring we control it, it doesn’t control us. But, we must be ever vigilant!
Diplomats have mandatory retirement at 65.
You guys are getting really good. Keeps a smile on my face.
Great comments by Sacks on Putin👍
Love your podcast. Nothing better than to spend time learning from really smart people. This episode was spot on, Sacks and Chamath nailed the Putin position to become a productive part of the western economy. Putin is a businessman, Clinton saw that. Oh by the way, my parents were forced out of Hungry by Russia and Khrushchev in 1957. They left everything behind when Russian tanks showed up on the front lawn. So yes, I do have an opinion on Russia's past leadership. However, if you have at least an average IQ, you should be able to see the difference between Putin and Stalin leadership agenda's. Thank you for elaborating on the NATO conversation, I believe more people will understand why Putin is upset about a Ukraine Nato inclusion. Discrimination comes to mind...
Navalny's fate seems pretty reminisent of how Stalin dealt with peaple who criticized him.
Sacks monologue at the end was sooo amazing. Sacks, the more I listen to this podcast, the more I respect you most of all. Great show.
Sacks’ straightforwardness is what we need more of in this country. No more sugarcoating. The problem is all of the educated are institutionalized and fail to see it. People need to start thinking for themselves.
Hey guys, great show again.
Just wanted to share a quick insight. So, at around 22:45 mins Sacks talks about the DALLE model giving a 100 different answers if you give the same prompt a 100 times. And he is right. This happens coz AI models are by construct probabilistic and the initial config is randomly chosen and then it iterates towards the optimized result. But in regular ML we have solved this problem some time back with a technique called “setting seed”, essentially meaning that you set the same initial configuration. OpenAI introduced that in their latest GPT 4 Turbo LLMs and I’m sure very soon this feature of repeatable results would be available in DALLE as well.
My reaction to Putin's interview with Tucker was very similar to Jason's. It's tough to pin down a guy with a quarter century of statecraft experience plus however many years he spent in the KGB. Another great and thought-provoking episode, it's why I keep coming back for more. Thanks guys!
But still for Sachs Putin is a guy that could be negotiated with. Just as another opposition leader was murdered in Russia.
@@einerus He can be negotiated with if you’re in a position of power. Navalny was clearly not in that position.
@@BBStyles777 take an history book and check everything he said, you will find out that there are multiple factually wrong statements he said during his "history teacher" section... I like listening to Putin, Chavez, Castro and many more dictators cuz we can learn a lot about strategy and game theory from them. But the issue is that they lie a lot, so is hard to trust their stories
Remember Tucker was at risk somewhat. When the NYT guys is in prison and Putin knows the US govt would just assume Tucker was locked up or killed …..if I’m Tucker, I DO NOT get too carried away with questions that shed him in a bad light. This interview I think was more to just allow to world to hear and see the bad man and see the Russian economy as well
Chamath point on the interview with Putin was one not to miss.
Basically after the Cold War ended the most logical steps in the eyes of everyone was to;
1. Dismantling NATO
If not the next acceptable option is
2. Accepting Russia in NATO by the logic that we are not enemies anymore and if we are going to have a global security force Russia should be included.
If not, the last acceptable option is
3. Not NATO membership but very close cooperation between Russia and NATO
- Unfortunately non of these happened. We could have different world today.
Refreshing to finally hear J Cal make statements against Biden and how awful of a candidate he is.
You call that critical?? He’s probably crying in his cups hoping Hillary will make a run! 😂
If America is indispensable nation, Russia is indispensable enemy, if Russia is a NATO member, who America will fight, how they can sell weapons? If they fight Cuba, how they will sell weapons?
All of this posturing by the government and it’s basically off of emotions and hunches, and what they may think is possible. How are you ever going to come to any type of resolve if you don’t sit down and negotiate I mean everything can’t be solved via war.
No censorship on the “All-In Podcast”! It doesn’t exist!
Pft!
