Dangerous Question: Has AI Been A Disappointment So Far? | Cal Newport

Clearly, a huge amount of money is being invested into new generative AI tools built on large language models such as chatbots. And it just goes without saying this has been massively disruptive. Most of the conversation is about how do we deal with the disruption that has been caused by this new technology. But think about this question for a second. What actually are the things that this massive investment in generative AI through language models, what actually are the things that this has changed? Because if you ask people, they don't always have an immediate answer to offer up. Is that possible? Well, I want to investigate this. So here's the way I'm going to do this. I found a Reddit thread that was from three months ago. I'll put this on the screen here for people who are watching. It says the same question. With all the billions or trillions going into it, what has AI actually done other than as a search engine and funny pictures? So this is a thread where people from all over the internet are going to write in and do their best to explain all the various things that we've gotten from this massive investment in new AI technologies. So whatever these benefits are, you would think they're in this very popular thread. So what we're going to do is I'm going to go through this Reddit thread. to try to come up with a list of the things that this recent massive invest in AI has produced. Now, a couple of ground rules to this. I'm not going to include AI applications that aren't based primarily on the new large language models that have received the bulk of this investment, so the type of models that OpenAI and Anthropic have been producing. There's a lot of existing machine learning driven AI tools that we've been working on for year after year. And over the time, those have made progress. I'm going to separate that because that's just the standard progress in AI and it's not connected to anything new that has happened. I also want to separate into its own category the one obvious place where we hear most often about language models making an impact, which is computer programming. Computer programming is like a best case scenario for generative AI. It's a highly structured language, and we have a huge amount of very relevant data on which to train these. So we know it's been really impactful there. So we'll kind of keep that as its own column. What I'm looking for as we go through here, then, what I'm actually going to write down as we go through here is examples that depend on the generative AI models that we put billions of dollars into the last three or four years, but it's not computer programming. Because clearly, if this has really changed our world, that list should be really long. All right, so I'm going to load this back up again. And I'm going to go through, I mean, this is a very long thread, so I'm just going to do the best I can to go through this. All right, so the first thing we see here is assisting medical and scientific research and translating ancient documents written in dead languages. Neither of those are really from generative AI. Assisting medical and scientific research. I mean, there are a lot of tools that we're using for medical research. We're building AI tools, but that's using existing technology that's been around pre-gen AI boom. Translation documents, that's also machine learning tools that have been around. We've been getting better at them, but that's not directly related to the recent boom. All right, let's keep going here. AI as a tool is incredible at very specific things, mostly things human finds being incredibly mind-numbing and meticulous. Its best use by far is pattern finding in... wide data sets that are hard to pick up by hand. So I'm going to put that down. Pattern finding. I'm just writing this on paper here. So pattern finding. I've heard this. You can give some data to a chatbot and it can perhaps find you some patterns. Let's keep going. AlphaFold protein structure database. AlphaFold is not directly based off of the recent investments in Genitive AI. They've been working on that model for well before the LLN breakthrough when it uses Its own technologies that have, it has some new versions have some overlap with transformer attention ideas, but that is sort of a separate trajectory. That's not a result of the billions and billions that have been invested in large language models. So I'm going to put that to the side. All right, next thing. It speeds up a lot of menial computer tasks. I had it make a PowerPoint from six MS word size document pages for me. It made 60 slides, like a minute would have taken me an hour to do the same thing. I'm going to write that down. Like I'll put down slide. production. It's more generally, it can produce slides. I've seen that in PowerPoint. That's a good, and that required the investments that have happened. So we'll put that down. Let's see. It's incredible tool for programming. Okay. So we're putting that aside. There's an argument about that. Someone else says it's a lot of bad code and then all the cloud code people come out with their torches and get mad at them. Here's a different one. I mean, It really is useful for quickly disseminating a large volume of boring info into bite-sized pieces and identifying points of interest. I think this is right. So I'm going to put down summarizing text. So we've seen this a lot. Like you can take a bunch of notes, like a meeting transcript, and say, let's give me the main points here, summarize this in the bullet points. LanguageWell is very good at that. All right, let's keep rolling here. Translation of any language to another, that's not primarily a language model based thing. We've been doing that for a while. It's being used to help detect cancer earlier than humans detect it. All of that sort of medical prediction model research is not, those are not underlying it language models. That's something people have been working on for decades and we keep getting a little bit better at it, but that's sort of unrelated to the billions being invested in large language models. We'll put that aside. Discussion, discussion, mapped every protein. We talked about that. That's not large language models. Let's keep moving here. Someone else described a bunch of stuff, and then a response to that said, most of the stuff you're describing isn't break-produced in the last three years. It's slow incremental process over the last 20 or more years. That's right. So he was pointing out a whole long example of things that was not actually generative AI. Someone says, take customer support. Call centers are already so heavily scripted, they might as well be AI. Make AI good enough to hold a rudimentary guided conversation, and it can replace basically every first-level support person. That's some speculation into the future, but I will put down customer support agents because there are more of