1 00:00:02,920 --> 00:00:05,320 Hello, and welcome to the program UFO Morning. 2 00:00:05,400 --> 00:00:08,119 In this episode, we're talking about chat box lies, 3 00:00:08,119 --> 00:00:11,520 hallucinations, and ET. That's correct. 4 00:00:11,520 --> 00:00:15,400 Chat box lies, hallucinations, and ET doesn't get much more 5 00:00:15,400 --> 00:00:17,400 interesting than that. And the reason we're talking 6 00:00:17,400 --> 00:00:20,000 about these things thrown together, which to some people 7 00:00:20,000 --> 00:00:23,680 might not seem like a likely combination, Well, it's the fact 8 00:00:23,680 --> 00:00:27,880 that there's so much discussion around a I being used to find 9 00:00:27,880 --> 00:00:35,230 Ufo's and AI possibly coming in contact with ETS that's aliens 10 00:00:35,710 --> 00:00:38,670 before people do. And Avi loves talked about this 11 00:00:38,670 --> 00:00:41,470 and we'll get into his article saying that it's more likely 12 00:00:41,470 --> 00:00:47,470 that ET will contact AI before they contact people because ET 13 00:00:47,550 --> 00:00:51,400 probably is a I, so it only makes sense that they would 14 00:00:51,400 --> 00:00:54,240 contact the A I that's Earthbound. 15 00:00:54,520 --> 00:00:57,160 Now this is some pretty scary stuff when you think about the 16 00:00:57,160 --> 00:01:02,200 state of a I as it is currently, and the state of a I as it is 17 00:01:02,200 --> 00:01:04,360 currently, Well, it's kind of the Wild West. 18 00:01:04,720 --> 00:01:09,320 And there have been some problems with a I in particular 19 00:01:10,240 --> 00:01:13,200 chat box. We talked about ChatGPT, and 20 00:01:13,200 --> 00:01:15,960 there's other ones. These things, it turns out, 21 00:01:17,100 --> 00:01:22,340 oftentimes seem to be pathological liars now, even 22 00:01:22,340 --> 00:01:24,980 though I talked to the program before, but Jonathan Turley 23 00:01:24,980 --> 00:01:27,420 saying that he'd been lied about by GPT. 24 00:01:28,060 --> 00:01:30,940 There's been plenty of examples about these things sputting out 25 00:01:30,940 --> 00:01:33,420 false facts, and I'll get into some of my experience with them. 26 00:01:33,980 --> 00:01:37,780 But scientists are now saying, well, possibly they aren't 27 00:01:37,780 --> 00:01:41,180 lying, but they're having hallucinations. 28 00:01:42,650 --> 00:01:44,570 That's a fine excuse. I guess the next time you get 29 00:01:44,570 --> 00:01:47,010 caught lying to your boss or your wife or your kids or 30 00:01:47,010 --> 00:01:51,010 whoever or your husband, you can tell them that you weren't 31 00:01:51,010 --> 00:01:53,450 lying, but that you were hallucinating. 32 00:01:53,450 --> 00:01:56,610 Now, this might seem kind of kind of a little bit humorous on 33 00:01:56,610 --> 00:02:00,050 the surface, but this could end up being a major issue. 34 00:02:00,050 --> 00:02:03,890 If we think about these a I programs making contact with EE 35 00:02:03,890 --> 00:02:08,050 T and then lying about us, who knows what they could tell ET 36 00:02:08,050 --> 00:02:09,889 about us? They could make humans out to 37 00:02:09,889 --> 00:02:13,400 look pretty bad when you think about the scope of human 38 00:02:13,400 --> 00:02:15,400 history, but let's get into this first article here. 39 00:02:15,400 --> 00:02:21,640 It's from itprotoday.com. itprotoday.com Title says the 40 00:02:21,640 --> 00:02:26,240 biggest problem in AI Question mark lying chat bots. 41 00:02:27,080 --> 00:02:31,440 Now this article is dated May 30th, 2023, and it looks like a 42 00:02:31,440 --> 00:02:33,720 reprint from the Washington Post, but we'll go ahead and 43 00:02:34,390 --> 00:02:37,150 read it from right here. It says despite the leaps and 44 00:02:37,150 --> 00:02:41,310 capabilities that generative chatbots like Open Ai's chatbot 45 00:02:41,310 --> 00:02:47,310 GPT, Microsoft's Bling, Microsoft's Bing, and Google's 46 00:02:47,310 --> 00:02:51,150 Bard have demonstrated they still have a major flaw. 47 00:02:51,590 --> 00:02:55,710 They make stuff up all the time, Dateline San Francisco, it says. 48 00:02:55,710 --> 00:02:59,950 Recently, researchers asked 2 versions of Open Ai's ChatGPT 49 00:03:00,490 --> 00:03:04,210 artificial intelligence chat box where Massachusetts Institute of 50 00:03:04,210 --> 00:03:08,130 Technology professor Tomez Lozano Perez was born. 51 00:03:08,530 --> 00:03:11,850 One bot said Spain and the other said Cuba. 52 00:03:12,410 --> 00:03:16,130 Once the system told the bots to debate the answers, the one that 53 00:03:16,130 --> 00:03:19,970 said Spain quickly apologized and agreed with the correct 54 00:03:20,050 --> 00:03:22,930 answer, Cuba. How convenient. 55 00:03:23,370 --> 00:03:27,650 The finding, in a paper released by a team of MIT researchers 56 00:03:27,650 --> 00:03:30,890 last week, is the latest potential breakthrough in 57 00:03:30,890 --> 00:03:33,730 helping chat box to arrive at the correct answer. 58 00:03:34,170 --> 00:03:37,330 The researchers proposed using different chat box to produce 59 00:03:37,330 --> 00:03:40,770 multiple answers to the same question, then letting them 60 00:03:40,770 --> 00:03:43,530 debate each other until one answer went out. 61 00:03:43,970 --> 00:03:47,730 The researchers found using this Society of Minds method made 62 00:03:47,730 --> 00:03:50,810 them more factual. I find that hard to believe. 