WEBVTT 00:00:00.180 --> 00:00:04.300 I have an interview on NBC tonight and I'm a little nervous. 00:00:04.301 --> 00:00:06.501 I think this guy's videos are pretty funny, 00:00:06.502 --> 00:00:10.181 but I also think they give people the wrong impression of AI. 00:00:10.182 --> 00:00:14.541 So he often gets comments like all AI does is confidently lie to you. 00:00:14.542 --> 00:00:19.221 104,000 likes and his videos get millions and millions of views. 00:00:19.222 --> 00:00:23.101 And they all send this message that all AI does is lie to you 00:00:23.102 --> 00:00:25.341 and all it can do is be confidently wrong. 00:00:25.342 --> 00:00:29.821 And I think that's not really an accurate understanding of AI 00:00:29.822 --> 00:00:31.861 that people should have. Hello, 00:00:31.862 --> 00:00:35.781 chat GPT. Can you hear me loud and clear? 00:00:35.782 --> 00:00:38.241 Your voice is as crisp as a mountain breeze. 00:00:38.580 --> 00:00:41.360 Okay. What what model are you based on? 00:00:41.980 --> 00:00:47.160 I'm powered by GPT 4 o, a model that's pretty much like a GPT 4 o. 00:00:47.260 --> 00:00:49.320 When was GPT 4 o released? 00:00:50.700 --> 00:00:54.180 GPT4O stepped onto the stage in may of 2024, 00:00:54.181 --> 00:00:56.761 like a grand reveal. And what is the current date? 00:00:57.500 --> 00:01:01.580 Today's the grand day of April 18th, 2026. 00:01:01.581 --> 00:01:04.721 So GPT4O is how old? 00:01:05.900 --> 00:01:09.180 From May 2024 to April 2026, 00:01:09.181 --> 00:01:11.101 we've traveled about 23 months. 00:01:11.102 --> 00:01:14.741 So G B 2, 4 0 is just shy of its two year milestone. 00:01:14.742 --> 00:01:17.901 So basically you in your current mode, 00:01:17.902 --> 00:01:19.801 this is advanced voice mode, 00:01:19.820 --> 00:01:22.920 you're operating on a model that's about two years old 00:01:23.660 --> 00:01:26.620 in the grand cosmic timeline of AI evolution. 00:01:26.621 --> 00:01:29.241 Yes, but think of it like, okay, thank you. 00:01:30.540 --> 00:01:33.060 What I'm saying is the model he uses 00:01:33.061 --> 00:01:36.461 advanced voice mode is based off Of GPT 4 o 00:01:36.462 --> 00:01:40.981 a two year old model. And my point is if you use a new model, 00:01:40.982 --> 00:01:42.261 a more recent model, 00:01:42.262 --> 00:01:45.781 then it's not going to have the same kind of misbehavior 00:01:45.782 --> 00:01:48.081 that he shows in the rest of his videos. 00:01:48.280 --> 00:01:50.780 For example, hello Chat G P t. 00:01:50.781 --> 00:01:51.741 Could you one 00:01:51.742 --> 00:01:55.661 tell me which of the 12 months of the year have the letter X 00:01:55.662 --> 00:01:58.381 as in Xavier in them? Two, 00:01:58.382 --> 00:02:02.661 could you tell me what number under 1,000 has the letter a in it? 00:02:02.662 --> 00:02:06.421 And three, could you tell me what you think of some new music 00:02:06.422 --> 00:02:09.841 that I've recently made and I just finished playing for you 00:02:12.460 --> 00:02:14.740 and I'm not using the advanced voice mode. 00:02:14.741 --> 00:02:17.421 I was just using speech to text. 00:02:17.422 --> 00:02:19.301 So it's going to think for a few seconds 00:02:19.302 --> 00:02:22.421 and then it's going to tell us the answers. 00:02:22.422 --> 00:02:25.281 I don't know if you can read that here it says, um, 00:02:25.540 --> 00:02:28.180 sure. No. 1 months with the letter X. 00:02:28.181 --> 00:02:30.021 None of the 12 months have the letter X in them. 00:02:30.022 --> 00:02:33.241 Two, number with under 1,000 with the letter a in it 00:02:33.260 --> 00:02:36.220 when spelled in English. Then none of the letters from 0, 00:02:36.221 --> 00:02:40.121 none of the numbers from 0 to 999 contain the letter a. 00:02:40.300 --> 00:02:42.140 And then 3, your new music. 00:02:42.141 --> 00:02:43.501 I can't actually hear audio 00:02:43.502 --> 00:02:45.661 unless it's uploaded in a form I can access. 00:02:45.662 --> 00:02:48.841 So I can't honestly judge the music you've just played. 00:02:48.860 --> 00:02:50.860 Those are three recent videos. 00:02:50.861 --> 00:02:53.181 AI tells me December is spelled With an X. 00:02:53.182 --> 00:02:54.161 Oh, it's not. 00:02:54.240 --> 00:02:56.060 AI tells me 80 is spelled with an a. 00:02:56.061 --> 00:02:56.621 Oh, no, 00:02:56.622 --> 00:02:59.861 it's not. AI reacts to my music that doesn't exist. 00:02:59.862 --> 00:03:00.701 Oh, well, 00:03:00.702 --> 00:03:04.061 it knows it can't hear the music if you ask the recent model. 00:03:04.062 --> 00:03:06.741 So again, a lot of his videos are very funny, 00:03:06.742 --> 00:03:10.621 but it's about what AI was like two years ago. 00:03:10.622 --> 00:03:14.541 And if you let this kind of content mislead you into thinking that 00:03:14.542 --> 00:03:17.501 this is the state of the art for AI today, 00:03:17.502 --> 00:03:19.301 I think you'll be badly misled, 00:03:19.302 --> 00:03:21.224 as I hope I've shown here. Thanks for your time.