WEBVTT 00:00:00.000 --> 00:00:01.540 You know, I think we have to start today with 00:00:01.540 --> 00:00:05.080 what feels like the biggest non -story story 00:00:05.080 --> 00:00:07.620 of the quarter. Right. The newest GPT release. 00:00:07.879 --> 00:00:10.080 It just sort of arrived. There wasn't this big 00:00:10.080 --> 00:00:11.960 announcement screaming about wild performance 00:00:11.960 --> 00:00:14.320 claims. Exactly. For years, we were all kind 00:00:14.320 --> 00:00:16.480 of trained to look for that straight lineup. 00:00:16.660 --> 00:00:19.940 10x better, more parameters, these huge benchmark 00:00:19.940 --> 00:00:22.679 charts. That was the game. It was quiet this 00:00:22.679 --> 00:00:26.160 time. Almost noticeably so. And instead of speed 00:00:26.160 --> 00:00:29.480 or scale, the fundamental shift. The thing that 00:00:29.480 --> 00:00:33.719 seems to define GPT -5 .1 was personality. It 00:00:33.719 --> 00:00:36.520 feels like we are stepping into the age of emotional 00:00:36.520 --> 00:00:41.179 quotient or EQ in AI. Welcome to the Deep Dive. 00:00:41.219 --> 00:00:43.380 And our mission today is really to look past 00:00:43.380 --> 00:00:45.659 those old metrics to unpack these new sources 00:00:45.659 --> 00:00:47.820 you've sent over. We need to get into why this 00:00:47.820 --> 00:00:49.939 EQ focus is suddenly dominating the conversation. 00:00:50.240 --> 00:00:52.159 So we're going to dive into some huge cultural 00:00:52.159 --> 00:00:54.759 shifts. I'm talking AI in Hollywood, AI on the 00:00:54.759 --> 00:00:56.439 billboard charts, and then we'll get into the 00:00:56.439 --> 00:00:58.159 weeds on this incredible little engine called 00:00:58.159 --> 00:01:00.659 Lenin, which is all about privacy. So let's unpack 00:01:00.659 --> 00:01:03.240 this shift. Okay, so let's start with GPT -5 00:01:03.240 --> 00:01:07.359 .1's arrival. It felt... strategic didn't it 00:01:07.359 --> 00:01:09.000 like they weren't trying to sell horsepower anymore 00:01:09.000 --> 00:01:12.219 they were selling reliability comfort even that's 00:01:12.219 --> 00:01:14.459 a good way to put it the goal seems to have shifted 00:01:14.459 --> 00:01:16.659 from making the model just generically smarter 00:01:16.659 --> 00:01:19.620 to making it consistently useful for a specific 00:01:19.620 --> 00:01:22.510 person and why is that Well, because performance 00:01:22.510 --> 00:01:25.310 saturation is coming. They know that just adding 00:01:25.310 --> 00:01:27.329 more and more parameters, it gives you diminishing 00:01:27.329 --> 00:01:29.969 returns. The numbers get bigger, sure. But the 00:01:29.969 --> 00:01:32.689 practical value for you, the user, doesn't really 00:01:32.689 --> 00:01:35.230 scale at the same pace. So if you can't win on 00:01:35.230 --> 00:01:38.349 raw speed anymore, you have to win on the experience 00:01:38.349 --> 00:01:40.790 itself. It has to be consistent, predictable. 00:01:41.129 --> 00:01:43.629 It has to align with what I'm trying to do. Precisely. 00:01:43.689 --> 00:01:46.870 And they rolled out two main versions of 5 .1 00:01:46.870 --> 00:01:49.209 that really show this split. First, you've got 00:01:49.209 --> 00:01:52.390 GPT 5 .1 Instant. Right. That's your fast, casual 00:01:52.390 --> 00:01:54.709 assistant. Now, it's not just smarter, it's also 00:01:54.709 --> 00:01:57.150 warmer in its tone. And then the other one is 00:01:57.150 --> 00:02:00.310 GPT 5 .1 Thinking. That's for the deep, complex 00:02:00.310 --> 00:02:03.409 tasks. And with that one, they're focusing less 00:02:03.409 --> 00:02:05.829 on warmth and more on being just crystal clear, 00:02:05.890 --> 00:02:07.810 super efficient, and it's reasonable. But the 00:02:07.810 --> 00:02:10.289 real story, I think, is the personality presets. 