I heard Christopher Nolan liked SORA so much, he is willing to stop logging around that heavy Imax cam on his DP's back. It's also been said Emmanuel Lubezki is going to stop camping in remote locations to find out the exact precise moment something magical happens with the sun for a scene in a movie like The Revenant and he is also switching to SORA!
Underated comment
Nick, can you pull up this chart? And while you are doing that run this model to add a ponytail to Chamath. Also, find the clip of me saying xyz a few weeks ago... Love Nick 😂
One simple improvement to video/media that I would appreciate would simply be to take a great movie and set the Parental Guidance level, or make it more/less suspenseful or whatever.
24:33 to create an understanding of the output, a potential solution could be to self label the output and index it, so that it can retrieved to provide context to make small tweaks. It makes sense that the foundation model alone cannot understand the output, most likely because it is not set up to do so.
Great discussion as always! 🎉
Finally talking about the Putin Puppet Show aka red meat for Sacks 🤣
Would have been good to bring in Caleb Ward as a guest and get his take on Sora
Newsom cleaned SF street for Winnie the Pooh. Very nice.
Thanks for bleeping the swears! Doesn't hurt the flow and lets me listen with the kids in the car.
The Trump rant about Nikki Haley causing Jan 6th in my mind is the scariest cognitive moment of our two presidential candidates. It wasn’t a simple mistake, it was a wholesale cognitive breakdown. Just yesterday, Trump told Michigan voters they need to vote on November 27th. Yes I think both are too old, but ignoring this in the discussion is dishonest. Sacks is a full blown Cult45 member. I'm surprised he didn't quote Trumps doctor who said he would live to be 200.
Thanks for the Putin discussion. While I’m not a fan of Russian or Putin today, having little historical context does help to understand the other side. We have to deal with Today’s Putin, despite possible historical policy mistakes, but perhaps understanding history can help find a way out.
Perhaps the near term there is hope for peace through negotiation Ukraine as a neutral zone (hands off for both Russian and NATO) (EU/Russian DMZ).
Frieberg @ 1:14 brilliant political analysis!
The sample A.I. movie video had teleporting background objects, and objects that transformed while in scene. Even the initial position of the walking people into the sceen was off scale.
Masa is not just a gambler. He has been operator in most of his career. His first business was MS software distribution and then he set up Yahoo! Japan(No.1 search engine in Japan) with Yahoo and acquired Vodafone and made it Japan's 2nd largest Telco operator. He is currently using most of his time in operating ARM. Also, even though Vision fund's big failure, his Telco business in Japan generates multi billion dollar of cashflow every year so he is always ready to make a big bet.
Seems like Palantir is one of the few that actually has a functional closed LLM model for business that is working .Ontology is required to be a winner in the business space
in context learning ... brotatos, the mega context window does help...
100% Chamath and Jason haven’t read the latest literature. I half expected DF to chime in and educate them.
This vid explains perfectly why the mega context window does indeed help:
ua-cam.com/video/Cs6pe8o7XY8/v-deo.htmlsi=p_tfbB6wxf6Of9tS
“Skip the last three”…nailed it
I had the same issue with editing images in Dalle as David is having. It really diminishes the utility.
Looking forward a follow up episode on which one worked or is more efficient
Options (via the ownership mentality) for everyone is a feature, not a bug.
What you all missed at the end is that NATO is not just some defense pact, it's a political alliance. Russian admittance to NATO was very much possible. All that would have meant is that the nature of the alliance would have evolved from its original founding purposes, as alliances and pacts have done throughout history.
Chamath is actually very wrong on the context window point. Jason also misses why the architectural improvements to Transformers (whether through Mamba or Mistral) are significant.
Excellent vid explaining it here:
ua-cam.com/video/Cs6pe8o7XY8/v-deo.htmlsi=p_tfbB6wxf6Of9tS
Chamath's face when it seems NVIDIA is going to tank 🙂vs when he realises they're halfway to the moon 🙃
Chamath doesn’t understand the benefit of a larger context window… increased memory will yield better results.