those. I don't think it's necessarily replacing the industry as quickly as people predicted, but there are better sort of first line chat agents that language models in particular made those good enough to deploy in a way that they wouldn't have been before. All right, let's go here. I guess that's partially language models are probably involved in that. The language translation is something we've been doing from before. It's not really a generative AI. breakthrough the open ai has been working on voice to text they have a special model for that their sort of whisper model which they then have on the front end of chat gpt so like if you talk to your phone into the chat gpt app you really have a separate smaller model that's been trained it's a different type of architecture than a language model that converts that the text and then the text is submitted to the large language model so that's cool but again that is not a result of the billions invested in large language models we were already doing that Then someone came along, this is interesting, and said, you can do things like tell it, read this congressional bill and note anything that seems illegal, unconstitutional, or would adversely affect separation of powers, checks and balances of civil liberties ranked by risk and describe the worst case scenario, the worst possible abuse of the change concerning an executive branch intent on consolidating power for itself. And the guy's like, oh, that's the type of thing you could ask GPT-5 to do. But then someone else wrote back and said, LLMs can't actually do that though. There's no reason, no imagination. They can't create and explore scenarios in the way you're explaining. They're not going to give you good advice. I'm not going to write anything down. Someone else said, check this year's Nobel Prize in chemistry. Well, that was given to DeepMind for AlphaFold, which again is non-language model technology that predates the billions invested. Someone else says it can make pretty good Hinta. And someone else said actually it's slop. So I guess I won't write that down. Someone else said, write the answers to my stupid quarterly self-review questions. That is true. write boring text. Yeah, so it is really good for that. Like, oh my God, no one cares about this. I have to write a self-review. It'll write boring, you know, good enough text that covers what you want. It's very good at that. All right, let's do a couple more because I think we're seeing some patterns here. Someone else says there's thousands of AI projects you don't hear about because they're corporate trade secrets. That's not true. There's a lot of that push of like, no, there's all these amazing things you just don't know about yet. They're coming. Let me tell you, as someone who's studied this for the last five years, they've, we've always been quote unquote, three months away from learning about the amazing thing that open AI or Anthropic or Google or Microsoft are doing behind the scenes. I have a lot of sources at these places. There aren't magic projects. There isn't, I mean, we heard this two years ago that open AI had AI was now doing all their programming. And within a few months, they're going to start like leaping ahead and models completely made by AI. That was just made up. It's just not true. They're using the same coding agents that anyone else is using. So I get that. We'll do one or two more. Small QL improvements for businesses automatically sending text messages to confirm appointments a few days prior, things like that. Sure, you can call it AI, but none of that has to do with advances in language models that have taken all these billions of investments. And then someone says fake videos and influencers pushing fascism. There is a lot of fake videos. That's not a... Good thing, though. That's a bad thing. And then I saw something else in here. I'll just write this down because I can't quite find it. But basically, someone was talking about it's good for pulling out or summarizing complicated information. And they talked about taking the municipal parking rules for where they lived, and then ChatGPT could kind of summarize them into something that's easier. So make sense of complicated text. All right, whatever. We've been going for a while here, right? Let me look at this list I just wrote. So this is, you know, AI is, the world is going to be unrecognizable. It's the biggest thing that's ever happened to us. How are, why are our kids even going to college? Like this is where we are now. Here is the list of things from this open discussion on the internet. Anyone who can come and tell us how the billions invested recently in AI have changed the world. Here is the total list of non-computer programming ideas that came out of this. Pattern finding, producing slides, summarizing text in the bullet points. customer service agents, writing a boring text automatically and helping to make sense of complicated text. Those are cool, Jesse. But if I just, if I five years ago said, imagine a tool that could do those six things, you weren't going to like Simpson style crash out of the window and then like jump on the back of a horse and ride off. It's like, oh, that's, yeah, it sounds like the good. I'm glad we're making progress on that. So it's kind of interesting. And look, I don't want to go too far here and be like this, revolution doesn't matter. But when you put aside the computer programming stuff, which is a discussion for another time, complicated, more complicated than people make it out to be, but that's a discussion for another time. And when you put aside other AI stuff that's unrelated to all these billions we invested, what we are getting right now out of the whatever it is, $500 billion that have invested into generative AI, there's not a ton of home runs yet. And I think this is a difficult point. for AI boosters is why there's a lot of like, what is going to happen next year? And we've done this five years in a row now almost. What is going to happen next year is where the conversation goes, because I think it's hard not to admit the real concrete things are coming a little slower than we think. Now, we see these amazing demos. We see these things do great on benchmarks. and we have all these scary scenarios which are real like look at these fake videos what's going to happen the truth or students can produce their papers on these things but if we really think about it the stuff that like what's changing your mind people are using this like a better version of google and then they're very narrow right now now this doesn't mean much bigger changes aren't coming but i thought this was an interesting sort of experiment to do that when we actually look at what has already come specifically from investing all of this money into the generative ai it's hard not to conclude that like so far um it's not massive yet hey if you like this video i think you'll really like this one as well check it out