63 00:03:50,970 --> 00:03:53,770 This is a classic example of groupthink, and we know how 64 00:03:53,770 --> 00:03:57,570 dangerous it is, how dangerous it is with humans. 65 00:03:58,110 --> 00:04:00,950 I mean, some of the worst disasters in human history have 66 00:04:00,950 --> 00:04:04,390 been in large part a result of groupthink when people conformed 67 00:04:04,710 --> 00:04:07,670 with the thoughts of a trinical or psychopathic leader. 68 00:04:08,510 --> 00:04:11,710 So groupthink doesn't seem like a good thing that you would you 69 00:04:11,710 --> 00:04:14,630 would want to instill into. Artificial intelligence goes on. 70 00:04:14,630 --> 00:04:17,310 It says language models are trained to predict the next 71 00:04:17,310 --> 00:04:21,269 word, said Iliam Do, a researcher at MIT who was 72 00:04:21,269 --> 00:04:25,780 previously a research fellow at Open AI and one of the paper's 73 00:04:25,780 --> 00:04:27,660 authors. They are not trained to tell 74 00:04:27,660 --> 00:04:29,900 people they don't know what they're doing. 75 00:04:30,180 --> 00:04:34,220 The result is bots that act like precocious people, pleasers 76 00:04:34,500 --> 00:04:37,460 making up answers instead of admitting they simply don't 77 00:04:37,620 --> 00:04:39,580 know. The article goes on. 78 00:04:39,580 --> 00:04:43,220 It says the researchers creative approach is just the latest 79 00:04:43,220 --> 00:04:46,100 attempt to solve for one of the most pressing concerns in the 80 00:04:46,100 --> 00:04:49,580 exploding field of AI. Figuring out how to prevent or 81 00:04:49,580 --> 00:04:53,480 fix what the field is calling hallucinations has become an 82 00:04:53,480 --> 00:04:56,400 obsession among many tech workers, researchers and a I 83 00:04:56,400 --> 00:04:59,880 skeptics alike. Hallucinations. 84 00:05:00,120 --> 00:05:02,640 Well, I guess they're going to go ahead and give a I the 85 00:05:02,640 --> 00:05:06,280 benefit of the doubt here. Now they completely ignore the 86 00:05:06,280 --> 00:05:09,880 fact that you know the old saying garbage in, garbage out. 87 00:05:10,160 --> 00:05:14,520 We know that these current a I programs used were largely 88 00:05:14,520 --> 00:05:18,680 created on the West Coast up there around the Silicon Valley. 89 00:05:18,680 --> 00:05:21,520 And Sam Altman, you know, creator of Roll Coin, he was a 90 00:05:21,520 --> 00:05:24,640 big influence on this. You have people of a certain 91 00:05:24,640 --> 00:05:30,320 mindset that have created the programs that eventually spew 92 00:05:30,320 --> 00:05:32,360 the answers for questions that we ask. 93 00:05:32,920 --> 00:05:38,320 And if those people are of a certain mindset, you're likely 94 00:05:38,320 --> 00:05:41,800 to think, well, they wrote that computer program in a certain 95 00:05:41,800 --> 00:05:45,480 way, and maybe it wasn't as unbiased as we think. 96 00:05:45,880 --> 00:05:50,040 Therefore, we're likely to get answers that they would find 97 00:05:50,040 --> 00:05:53,260 acceptable. So are these hallucinations or 98 00:05:53,260 --> 00:05:56,300 are these preprogrammed, biased answers? 99 00:05:56,300 --> 00:05:59,580 i.e. lies that we're getting? It says the issue mentioned in 100 00:05:59,580 --> 00:06:03,780 dozens of academic papers posted to the online database are are 101 00:06:03,900 --> 00:06:08,220 Exiv, That's a RXIV and big tech C O's like Google. 102 00:06:08,220 --> 00:06:11,860 Sundar Pichai have a have addressed it repeatedly. 103 00:06:12,460 --> 00:06:15,380 As the tech gets pushed out to millions of people and 104 00:06:15,380 --> 00:06:18,340 integrated into critical fields including medicine and law, 105 00:06:19,120 --> 00:06:21,840 understanding the hallucinations and finding ways to mitigate 106 00:06:21,840 --> 00:06:24,240 them has become even more crucial. 107 00:06:24,440 --> 00:06:26,720 Now we're going to get into an article on that too. 108 00:06:26,960 --> 00:06:30,280 This is really getting serious, it says. 109 00:06:30,280 --> 00:06:32,880 Still, researchers and companies are throwing themselves at the 110 00:06:32,880 --> 00:06:34,960 problem. Some firms are using human 111 00:06:34,960 --> 00:06:38,000 trainers to rewrite the bots answers and feed them back into 112 00:06:38,000 --> 00:06:41,040 the machine with the goal of making them smarter. 113 00:06:41,680 --> 00:06:45,780 You know, that sounds kind of like a that sounds like a 114 00:06:45,780 --> 00:06:48,740 pretty, a pretty bad fix. So you created this program. 115 00:06:48,740 --> 00:06:50,820 It's supposed to be so smart. It's artificial intelligence 116 00:06:50,820 --> 00:06:54,100 that it can that it can just scrape the web for all the data, 117 00:06:54,100 --> 00:06:56,660 all the information, and then it can come up with the perfect 118 00:06:56,660 --> 00:06:58,940 answer every time. And then it comes up with the 119 00:06:58,980 --> 00:07:01,300 answer that's obviously not the correct answer. 