Putin is career KGB officer guys. You’re way too smart to not understand what that means. Thus his current dictatorship and how he deals with any opposition.
This is so tough to listen to 4 guys I really enjoy, kissing up to Putin
And the US politician is better?
The difference between your total tax liability and your regular income tax, you get back as a credit in future tax years. This means you still get the money back even if you pay taxes up front for AMT, so long as you make income such that your tax liability eventually adds up to the credit
Sora takes one prompt and does everything to make the video. Once it is done it doesn't really know anything about what it just spit out.
What it needs to be made to do is create a skeleton of prompts, and then recursively create each bone individually, before rendering the output using the pieces it created. I think this is also the key to eliminating hallucinations, such as dogs multiplying out of a single dog.
This is sort of the merging of models and deterministic programming the boys talk about.
Free speech - Just not in the comments section. Lots of comments being deleted.
Chamath is correct. Putin was serious about joining NATO and the purpose of joining was to potentially fight terrorism (especially islamic) that both sides were dealing with.
Nice timing. Time to get our Friday night fix.
Chamath wrong about the importance of context window size in regard to Gemini Ultra. The context window is not like parameter size, it allows you put a larger code base in there to ask questions, it allows to summarize larger texts, it allows for more memory in conversations. Annoying how overconfident and wrong he is.
He's insufferable arrogant and things he's a genius because he rode Zuckerbergs coattails to fame and glory. He was VP of growth at Facebook, a company that was already growing like wildfire. A little like being VP of happiness at Disney.
Leonardo ai allows consistent characters through images. Value will confer to application builders on top of these models that have unique product insight imo
25:28 - I think so as well, that way they might for example maybe solve the clipping of hands/legs issue using an animation rig and a 3d space environment for it (so it would allucinate coherent images on top of the rig and according to it to use it as helping guide for the AI model)
An awful lot of words to avoid saying out loud that “democracy”, whatever that means these days, obviously doesn’t work.
Why not try Mezcal, makes all Tequila taste like shit haha
Where Friedberg says "fusion" of 3D rendering & LLM training, I say "feed-forward axioms to constrain how a model learns parameters". Axioms are wisdom that guide perception of new data. It's going to make them smarter per parameter. The axioms can be the parameters of domain-specific models that are the most efficient (smartest with the fewest parameters) via competitive training on ideal data.
@Chamath, the Boeing 737 MAX 8 issues are actually believed to be the cause of the issues that resulted in hundreds of deaths and the mass ground a few years ago. Boeing sought to save costs by moving from manual testing to automated and they overreached - automated testing tests exclusively for what it was told to test for. I believe in automated testing, but I also believe it’s situational and humans are still required (for the foreseeable future, at least).
Great episode!! Thank you
Sick episode guys
40:00 yes but the bigger and more immediate threat would be to video producers who worked on producing short ads. we can upload our own video and make changes to it which gives massive power to us in producing short high level content within minutes and 1/100th of the cost if not even less.
Chamath’s Putin interview analysis lacked so much depth and was so superficial. “Putin, he is just misunderstood” on the day that Navalny dies in a remote penal colony. Really frustrating!
Sacks is very one sided as well, though he has some valid points.
Chamath should take a cognitive test before the next episode.
I'm with Friedburg. This will be the first time I vote third party. I can't support either at this point.
1:42 wrong. we tried to negotiate with Putin but we failed. Macron was giving him calls without success. Lets understand one thing - Ukraine joining NATO was not realistic. They would not have joined even if Russia were ok with that. Putin is lunatic. He justifies his atack by saying that those teritories belonged to Rus'' some time ago. Then Italy could claim all teritories that belonged to Roman empire. Mongols could think of rebuilding back of what they had during times of Golden Horde. Total lunatic. The ones who were thining to "negotiate" with him are simply delusional
Tequila sales? All-In should next invest in ($) or promote tobacco products.