Cal Newport examines if AI has been a disappointment so far in this clip from a recent episode of the Deep Questions podcast.

Buy Cal Newport's latest book, “Slow Productivity” at www.calnewport.com/slow 

Download my FREE Deep Life Guide HERE: https://bit.ly/3QBIcug

Listen to Episode Here:  https://www.thedeeplife.com/listen/

Get your questions answered by Cal! Here’s the link: https://bit.ly/3U3sTvo

0:00 Is AI a disappointment?
3:20 Pattern finding
6:10 AI and customer support
8:50 Corporate trade secrets

Connect with Cal Newport:

🔴Visit Cal's BLOG and website:             https://calnewport.com/blog/
🔴Check out Cal's books:                         https://calnewport.com/writing/
🔴Check out The Deep Life:                     https://thedeeplife.com

About Cal Newport:
Cal Newport is a computer science professor at Georgetown University. In addition to his academic research, he writes about the intersection of digital technology and culture. Cal's particularly interested in our struggle to deploy these tools in ways that support instead of subvert the things we care about in both our personal and professional lives.

Cal is a New York Times bestselling author of seven books, including, most recently, A World Without Email, Digital Minimalism, and Deep Work. He's also the creator of The Time-Block Planner.

The videos are considered to be used under the "Fair Use Doctrine" of United States Copyright Law, Title 17 U.S. Code Sections 107-118. Videos are used for editorial and educational purposes only and I do not claim ownership of any original video content. I don't use said video clips in advertisements, marketing or for direct financial gain. All video content in each clip is considered owned by the individual broadcast companies.

#CalNewport #DeepWork #DeepLife #DeepQuestions #TimeblockPlanner
#WorldWithoutEmail #DeepQuestionsPodcast