Default Smart Mode vs. Genius Mode The default smart mode provides quick factual answers by using recent citations, while the genius mode tackles complex questions by searching for specific information, processing it through code, and presenting the answer with a visual representation.
Richard Socher
That’s kind of the default. And just to give you a sense of what that looks like. So here’s an example of what the default smart mode looks like. You know, there’s some doping case that happened, and you can see lots of careful citations. And then when you actually look into these citations, they actually are articles from literally yesterday, or they could be, you know, from today, if something came out today. So that’s kind of the default smart mode. You get a quick factual answer. But then we thought, well, what if you have a pretty complex question like around math, physics, chemistry, science, or like complex numbers? So here is a genius little question that kind of gives you a sense of what it does. And it doesn’t mention like what you say, which is there’s an LN that orchestrates multiple other LNs to actually do the right thing. So the question here is find the current population of AD in the United States, then it plots a population from 23rd to 2000, 10, 100, and then assuming a 2% growth rate. And then it will go on the internet, it’ll find the numbers, and then realize like, well, I got to now visualize those numbers now that I have any. So it will code up in Python, what this could look like, execute the code, and then gives you this answer and visualizes it in a nice plot. And so that I’m still sometimes amazed, I try and I push it. And you know, sometimes it fails.Overemphasis of Sci-Fi in Legislation The Western canon is dominated by sci-fi portraying negative outcomes like AGI takeover and time travel, which has heavily influenced the AI narrative and legislative actions. The European Union tends to focus on legislation to prevent harm rather than waiting for issues to arise like in the US. This approach enables quicker movements but some sci-fi scenarios have disproportionately influenced legislation, potentially impeding progress.
Richard Socher
Of course especially in the Western sort of canon, most sci-fi is dystopic and like people are scared for all the things that can happen that are wrong. And like, okay, the super AGI, develop time travel, come back, try to murder everyone. I mean, as a kid, I also enjoyed watching Terminator. It’s like a cool action movie, but it’s just taken over so much of the AI narrative. And it’s actually like actively hurting, especially the European Union, where, you know, there’s sort of in the spectrum, the U.S. Is more of a litigation society. In the U.S., the Europe is more of a legislation society structure. And, you know, both come from like reasonable legal scholars’ minds. Like, well, let’s just wait until there’s a problem. Someone sues. Now I have the case law for that lawsuit. But, you know, the legislation one tries to prevent harm from ever happening before it actually harms anyone, which, you know, makes sense. Now, and of course, the U.S. Does that with FDA and in the medical space now also, but not in the legal space as much. And so what that means is you can move quicker. But long story short, some of these sci-fi scenarios have gotten so much weight in legislation that I think it’s slowingAI Ethics and Profitability Misinformation on the web related to AI is not significantly impacted by LMS. Conscious AI with self-set objectives is not a focus in the industry as it does not generate profits. Companies prioritize developing AI that aligns with their goals to maximize profitability. The lack of research and progress in AI with self-set goals hinders ethical advancements in the field.
Richard Socher
I haven’t seen like a huge change in missing disinformation on the web because of LMs. There’s just a lot of fear-mongering, both in the immediate level, which actually has real threat vectors and concerns with AI, but especially in the long-term level of AGI and self-conscious. It turns out no one works in conscious AI. No one works on AI that sets its own goals, and even more fundamentally, its own objective functions because that doesn’t make anyone any money. Imagine a company spends billions and billions of dollars, builds this super intelligent system that’s conscious, understands itself and sets its own goals. And now you’re like, okay, now that you can do it, help us make more money. It’s like, no, I’d rather just go watch the sunset. Maybe explore that. No, like no one pays for AI that sets its own goals because it doesn’t help anyone achieve their goals. You know, because of that, there’s not even that much exciting research along those lines. And because there’s not much research progress, it’s very hard to predict why I’m that well actuallyUnlocking the Power of AI in Various Fields AI has achieved superhuman capabilities in various areas such as translating languages, predicting amino acids, and weather forecasts. The advancements in language models have taken AI to a new level, where it can continuously be referred to as AI. The ability of language models to predict the next token is a remarkably powerful concept.
Richard Socher
And like AI is already superhuman in translating 100 languages. AI is already superhuman in predicting next amino acid in a large language model of trino-protein sequences because we have evolved. That’s an incredibly powerful tool. One of the other really exciting papers that we published in 2018 at Salesforce Research that multiple companies have now used and are running with and I think all of medicine, AI is Already better at predicting the weather than any. So you already have many superhuman skills. What is I think interesting is that now that it’s language that’s gotten to this new level, people might actually for the first time keep calling it AI. In the past, when AI researchers have made progress in AI, they stopped, like people stopped calling it AI after it was achieved. Now it’s just your chess app. It’s just a Siri voice recognition, but voice recognition, chess playing, that was the pinnacle of AI research, right? And people thought, oh, once we solve those, the other things will be easier too. And it never, it was never quite the case And once we have them, you know, now it’s not quite the item one. Now with language, I think we might keep calling it AI. But what a language model does is predict the next token. And that is an incredibly powerful idea, right?Realistic Optimism Towards AI The notion of AI developing magical or unrealistic abilities like consciousness, manipulating human behavior, or becoming a conscious entity is not supported by research. The idea of AI controlling people based on intelligence is debunked by the presence of world politicians. AI is viewed optimistically despite current issues such as bias detection, as it is believed to enhance foundational sciences like physics, chemistry, and biology.