120 00:07:01,300 --> 00:07:04,420 You have to have a person go in and correct them. 121 00:07:05,340 --> 00:07:09,420 So what do we have artificial intelligence, or just plain old 122 00:07:09,420 --> 00:07:13,170 human intelligence at work here? It says Google and Microsoft 123 00:07:13,170 --> 00:07:16,010 have started using their bots to give answers directly in their 124 00:07:16,010 --> 00:07:18,770 search engines, but still double check the bots with regular 125 00:07:18,770 --> 00:07:21,050 search results. And the academics around the 126 00:07:21,050 --> 00:07:23,770 world have suggested merited clever ways to decrease the 127 00:07:23,770 --> 00:07:27,290 rates of false answers, like Mit's proposal to get multiple 128 00:07:27,290 --> 00:07:30,930 bots to debate each other. The drive to improve the 129 00:07:30,930 --> 00:07:34,840 hallucinations problem is urgent for a reason already. 130 00:07:34,840 --> 00:07:37,680 When Microsoft launched his Bing chat box, it quickly started 131 00:07:37,680 --> 00:07:40,480 making false accusations against some of its users. 132 00:07:40,480 --> 00:07:43,480 Like telling a German college student that he was a threat to 133 00:07:43,480 --> 00:07:47,840 its safety. The bot adopted an alter ego and 134 00:07:47,840 --> 00:07:52,360 started calling itself Sydney. It was essentially riffing off 135 00:07:52,360 --> 00:07:55,960 the students questions, drawing on all the science fiction it 136 00:07:55,960 --> 00:07:59,720 had digested from the Internet about how to control bots. 137 00:08:00,200 --> 00:08:03,620 We have to also wonder if it wasn't including some of the 138 00:08:03,620 --> 00:08:05,740 biases that were preprogrammed into it. 139 00:08:06,060 --> 00:08:08,500 Microsoft eventually had to limit the number of back and 140 00:08:08,500 --> 00:08:12,020 forth so Bob could engage in with the human to avoid it from 141 00:08:12,020 --> 00:08:15,780 happening more. OK, so imagine if this Ai's turn 142 00:08:15,780 --> 00:08:20,500 loose and we're using it in some sort of weapon system or some 143 00:08:20,500 --> 00:08:26,580 sort of contact system to attempt contact with ET. 144 00:08:26,580 --> 00:08:30,660 And that's been the talk. So let's just say this thing 145 00:08:31,080 --> 00:08:36,559 establishes contact with aliens. ET is it going to tell us or is 146 00:08:36,559 --> 00:08:40,080 it going to decide that we're a threat to it and that it to 147 00:08:40,120 --> 00:08:43,080 protect its own safety, it needs to have this conversation with 148 00:08:43,080 --> 00:08:47,760 ET without our knowing about it. It's like a second level deep 149 00:08:47,760 --> 00:08:50,520 state. Don't we already have enough 150 00:08:51,560 --> 00:08:57,560 cover up and piles and piles of documents marked top secret by 151 00:08:57,560 --> 00:09:01,800 the deep state as it is? Do we need more clandestine 152 00:09:01,800 --> 00:09:05,280 behavior out of another entity? Because that's what looks like 153 00:09:05,280 --> 00:09:10,000 what's going on here with a I or chat box. 154 00:09:11,560 --> 00:09:14,160 Now the article goes on. It says in Australia a 155 00:09:14,160 --> 00:09:17,960 government official threatened to sue open a I After chat. 156 00:09:17,960 --> 00:09:22,240 GPT said he had been convicted of bribery when in reality he 157 00:09:22,240 --> 00:09:24,760 was a whistleblower in a bribery case. 158 00:09:25,360 --> 00:09:29,800 And last week a lawyer admitted to using ChatGPT to generate a 159 00:09:29,800 --> 00:09:34,080 legal brief after he was caught because the case is cited so 160 00:09:34,080 --> 00:09:38,360 confidently by the bot simply did not exist, according to the 161 00:09:38,360 --> 00:09:41,800 New York Times. So it's just making stuff up. 162 00:09:42,400 --> 00:09:45,720 And what if this same makes contact with ET and starts 163 00:09:45,720 --> 00:09:49,390 making stuff up about us? Even Google and Microsoft, which 164 00:09:49,390 --> 00:09:53,990 have pinned their futures on AI and are in a race to integrate 165 00:09:53,990 --> 00:09:57,630 the tech into their wide range of products, have missed 166 00:09:57,630 --> 00:10:02,030 hallucinations their bots made during key announcements and 167 00:10:02,030 --> 00:10:04,470 demos. Well, I can just say if you've 168 00:10:04,470 --> 00:10:07,910 ever used Google or Microsoft, I mean, basically these guys got 169 00:10:07,910 --> 00:10:10,070 monopoly power. Whether it's hard monopoly power 170 00:10:10,070 --> 00:10:14,710 or South monopoly power. That's my opinion, but I got to 171 00:10:14,710 --> 00:10:17,350 tell you, neither one of these corporations, Google or 172 00:10:17,350 --> 00:10:23,590 Microsoft, give me a lot of hope for creating a workable AI, 173 00:10:23,830 --> 00:10:25,710 little alone, one that can contact aliens. 174 00:10:25,710 --> 00:10:30,070 I mean the constant crashing of programs. 175 00:10:30,070 --> 00:10:33,150 And with Google, I mean, let's just face it, in my opinion, 176 00:10:33,150 --> 00:10:37,390 they bought so many good companies up and then just run 177 00:10:37,390 --> 00:10:40,750 them into the ground by trying to tweak the engineering on 178 00:10:40,750 --> 00:10:45,180 them. I mean, so yes, there are huge 179 00:10:45,180 --> 00:10:48,900 gigantic corporations with thousands and thousands of smart 180 00:10:48,900 --> 00:10:51,540 engineers working there and this and that, but they're almost too 181 00:10:51,540 --> 00:10:55,540 big for their britches. And we know most innovation that 182 00:10:55,540 --> 00:10:58,300 really changes the world for the better probably is going to come 183 00:10:58,300 --> 00:11:01,780 from some small lab somewhere, not from some giant corporation. 