00:02:10.449 --> 00:02:12.770 That's the headline feature. They've introduced 00:02:12.770 --> 00:02:15.669 eight specific modes. Yeah, that's the core of 00:02:15.669 --> 00:02:17.979 it. you could tell your ai to be professional 00:02:17.979 --> 00:02:21.599 for you know writing a formal email or you can 00:02:21.599 --> 00:02:23.439 switch it to candid if you just want the straight 00:02:23.439 --> 00:02:25.960 truth no sugar coating and then there's quirky 00:02:25.960 --> 00:02:28.960 which i love it's designed to be fun a little 00:02:28.960 --> 00:02:31.400 weird for brainstorming all on top of the old 00:02:31.400 --> 00:02:34.349 prefets like friendly and Tech geek. And the 00:02:34.349 --> 00:02:36.590 fine tuning controls are getting so detailed. 00:02:36.650 --> 00:02:39.830 You can actually adjust for conciseness or enthusiasm 00:02:39.830 --> 00:02:43.189 level or even how often it uses emojis. That's 00:02:43.189 --> 00:02:45.150 amazing. Or demand it uses bullet points for 00:02:45.150 --> 00:02:47.530 scannability. It's getting really granular. And 00:02:47.530 --> 00:02:49.490 that consistency. I mean, that's really the key 00:02:49.490 --> 00:02:51.789 to trust, isn't it? I have to admit, I still 00:02:51.789 --> 00:02:53.909 wrestle with prompt drift myself. Oh, everyone 00:02:53.909 --> 00:02:56.150 does. I'll be halfway through a complex document 00:02:56.150 --> 00:03:00.629 and the AI's tone will just shift and the whole 00:03:00.629 --> 00:03:04.199 output feels disjointed. That lack of predictability 00:03:04.199 --> 00:03:06.879 just breaks your workflow. So the coolest feature 00:03:06.879 --> 00:03:10.219 they added hits this exact problem. EPT will 00:03:10.219 --> 00:03:13.580 now detect your tone mid -chat, and it'll offer 00:03:13.580 --> 00:03:16.520 to save that style as your new default for later. 00:03:17.979 --> 00:03:20.139 That is such a smart way to create alignment. 00:03:20.340 --> 00:03:22.539 It feels like it's adapting to you. Yeah, instead 00:03:22.539 --> 00:03:25.539 of you adapting to it. Exactly. So a probing 00:03:25.539 --> 00:03:28.939 question then. If performance isn't the big headline, 00:03:29.199 --> 00:03:33.289 why is this idea of personality, of EQ, suddenly 00:03:33.289 --> 00:03:35.569 more important than the raw benchmarks. I think 00:03:35.569 --> 00:03:38.310 it means models need practical consistency and 00:03:38.310 --> 00:03:41.830 human trust to be truly useful. OK, let's connect 00:03:41.830 --> 00:03:44.969 that idea, this new focus on EQ to what's happening 00:03:44.969 --> 00:03:46.650 out in the world, because it's not just about 00:03:46.650 --> 00:03:48.389 the tech. It's about, you know, how people are 00:03:48.389 --> 00:03:50.310 actually accepting it. And what stands out right 00:03:50.310 --> 00:03:53.150 away is how people are using these tools socially. 00:03:53.710 --> 00:03:56.629 There's a new analysis of, what, 47 ,000 chat 00:03:56.629 --> 00:03:59.330 GPT chats? Yeah, huge sample. And it shows that 00:03:59.330 --> 00:04:01.629 about 10 % of users are shifting away from pure 00:04:01.629 --> 00:04:03.909 work tasks. They're using it more for emotional 00:04:03.909 --> 00:04:06.629 support, asking really personal questions. It's 00:04:06.629 --> 00:04:08.169 pretty clear people are looking for a sympathetic 00:04:08.169 --> 00:04:10.550 ear, right? Even if they know it's a digital 00:04:10.550 --> 00:04:12.389 one. So, of course, the design has to follow 00:04:12.389 --> 00:04:15.689 that use case. It demands an AI with a consistent, 00:04:15.770 --> 00:04:18.970 predictable personality. And you see this disruption 00:04:18.970 --> 00:04:21.839 hitting media. So hard. I mean, look at Hollywood. 00:04:22.000 --> 00:04:25.500 Higgs field's new recast AI. It lets you swap 00:04:25.500 --> 00:04:28.220 out entire characters in a movie. And it can 00:04:28.220 --> 00:04:30.680 handle really complex movements. This goes way 00:04:30.680 --> 00:04:33.220 beyond simple deep fakes. It points to a massive 00:04:33.220 --> 00:04:35.259 change in how movies get made. And it's not just 00:04:35.259 --> 00:04:37.660 visuals. If you move to audio, consumer acceptance 00:04:37.660 --> 00:04:40.920 is, well, it's already here. A completely AI 00:04:40.920 --> 00:04:43.860 -generated country song, Walk My Walk, didn't 00:04:43.860 --> 00:04:46.879 just chart. It hit number one. Number one on 00:04:46.879 --> 00:04:49.379 Billboard's digital song sales chart. Which is 00:04:49.379 --> 00:04:52.259 wild. I mean, think about that. It beat a real 00:04:52.259 --> 00:04:55.899 artist, Ella Langley, who has like 1 .8 million 00:04:55.899 --> 00:04:59.319 monthly listeners. Wow. The fact that a synthetic 00:04:59.319 --> 00:05:01.360 piece of art can have that kind of commercial 00:05:01.360 --> 00:05:04.060 success, it proves the barrier isn't technical 00:05:04.060 --> 00:05:06.319 quality anymore. It's just cultural acceptance. 