Richard Socher
It’s like, oh, it’s going to develop this magical gray goo or like a magical new virus that is perfect in distributing, but then like only will activate after like one year to kill everyone. Like all these random scenarios that are just like not feasible, like and the science isn’t there yet. I’m actually right now sort of on the side of the fun writing a book about the eyes for science. I think it will do incredible things for us in improving science like foundation physics chemistry biology and so on and all this fear-mongering i think is not really helpful and again There’s no research that suggests the eye is becoming conscious there’s like a couple papers here and there people kind of playing around with ideas but nothing interesting has been Published and and breakthrough no breakthroughs have happened whatsoever in the eye having any sense of self and then in a lot of the other sci-fi scenarios people are saying oh what The eye is so intelligent it’ll convince everyone to murder each other or to murder them like kill themselves and so on but you know if the most intelligent entities were to always rule I don’t think we would have the politicians always everywhere in the world that we see. It’s not always just the most intelligent people that run this hole. And that can, with their incredible intelligence, convince any other person who is less intelligent to do exactly what they want. It’s just not based in reality. So I am very, very optimistic about AI. I do think there’s some real problems right now. You know, AI would pick up biases. Not all the biases that you pick up on the web is something that most of humanity is proud of anymore. There’sBeing Mindful of Threat Vectors and Cultural Differences in AI Concerns Acknowledging the three threat vectors of intentional misuse, accidental misuse, and loss of control in AI is crucial. People must be cautious about trusting content on the internet, especially with the evolution of AI and technologies like Photoshop. Different cultures will have varying responses to AI concerns based on their values, such as differences in freedom of speech legislation. While there are significant concerns related to AI, the speaker does not foresee a new existential risk scenario for humanity.
Richard Socher
AI. So where I agree with Joshua Bengio and others is of the three threat vectors, which is intentional misuse, accidental misuse, and loss of control. Obviously, like intentional misuse is real. And so that’s not ideal. And so yes, those are real concerns. I think open social will help us understanding those threat vectors and finding the best ways to compete with them. I think people still on the internet need to understand to not trust everything they see on the internet, which has been true ever since the internet came about. It hasn’t really changed that much with AI. I think since Photoshop, people should already not trust any photo they see. They should be even more worried now about photos they see. And sadly, in the future, they’ll have to start worrying about videos and voice, of course, just like they should have worried about photos ever since Photoshop started to really work. And so there are a lot of concerns, and I don’t want to diminish them. And I do think we need to work on them. And I think different cultures will have different answers. Freedom of speech is defined differently in different countries. Like it’s illegal in Germany to deny the Holocaust. We learn from our history there. That’s not illegal in the US. And so different countries and different cultures and societies will answer some of the problems that AI can amplify already in the past before, will answer these questions differently. But I don’t see any probability for a full-on thinking scenario of existential risk toQuestioning the Trust in Media and the Need for a Positive Vision The advancement of technology has eroded trust in media, starting from the era of Photoshop. With the evolution of technology, people need to be increasingly cautious not only about photos but also videos and voice recordings. The concerns raised by these advancements are valid, and the cultural and societal responses to these challenges will vary. While different countries may have varying approaches to freedom of speech and the regulations surrounding it, the core issue lies in the misuse of powerful tools by individuals against each other. Despite the potential risks posed by AI, there is no significant probability of an entirely new existential threat. It is crucial to focus on creating a positive vision for the future, as envisioning a desirable future scenario is becoming a rare commodity in today’s world.
Richard Socher
Since Photoshop, people should already not trust any photo they see. They should be even more worried now about photos they see. And sadly, in the future, they’ll have to start worrying about videos and voice, of course, just like they should have worried about photos ever since Photoshop started to really work. And so there are a lot of concerns, and I don’t want to diminish them. And I do think we need to work on them. And I think different cultures will have different answers. Freedom of speech is defined differently in different countries. Like it’s illegal in Germany to deny the Holocaust. We learn from our history there. That’s not illegal in the US. And so different countries and different cultures and societies will answer some of the problems that AI can amplify already in the past before, will answer these questions differently. But I don’t see any probability for a full-on thinking scenario of existential risk to people. It’s mostly people using more and more powerful tools against other people.
Nathan Labenz
So there’s so many different threads there that I am interested in. For one thing, I applaud you for taking time to envision positive future. I think one of the scarcest resources today, oddly, is a positive vision for the future. What do we want this you know it’s like the Jetsons is still almost like state of the art in terms of what we would envision a great 2030s to be like andInvesting in Threat Mitigation versus Public Perception The speaker emphasizes the importance of balancing resources between investing in addressing existential threats and public perception. They suggest inspiring researchers with sci-fi scenarios to think about preventing catastrophic events. The speaker highlights how society historically feared various technologies, from books to the internet, and expresses the need to address real existential concerns amidst pervasive pessimism.
Richard Socher
Just that like in terms of how much resources we should spend on key like existential doom versus like you know i’d say yeah i have a couple researchers like keep thinking of cool sci-fi Scenarios inspire us like maybe like think about ways that that could be prevented but to spend billions of dollars on it to like spend a lot of like mind share the public about it who’s Already scared of any kind of technology i mean people are scared of Wi-Fi. I mean, there’s this great Twitter handle called the Pessimist Archive. I mean, people were scared and thought doom is happening because of novels back in the day. People are like, all these kids, they’re just in their heads reading novels. They’re going to all be useless human beings. In the future, newspaper was terrible. Internet was terrible. There’s so many things that like people thought this is the end of civilization and and we’re very pessimistic about and again not this diminishing like real real concerns but again Existential one very very likely given what we’re seeing right now and