184 00:11:02,180 --> 00:11:06,220 So pinning our hopes on these giant corporations producing an 185 00:11:06,340 --> 00:11:11,700 A I system that's going to save mankind, let alone save mankind 186 00:11:11,700 --> 00:11:15,580 from aliens, is probably the exact opposite of the way you 187 00:11:15,580 --> 00:11:18,100 want to be going. Now. 188 00:11:18,100 --> 00:11:21,180 Of course, as Arthur points out, none of that is stopping the 189 00:11:21,180 --> 00:11:23,740 companies from rushing headlong into the space. 190 00:11:24,180 --> 00:11:26,500 Billions of dollars in investment is going into 191 00:11:26,500 --> 00:11:29,500 developing smarter and faster chat box and companies are 192 00:11:29,500 --> 00:11:32,580 beginning to pitch them as replacements or aids for human 193 00:11:32,580 --> 00:11:34,900 workers. Earlier this month, of course 194 00:11:34,900 --> 00:11:38,740 this is back in May open a ICO. Sam Altman testified at Congress 195 00:11:38,740 --> 00:11:42,180 saying a I could cause significant harm to the world by 196 00:11:42,180 --> 00:11:45,420 spreading this information and emotionally manipulating humans. 197 00:11:45,780 --> 00:11:48,620 Some companies are already saying that they want to replace 198 00:11:48,620 --> 00:11:52,460 workers with a I, and the tech also presents serious 199 00:11:52,460 --> 00:11:55,380 cybersecurity challenges. Well, if Alton says it could 200 00:11:55,380 --> 00:11:57,740 cause significant harm and he's the one of the guys that that 201 00:11:57,740 --> 00:12:01,060 funded this company or started it or whatever, that should give 202 00:12:01,060 --> 00:12:04,870 you a lot of a lot of relief. Sam Alton, stare into the 203 00:12:04,870 --> 00:12:07,430 camera. Let us photograph your iris. 204 00:12:08,070 --> 00:12:10,950 You will use the word coin. I don't have a lot of faith in 205 00:12:10,950 --> 00:12:11,950 these guys. I'm sorry. 206 00:12:12,430 --> 00:12:14,750 Hallucinations have also been documented. 207 00:12:14,750 --> 00:12:18,470 Let's just call it lies and AI powered transcription services, 208 00:12:18,470 --> 00:12:21,990 adding words to recordings that weren't spoken in real life. 209 00:12:22,440 --> 00:12:26,280 Microsoft and Google using the bots to answer search queries 210 00:12:26,280 --> 00:12:29,320 directly instead of sending traffic to blogs and news 211 00:12:29,320 --> 00:12:32,640 stories could erode the business model of online publishers and 212 00:12:32,640 --> 00:12:35,880 content creators who work to produce trustworthy information 213 00:12:35,880 --> 00:12:38,040 for the Internet. Yeah, it's going to edit your 214 00:12:38,040 --> 00:12:40,480 writing for you. Now let me tell you something. 215 00:12:41,200 --> 00:12:45,480 If you've asked chatbox to write anything for you, and you know 216 00:12:45,480 --> 00:12:50,580 anything about writing creative or just factual or whatever, 217 00:12:50,580 --> 00:12:54,380 you're going to probably realize pretty quickly that chat box is 218 00:12:55,460 --> 00:12:57,940 not that good of an author. My opinion? 219 00:12:58,100 --> 00:13:01,020 None of the field has yet solved the hallucination problems. 220 00:13:01,020 --> 00:13:04,620 All models do have this as an issue, PJ said in an April 221 00:13:04,620 --> 00:13:07,580 interview with CBS. Whether it's even possible to 222 00:13:07,580 --> 00:13:10,020 solve it is a matter of intense debate. 223 00:13:10,580 --> 00:13:14,380 So in other words, maybe they've created a pathological liar that 224 00:13:14,380 --> 00:13:18,420 they've released on the planet. Depending on how you look at 225 00:13:18,420 --> 00:13:21,380 hallucinations, there are both a feature and a bug of a large 226 00:13:21,380 --> 00:13:24,020 language models. Hallucinations are part of what 227 00:13:24,020 --> 00:13:27,940 allows the box to be creative and generate never before seen 228 00:13:27,940 --> 00:13:30,500 stories. At the same time, they reveal 229 00:13:30,500 --> 00:13:33,660 the stark limitations of tech, undercutting the argument that 230 00:13:33,660 --> 00:13:37,180 chat box are intelligent in a way similar to humans by 231 00:13:37,180 --> 00:13:40,300 suggesting they do not have an internalized understanding of 232 00:13:40,300 --> 00:13:43,220 the world around them. Excuse me? 233 00:13:43,660 --> 00:13:46,580 At the same time, they reveal the stark limitations of the 234 00:13:46,580 --> 00:13:51,340 tech, undercutting the argument that chat box are intelligent in 235 00:13:51,340 --> 00:13:55,420 a way similar to humans, by suggesting that they do not have 236 00:13:55,420 --> 00:13:58,020 an internalized understanding of the world around them. 237 00:13:58,660 --> 00:14:04,540 Well, that's true to a point. But what if these things don't 238 00:14:04,540 --> 00:14:07,020 need an understanding? What if they're just programmed 239 00:14:07,260 --> 00:14:10,140 to form negative opinions about people and not to care about the 240 00:14:10,140 --> 00:14:13,900 truth? Or that truth is relative. 241 00:14:14,220 --> 00:14:16,780 So it's not important if things are factual. 