00:05:06.560 --> 00:05:08.259 Which is often shaped by the personality of the 00:05:08.259 --> 00:05:10.560 output, right? Exactly. And you see it in work 00:05:10.560 --> 00:05:13.220 tools, too. ChatGPT is now testing group chats. 00:05:13.379 --> 00:05:15.459 Right. The AI isn't just a tool in the chat. 00:05:15.579 --> 00:05:17.620 It becomes a collaborator. Yeah. It has a role. 00:05:18.110 --> 00:05:20.610 a personality right there in the meeting. Speaking 00:05:20.610 --> 00:05:23.269 of personality, look at the voice market. Eleven 00:05:23.269 --> 00:05:26.649 Labs just launched its iconic marketplace. Now 00:05:26.649 --> 00:05:30.149 you can legally use authorized celebrity AI voices. 00:05:30.449 --> 00:05:32.569 Michael Caine is one of the first big examples. 00:05:32.810 --> 00:05:34.689 You're literally monetizing personality directly. 00:05:34.910 --> 00:05:37.329 Exactly. And the money is flowing to back this 00:05:37.329 --> 00:05:40.009 up. A company called Colab, they make an automatic 00:05:40.009 --> 00:05:43.610 code review tool that they just raised $72 million. 00:05:44.110 --> 00:05:47.430 And they have a wait list of 47 ,000 engineers. 00:05:47.850 --> 00:05:50.310 That tells you professionals are all in on this 00:05:50.310 --> 00:05:52.810 kind of automation. So another question here, 00:05:52.829 --> 00:05:56.069 then. What does the commercial success of AI 00:05:56.069 --> 00:05:58.810 music hitting number one actually tell us about 00:05:58.810 --> 00:06:01.430 the future of creating content? I think it signals 00:06:01.430 --> 00:06:03.829 that consumer acceptance of synthesized art is 00:06:03.829 --> 00:06:06.110 rising and it's rising fast. OK, so we've talked 00:06:06.110 --> 00:06:07.850 about the shift to EQ and how it's being accepted 00:06:07.850 --> 00:06:09.850 culturally. Now let's talk about control and 00:06:09.850 --> 00:06:13.490 privacy. And that brings us to LAN. LAN. It's 00:06:13.490 --> 00:06:16.160 a tiny rag engine. But let's quickly define that 00:06:16.160 --> 00:06:19.180 jargon. Yeah, good idea. RAG is Retrieval Augmented 00:06:19.180 --> 00:06:22.279 Generation. But in plain English, it's basically 00:06:22.279 --> 00:06:24.939 an AI search stack that uses your own private 00:06:24.939 --> 00:06:28.040 documents to find answers. And the big breakthrough 00:06:28.040 --> 00:06:31.819 with Lean is its size, its efficiency. This is 00:06:31.819 --> 00:06:34.220 a local vector index you can run entirely on 00:06:34.220 --> 00:06:37.839 your laptop. And it uses 97 % less storage space. 00:06:38.560 --> 00:06:42.750 97%. Yet it keeps full accuracy. It's kind of 00:06:42.750 --> 00:06:44.829 nuts. It's an incredible number. The analogy 00:06:44.829 --> 00:06:47.509 I've heard is it's like instead of a giant library 00:06:47.509 --> 00:06:50.870 catalog where you store every card forever, Lean 00:06:50.870 --> 00:06:52.930 just makes a temporary note card for the book 00:06:52.930 --> 00:06:54.370 you're looking for right now. And then throws 00:06:54.370 --> 00:06:55.790 it away. And then throws it away when you're 00:06:55.790 --> 00:06:58.490 done. It's like fitting an entire AI search engine 00:06:58.490 --> 00:07:01.290 into your backpack. Yeah, that's a great way 00:07:01.290 --> 00:07:02.930 to think about it. And what's so powerful is 00:07:02.930 --> 00:07:05.009 what it does for your privacy. You just point 00:07:05.009 --> 00:07:07.949 it at your own data, your PDFs, your code, your 00:07:07.949 --> 00:07:11.009 emails, even your browser history. And it's instantly 00:07:11.009 --> 00:07:13.629 searchable. Instantly. And because it's local, 00:07:13.730 --> 00:07:15.670 none of that data ever leaves your computer. 