Brigadoon cigar co
There should be a max age to run for a president not just a minimum age. A person over 70 shouldn't be a president.
This is the start of off-shoring or AI-shoring of actors. If i was an actor, i would invest time in understanding these technologies and focus on developing/producing movies instead of acting...I give that labor force 10 years to start receding and feeling the impact..and the major studios rejoicing.
Why can’t people just say: bullshit, it’s over. The US population is an abused spouse at this point. Leaving is the only option.
outstanding guys!
Yo Sacks, you need to use the same seed on your prompt, when you do iterations..
Right wing grifters gonna grift.
@saks try using 'seed' in your image creation - that will keep the bulk of the image the same. Or use in-painting to change a small part
PLEASE get Stephen Kotkin on. He's a wealth of knowledge on the Russia-Ukraine region. A conversation between him and Sacks would be enlightening
The other thing about Putin interview he was not arrogant and dismissive ... very important
It's my moral imperative to vote for the candidate I think is best, irrespective of their political party. Regardless of whether it is 'throwing away' a vote, I'll be voting for a 3rd party candidate (Robert Kennedy in my case) because I can't conscionably vote for either Biden or Trump. I would like to believe that veterans that fought for my rights to vote, would approve of this choice.
Great discussion , thank you
Open AI has Text to Image, Text to Video, and LLM, If they combine these three, they already have a phenomenal way to create artificial intelligence... OpenAI congrats on creating the first AGI
Have to agree with Sachs with curation. I'd rather see something thoughtful someone else has put together than just give a half assed prompt to AI myself. Conversely, if I'm doing the prompting to AI, I'm probably going to put a decent amount of effort into it and that's the product that will be more likely to have legs.
1:42:10 how come the world's greatest moderator decides to shut down what may be the most important discussion of the week? J-Cal, Are you actually afraid of the potentially uncomfortable truth, that we as Americans are responsible in part or whole for the death of all of the Ukrainian soldiers and civilians?
Or worse, that our democracy is actually not a democracy at all? That in fact Americans are not the good people anymore? That perhaps we as Americans owe amends to the world?
R.I.P. Navalny
You will not be forgotten
Chamath going full Dunning-Kruger on AI. Surprised Friedberg (who actually understands AI) didn’t correct him on context window.
Googles Lumiere is set to be released soon. It has editing video capabilities and will compete with this.
Jason you don’t need an overly large context window if you are using RAG.
Putins ask to join nato was not to join the anti russian group, but it would have started the process of reducing arms race, & move towards resolving socio economic issues of poverty or infrastructure. Pure theoretical question that tests the desire for peace.
Get well soon, Chamath! All the best.
Filmmakers will be vertically integrated if they utilize the blockchain to tokenize their films and distribute the via the blockchain. The opportunity to buiild your own global audience is the future
Hell yea, screw the plebs... bots ftw. Can't believe the most profitable companies in human history had the audacity to ever give their employees equity and foster an environment of individual ownership and commitment. Luckily those days are coming to an end. Scary indeed
You need to take tests for a California drivers license (vision, driving test, paper test)
Seems like a no brainer to at least be aware of cognitive abilities of candidates for president, along with current president. Why are we allowing leaders, who can’t see 3 ft in front of them, license to drive the car off a cliff?
I disagree with sacks, on offshore work. Measured offshore tasks, is way more productive.
We have a minimum age for POTUS, but I'm sure there are people under the age of 35 that are cognitively able to take on the job. Why would we not do the same thing for a max age? I don't really care if they can or can't pass a cognitive test... it's just the rules. Should be the same age as when someone is eligible for social security.
1:35:31 this is probably why we need new leaders and term limits. If anyone is in power for decades, they’ll accumulate a lot of these personal grievances over time that might affect their judgement on present matters and realities.