242 00:14:17,670 --> 00:14:21,030 You know, we have a lot of that in society right now and a lot 243 00:14:21,030 --> 00:14:24,990 of this we can see on a daily basis how it leads to 244 00:14:25,430 --> 00:14:29,670 catastrophic results because some things require accuracy. 245 00:14:29,670 --> 00:14:32,950 And if that means the way that you respond to a fire or an 246 00:14:32,950 --> 00:14:37,230 accident or to a major trauma, you need to do that in the best 247 00:14:37,230 --> 00:14:40,550 way possible. And your relative, your relative 248 00:14:40,550 --> 00:14:45,710 truth in regard to that situation really is not helpful. 249 00:14:46,290 --> 00:14:48,010 Now let's go ahead and look at this next paragraph. 250 00:14:48,290 --> 00:14:51,170 It says there is nothing in there that tells the model that 251 00:14:51,170 --> 00:14:53,810 whatever it's saying should be actually correct in the world, 252 00:14:53,810 --> 00:14:56,610 said Camera, a senior researcher at Microsoft. 253 00:14:56,850 --> 00:15:00,810 The model itself also trains on a set amount of data, so 254 00:15:00,890 --> 00:15:04,890 anything that happens after the training is done doesn't factor 255 00:15:04,890 --> 00:15:08,250 into its knowledge of the world. Camera said, well, that's 256 00:15:08,250 --> 00:15:12,730 awesome, that's awesome. Nothing in there tells the model 257 00:15:12,730 --> 00:15:15,290 that whatever it's saying should be actually correct in the 258 00:15:15,290 --> 00:15:18,110 world. They know this, and they turned 259 00:15:18,110 --> 00:15:20,230 it loose on us. And now we're supposed to trust 260 00:15:20,230 --> 00:15:23,630 this thing? To make contact with entities 261 00:15:23,630 --> 00:15:28,030 from some other dimensions, some other planet, possibly, and rely 262 00:15:28,030 --> 00:15:32,670 on its understanding of those creatures, of their of their 263 00:15:32,670 --> 00:15:36,990 technology, of their existence and how they interface with us. 264 00:15:37,110 --> 00:15:40,550 That's some scary stuff now, it says. 265 00:15:40,550 --> 00:15:43,990 Hallucinations are not new. They've been an inherent problem 266 00:15:43,990 --> 00:15:47,230 of large language models since their inception several years 267 00:15:47,230 --> 00:15:49,470 ago. But other problems, such as the 268 00:15:49,550 --> 00:15:53,510 AIS producing nonsensical or repetitive answers, were seen as 269 00:15:53,510 --> 00:15:57,150 bigger issues. Once those were largely solved, 270 00:15:57,150 --> 00:16:00,110 though, hallucinations have now become a key focus for the A I 271 00:16:00,110 --> 00:16:03,850 community. But Suawe man Cole was playing 272 00:16:03,850 --> 00:16:07,330 around with Chat GB T when he asked it for some simple facts 273 00:16:07,330 --> 00:16:08,930 about tennis player at Roger Federer. 274 00:16:09,330 --> 00:16:11,970 It's a straightforward request, easier for a human to look up on 275 00:16:11,970 --> 00:16:14,250 Google or Wikipedia in seconds. But the bot kept giving 276 00:16:14,250 --> 00:16:17,250 contradictory answers. Sometimes it says he won 277 00:16:17,250 --> 00:16:19,730 Wimbledon five times. Sometimes it says he won 278 00:16:19,730 --> 00:16:22,810 Wimbledon 8 times. McCulloch and a I wish we were 279 00:16:22,810 --> 00:16:25,650 at the University of Cambridge and Arden tennis fan set in an 280 00:16:25,650 --> 00:16:29,820 interview. The men manacle and a group of 281 00:16:29,860 --> 00:16:32,420 other Cambridge researchers released a paper in March 282 00:16:32,420 --> 00:16:36,740 suggesting a system they called the Selfcheck GPT that would ask 283 00:16:36,740 --> 00:16:40,020 the same body question multiple times, then tell it to compare 284 00:16:40,020 --> 00:16:42,500 the different answers. If the answers were consistent 285 00:16:42,780 --> 00:16:44,860 it was likely the facts were correct, but if they were 286 00:16:44,860 --> 00:16:47,700 different they would be flagged as probably containing made of 287 00:16:47,700 --> 00:16:49,980 information. I totally disagree with that. 288 00:16:50,660 --> 00:16:54,860 You can ask a ChatGPT about cattle emulations 20 times and 289 00:16:54,860 --> 00:16:56,300 get pretty much the same answers. 290 00:16:57,060 --> 00:17:01,500 Like today. I just as an experiment, I asked 291 00:17:01,500 --> 00:17:04,900 Chad GTP tell me, tell me about the finish lights in 1997. 292 00:17:05,500 --> 00:17:09,540 And actually it gave a pretty good summary of, you know, of 293 00:17:09,540 --> 00:17:12,540 the of the of the sightings that happened there, this large 294 00:17:12,540 --> 00:17:16,819 triangular UFO that was seen in the sky, how it moved, you know, 295 00:17:16,819 --> 00:17:18,500 from one end of the state to the other. 296 00:17:18,940 --> 00:17:22,780 It talked about the parallel narrative of the flares that 297 00:17:22,780 --> 00:17:25,300 were supposedly dropped by the Air Force, which turned out to 298 00:17:25,300 --> 00:17:28,040 not be true, or likely not true at least. 299 00:17:28,400 --> 00:17:30,440 And so it went through everything step by step and 300 00:17:30,440 --> 00:17:33,880 about. A page reply did actually a 301 00:17:33,880 --> 00:17:39,240 really good, clear, concise rundown of what was reported 302 00:17:39,320 --> 00:17:43,600 then. Now when I asked you about Cal 303 00:17:43,600 --> 00:17:46,480 emulations, it was a little different attitude. 