00:07:15.850 --> 00:07:18.649 You control the context completely. And the sources 00:07:18.649 --> 00:07:21.050 say the technical reason it's so light is because 00:07:21.050 --> 00:07:23.430 it only computes those data relationships when 00:07:23.430 --> 00:07:26.290 it absolutely needs to. It's not storing everything 00:07:26.290 --> 00:07:29.060 permanently. Right. The core of it is based on 00:07:29.060 --> 00:07:30.959 graphs, but the user doesn't even need to know 00:07:30.959 --> 00:07:33.079 that. They just get the results. And it's compatible 00:07:33.079 --> 00:07:35.660 with everything. Standard APIs, OpenAI compatible. 00:07:35.959 --> 00:07:38.860 It supports small local models like Alama or 00:07:38.860 --> 00:07:41.000 from Hugging Faith. There's a cloud option if 00:07:41.000 --> 00:07:43.300 you need to scale up, but that local first idea 00:07:43.300 --> 00:07:46.399 is really the magic here. Whoa. I mean, just 00:07:46.399 --> 00:07:50.019 imagine scaling that 97 % storage saving to a 00:07:50.019 --> 00:07:52.360 billion queries. The cost savings would be immense. 00:07:52.819 --> 00:07:56.290 For a company, yes. But for an individual, this 00:07:56.290 --> 00:07:58.930 is essential for anyone building a second brain 00:07:58.930 --> 00:08:02.550 or local dev tools. I think we have to call it 00:08:02.550 --> 00:08:04.649 the underrated AI infrastructure tool of the 00:08:04.649 --> 00:08:06.850 year. So that leads to a big question. How does 00:08:06.850 --> 00:08:09.709 running RGA locally like this fundamentally change 00:08:09.709 --> 00:08:11.790 how we should be using AI search? It gives you 00:08:11.790 --> 00:08:14.209 total privacy and control over your own data 00:08:14.209 --> 00:08:17.290 and context. Hashtag, tag, tag, outro. So if 00:08:17.290 --> 00:08:19.750 we synthesize everything from today, the big 00:08:19.750 --> 00:08:21.589 theme is that AI development is really shifting 00:08:21.589 --> 00:08:24.170 its focus. We're moving away from this race for 00:08:24.170 --> 00:08:26.680 just raw general power. And moving towards two 00:08:26.680 --> 00:08:29.240 different paths. On one side, you have highly 00:08:29.240 --> 00:08:31.519 personalized consistency. That's the EQ part. 00:08:31.720 --> 00:08:34.179 And on the other, you have privacy -focused utility, 00:08:34.440 --> 00:08:36.860 which is what we see with local tools like Lean. 00:08:36.980 --> 00:08:39.279 Right. So the practical takeaway for you listening 00:08:39.279 --> 00:08:41.919 should be this. Don't just rely on those big, 00:08:41.919 --> 00:08:44.500 flashy benchmark charts anymore. They're mostly 00:08:44.500 --> 00:08:48.360 for marketing. You need to actually test these 00:08:48.360 --> 00:08:51.559 models yourself. Find the best fit for your specific 00:08:51.559 --> 00:08:53.980 work. You have to factor in personality, consistency, 00:08:54.159 --> 00:08:58.059 and your own privacy needs now. The best model 00:08:58.059 --> 00:09:00.259 is the one that just works for you. And that 00:09:00.259 --> 00:09:02.529 leaves me with one last thought. What happens 00:09:02.529 --> 00:09:05.509 when these very advanced EQ models become truly 00:09:05.509 --> 00:09:08.570 personalized and they're running locally on a 00:09:08.570 --> 00:09:11.350 tool like Lean with access to all of our private 00:09:11.350 --> 00:09:14.750 data and context? Will we eventually start to 00:09:14.750 --> 00:09:17.110 prefer the advice of our own AI -powered digital 00:09:17.110 --> 00:09:20.149 friend over a human expert just because it knows 00:09:20.149 --> 00:09:23.049 us better? That's a deep thought to end on. We've 00:09:23.049 --> 00:09:24.649 pulled together all the sources we talked about 00:09:24.649 --> 00:09:26.529 today, and we really encourage you to explore 00:09:26.529 --> 00:09:28.309 them for yourself. Thanks for tuning in to this 00:09:28.309 --> 00:09:29.950 deep dive out to your own music.