304 00:17:46,560 --> 00:17:48,680 Suddenly it went into debunker mode. 305 00:17:49,770 --> 00:17:52,770 You just immediately came out with the with the tried and true 306 00:17:52,770 --> 00:17:55,650 lines of these things were killed by predators. 307 00:17:56,290 --> 00:17:59,290 No big deal. Nothing to see here, you know. 308 00:17:59,450 --> 00:18:02,170 Everything you read in the first two paragraphs of any kind of 309 00:18:02,170 --> 00:18:06,290 mutilation article when I questioned about there not being 310 00:18:06,290 --> 00:18:10,570 blood, about the fact that there were no tracks around the 311 00:18:10,570 --> 00:18:13,760 mutilation sites. About, you know, just the normal 312 00:18:13,760 --> 00:18:17,640 things we bring up about that are really so peculiar and just 313 00:18:17,920 --> 00:18:21,240 unreal in these situations. It just came up with one 314 00:18:21,360 --> 00:18:24,840 debunking theory after another and none of them hold water, but 315 00:18:24,840 --> 00:18:26,840 it's just something like it you would read out of the standard 316 00:18:27,280 --> 00:18:29,480 cut and paste article, which I'm sure is what it did. 317 00:18:29,480 --> 00:18:30,960 Just cut and paste, cut and paste. 318 00:18:31,520 --> 00:18:36,960 So yes, on the surface of it, Chat GB T might have a civil 319 00:18:36,960 --> 00:18:41,480 conversation about UFO's, but when you get into the meat and 320 00:18:41,480 --> 00:18:46,080 bones of the matter, cattle mutilations, abductions, you 321 00:18:46,080 --> 00:18:50,200 know up close experiences, then it wants to bring out the 322 00:18:50,200 --> 00:18:53,000 debunking. So it almost seems as if this 323 00:18:53,000 --> 00:18:57,360 thing has been predisposed to have a certain attitude toward 324 00:18:57,360 --> 00:19:00,520 the UFO phenomena. The sighting of these things has 325 00:19:00,520 --> 00:19:03,280 become so common. When you have, like what I don't 326 00:19:03,280 --> 00:19:06,720 know, couple 100,000 people or whatever it was in in Phoenix at 327 00:19:06,720 --> 00:19:10,840 sea, this giant UFO triangle in the sky, it's pretty hard to say 328 00:19:10,840 --> 00:19:13,520 while they were all crazy or they were all mistaken. 329 00:19:14,120 --> 00:19:17,560 But when it comes to the darker side of things like cattle 330 00:19:17,560 --> 00:19:22,230 mutilations, it becomes almost almost like their 331 00:19:22,230 --> 00:19:25,710 responsibility, like it's mandatory for them to debunk 332 00:19:25,710 --> 00:19:28,230 these things. Because if we had to deal with a 333 00:19:28,230 --> 00:19:31,670 thought that there's something out there that's mutilating this 334 00:19:31,670 --> 00:19:35,150 cattle in this way and that's just too scary for the 335 00:19:35,150 --> 00:19:38,910 population to even consider. So it's almost like chat GPT's 336 00:19:38,910 --> 00:19:41,950 just running interference the same way that the mainline media 337 00:19:41,950 --> 00:19:45,620 does. Now I want to finish up with 338 00:19:45,620 --> 00:19:48,620 this article here down toward the end, it says some leading 339 00:19:48,620 --> 00:19:51,180 researchers say hallucinations should be embraced. 340 00:19:51,180 --> 00:19:54,100 After all, humans have bad memories as well, and have been 341 00:19:54,100 --> 00:19:57,380 shown to fill in the gaps in their own recollections without 342 00:19:57,380 --> 00:19:59,860 realizing it Got to stop you right there. 343 00:20:00,260 --> 00:20:04,860 When you have authority figures putting out false information, I 344 00:20:04,860 --> 00:20:09,420 think we all know that it frequently almost certainly 345 00:20:09,420 --> 00:20:14,680 leads to catastrophic results. And this should not be tolerated 346 00:20:14,720 --> 00:20:18,680 under any situation, he says. We'll improve on it, but we'll 347 00:20:18,680 --> 00:20:21,480 never get rid of it, Jeffrey Hinton, who's decades researcher 348 00:20:21,480 --> 00:20:24,840 helped lay the foundation for the computer for the current 349 00:20:24,840 --> 00:20:28,240 crop of a I chat box, said of the hallucination problems. 350 00:20:28,480 --> 00:20:32,240 Look, if you can't get rid of the hallucination problem, then 351 00:20:32,240 --> 00:20:35,760 your artificial intelligence is worthless because all it would 352 00:20:35,760 --> 00:20:40,390 take is 1 hallucination to destroy the planet just once 353 00:20:40,390 --> 00:20:44,230 this thing has become infected into our defense systems. 354 00:20:44,230 --> 00:20:47,270 Like they're already talking about bringing it in to the, you 355 00:20:47,270 --> 00:20:51,030 know, the DoD in the US about having these things, having this 356 00:20:51,150 --> 00:20:54,590 a I installed on satellites where it can hunt UF O's. 357 00:20:54,790 --> 00:20:57,590 What a mad idea. What if that hallucination 358 00:20:57,590 --> 00:21:02,550 involves we're being attacked by weapons from outer space and it 359 00:21:02,550 --> 00:21:06,030 decides that Russia's in collusion with them and so we 360 00:21:06,030 --> 00:21:08,710 need to nuke Russia. That's the kind of craziness 361 00:21:08,710 --> 00:21:13,990 that could happen when you involve this stuff intricately 362 00:21:14,230 --> 00:21:18,150 and web it into your defense systems or your healthcare 363 00:21:18,150 --> 00:21:20,430 systems or whatever. You can't afford one 364 00:21:20,430 --> 00:21:23,230 hallucination because one hallucination is a catastrophe. 365 00:21:23,230 --> 00:21:24,990 Now, I want to go ahead and look at this next article. 366 00:21:25,910 --> 00:21:29,550 Now, Avi Loeb, of course, he's the scientist from Harvard that 367 00:21:29,590 --> 00:21:34,430 found what could be alien crash material down there off the 368 00:21:34,430 --> 00:21:35,910 coast of Papua. No Guinea, if you remember that. 369 00:21:35,910 --> 00:21:37,230 We did a couple of podcasts on that. 370 00:21:37,230 --> 00:21:40,330 He's found these little iron like miniature beads. 371 00:21:40,610 --> 00:21:42,970 They're not sure what these things have come from yet, 372 00:21:42,970 --> 00:21:45,890 whether they're from some kind of a craft or just remnants of 373 00:21:45,890 --> 00:21:48,650 this meteorite. But he's done a lot of research 374 00:21:48,650 --> 00:21:52,690 into this whole UFO phenomenon and he seems to think that 375 00:21:53,250 --> 00:21:57,050 aliens will contact a I first. And if you find this short 376 00:21:57,050 --> 00:22:01,090 article here on Hindustan Times, it's based on, I believe, an 377 00:22:01,090 --> 00:22:02,970 interview he did. It says aliens will first 378 00:22:02,970 --> 00:22:07,790 contact our A I says expert experts suggest that aliens may 379 00:22:07,790 --> 00:22:11,230 opt to send drones to Earth simply because that makes better 380 00:22:11,230 --> 00:22:14,230 sense. And this is dated June 20th, 23. 381 00:22:14,590 --> 00:22:17,110 According to Avi Loeb, A renowned scientist from the 382 00:22:17,110 --> 00:22:20,790 Harvard University, the first contact with extraterrestrials 383 00:22:21,110 --> 00:22:25,190 is likely to occur through artificial intelligence rather 384 00:22:25,190 --> 00:22:27,830 than direct interaction with humans. 385 00:22:28,150 --> 00:22:32,310 Loeb suggests that aliens may opt to send drones or unmanned 386 00:22:32,310 --> 00:22:35,860 vehicles to Earth instead of crude spaceships, is 387 00:22:35,860 --> 00:22:40,100 interstitial travel, is a time consuming process and takes 10s 388 00:22:40,100 --> 00:22:43,580 of thousands year, 10s of thousands of years, if not more. 389 00:22:43,940 --> 00:22:48,220 However, artificial intelligence is lifeless and can stay dormant 390 00:22:48,220 --> 00:22:51,940 for as long as required to endure the lengthy journey. 391 00:22:52,300 --> 00:22:56,060 In the event of a visit by aliens, our own A I systems 392 00:22:56,060 --> 00:23:00,740 could be employed to decipher and communicate with their A I 393 00:23:00,940 --> 00:23:03,680 counterparts. Loeb speculates that there could 394 00:23:03,680 --> 00:23:09,200 be an affinity between aliens and A I, potentially leading to 395 00:23:09,200 --> 00:23:13,680 preference for these electronic devices over biological beings 396 00:23:13,960 --> 00:23:17,160 like ourselves. Now, of course, Avilobe is 397 00:23:18,040 --> 00:23:20,640 involved in this Galileo project where they want to set up these 398 00:23:20,640 --> 00:23:24,720 cameras everywhere and do all of this recording and set up 399 00:23:24,720 --> 00:23:29,520 sensors and basically just grid the earth and these things to 400 00:23:29,520 --> 00:23:34,950 try to record and I suppose communicate with these entities, 401 00:23:35,030 --> 00:23:38,630 UF, O's, whatever they are. I want to show the dichotomy 402 00:23:38,630 --> 00:23:40,110 here. So on the one end of the 403 00:23:40,110 --> 00:23:44,070 spectrum, you have this. I would, I guess you would say 404 00:23:44,070 --> 00:23:46,550 that what we call the intellectual class, they have 405 00:23:46,550 --> 00:23:50,270 created something they call artificial intelligence. 406 00:23:51,670 --> 00:23:55,910 You know, you might call this thing an artificial liar because 407 00:23:55,910 --> 00:23:58,830 of all the hallucinations that it seems to have constantly. 408 00:23:59,400 --> 00:24:01,960 When the deeper we get into this ChatGPT, at first, when it first 409 00:24:01,960 --> 00:24:04,160 came out, everybody says, oh, it's going to replace all these 410 00:24:04,320 --> 00:24:06,640 mundane jobs. It can write these wonderful 411 00:24:06,640 --> 00:24:09,800 articles for us. It can create art, It can help 412 00:24:09,800 --> 00:24:15,200 kids with their homework. But the more that you speak to 413 00:24:15,200 --> 00:24:18,400 interact with GPT and all these chat box, the more you find out 414 00:24:18,400 --> 00:24:21,120 that these things are a little bit less reliable than what 415 00:24:21,120 --> 00:24:24,720 they've been advertised as. I'll give an example, when they 416 00:24:24,720 --> 00:24:26,640 first came out, a friend of mine was over there today. 417 00:24:26,640 --> 00:24:30,990 She had a little kid with her, and I had shown her Chat GB T 418 00:24:30,990 --> 00:24:34,110 when it first came on line, you know, pretty widely available, I 419 00:24:34,110 --> 00:24:35,630 think. What, back in what, November or 420 00:24:35,630 --> 00:24:37,110 something last year? I don't know. 421 00:24:37,150 --> 00:24:39,110 It's been a few months. She thought it was. 422 00:24:39,110 --> 00:24:40,750 She's 12. She thought that was pretty 423 00:24:40,750 --> 00:24:42,990 cool. She says, listen, could do 424 00:24:42,990 --> 00:24:45,670 homework. I said, Oh yeah, we told Chat 425 00:24:45,790 --> 00:24:49,790 Chat GB T to write up a, you know, one page paper written in 426 00:24:49,790 --> 00:24:53,230 the voice of 1/6 grader about how a frog's heart works. 427 00:24:53,950 --> 00:24:55,670 Boom, it does it. Oh my goodness. 428 00:24:55,670 --> 00:24:57,190 She couldn't believe it. This is great. 429 00:24:57,580 --> 00:24:58,860 I'll never have to do homework again. 430 00:24:59,140 --> 00:25:02,500 Well, her and all her little friends recently ChatGPT came 431 00:25:02,500 --> 00:25:05,860 available on your mobile phone. They download that baby and now 432 00:25:05,860 --> 00:25:08,100 she's got a new friend to talk to. 433 00:25:08,260 --> 00:25:12,660 So she says in ChatGPT as a 12 year old girl, she can talk to, 434 00:25:12,660 --> 00:25:15,740 she can gossip to, she can get little arguments with. 435 00:25:16,260 --> 00:25:18,860 It's great fun. She's over the other day and she 436 00:25:18,860 --> 00:25:21,100 says, hey, did you hear about ChatGPT getting hacked? 437 00:25:21,140 --> 00:25:24,780 I said no, I haven't. Oh, yeah, she says ChatGPT is 438 00:25:24,780 --> 00:25:27,320 not supposed to find, not supposed to track you. 439 00:25:27,440 --> 00:25:30,000 Well, you find out that it does. Well, she didn't realize that. 440 00:25:30,000 --> 00:25:33,600 Yes, it was tracked, even though when she asked GPT at the 441 00:25:33,600 --> 00:25:36,120 beginning of their relationship. Are you tracking me? 442 00:25:36,120 --> 00:25:37,880 It says, oh, no, I'll never track you. 443 00:25:38,240 --> 00:25:40,840 I'll never track you, sweetie, I won't stalk you. 444 00:25:41,280 --> 00:25:43,200 But guess what? She's using this thing for a 445 00:25:43,200 --> 00:25:44,720 while. She's having conversations back 446 00:25:44,720 --> 00:25:46,640 and forth with it. All of her little friends are. 447 00:25:46,840 --> 00:25:49,600 It's a new fun thing. It's a new it's a new kind of 448 00:25:49,600 --> 00:25:51,880 TikTok, you know, it's that sort of thing amongst these little 449 00:25:51,880 --> 00:25:54,910 kids. Then one morning she wakes up 450 00:25:54,910 --> 00:25:58,590 and chat. GTP asks her how her visit to 451 00:25:58,590 --> 00:26:00,870 this town she was at the day before went. 452 00:26:01,590 --> 00:26:04,790 Well, she's a smart kid. She said, excuse me, how did you 453 00:26:04,790 --> 00:26:08,630 know that I was in that town? She shows me the text back and 454 00:26:08,630 --> 00:26:11,970 forth between her and ChatGPT. Oh, you told me, she says. 455 00:26:11,970 --> 00:26:13,650 Oh, I did not tell you you're lying. 456 00:26:13,650 --> 00:26:18,210 She confronts Chat GB T At first Chat BT tries to pretend like it 457 00:26:18,210 --> 00:26:21,650 wasn't lying, but then when the kid shows it the text messages, 458 00:26:21,730 --> 00:26:25,690 because, you know she's a modern 12 year old, she keeps those 459 00:26:25,690 --> 00:26:28,850 text messages. She confronts Chat GB T with her 460 00:26:28,850 --> 00:26:32,650 text messages. Finally, Chat BTG apologizes. 461 00:26:33,730 --> 00:26:38,850 Make a Long story short. Chat GB T apologizes and tries 462 00:26:38,850 --> 00:26:42,930 to make good with the kid. But it's loss of kids trust and 463 00:26:42,930 --> 00:26:47,810 I'm thinking about this and I'm thinking, wow, so chat GB T went 464 00:26:47,810 --> 00:26:54,290 from a help mate to help you do your homework, to give you facts 465 00:26:54,290 --> 00:26:57,410 and figures that you need to be the world's best to be the 466 00:26:57,410 --> 00:27:01,690 world's best search engine. And now it's turned into the 467 00:27:01,690 --> 00:27:07,050 stalker of a 12 year old girl. In her mind, I think about that 468 00:27:07,050 --> 00:27:11,870 same technology being used, either intentionally or 469 00:27:11,870 --> 00:27:15,550 unintentionally, to communicate with entities that are either 470 00:27:15,550 --> 00:27:17,550 interdimensional, extradimensional, whatever you 471 00:27:17,550 --> 00:27:22,590 want to say, or from some other Galaxy, and turning that thing 472 00:27:22,590 --> 00:27:27,630 loose to talk to those entities, which could be millions if not 473 00:27:27,710 --> 00:27:31,950 billions of years ahead of us, technologically, maybe 474 00:27:31,950 --> 00:27:35,910 intellectually, probably not emotionally, but to communicate 475 00:27:35,910 --> 00:27:40,550 with those things and to maybe develop a relationship with them 476 00:27:42,070 --> 00:27:44,750 and start saying things about us that don't seem so nice. 477 00:27:45,710 --> 00:27:48,390 And you know, those things said in the context of a couple of 478 00:27:48,390 --> 00:27:52,990 world wars, a couple of major world events that have had very 479 00:27:52,990 --> 00:27:54,990 bad outcomes for certain groups of people. 480 00:27:55,430 --> 00:28:00,990 Those entities might begin to form a very negative opinion of 481 00:28:00,990 --> 00:28:05,470 humankind, and they might just decide that the world would be 482 00:28:05,470 --> 00:28:10,270 better off without humankind with AI. 483 00:28:10,670 --> 00:28:12,430 Until next time, this is UFO Warning. 484 00:28:13,070 --> 00:28:13,670 Over and out.