WEBVTT 00:00.000 --> 00:15.000 So, we have to start quickly, so let me introduce Olivia, who will be talking about building 00:15.000 --> 00:19.000 your LN second brain and Mike is yours. 00:19.000 --> 00:23.000 All right, we're just going to have to go because we've got five minutes flat in order 00:23.000 --> 00:26.000 to do a whirlwind tour of building your second brain. 00:26.000 --> 00:32.000 I'm Olivia, I've been doing AI, I'm machine learning for about 15 years and I'm new here 00:32.000 --> 00:33.000 at Phosphum. 00:33.000 --> 00:39.000 Here's what you already know, AI is everywhere and we need Phosph AI solutions now. 00:39.000 --> 00:44.000 What you might know, you can download open source models today, local open source frameworks, 00:44.000 --> 00:51.000 make them easy to use, and they even mimic the open AI API spec. 00:51.000 --> 00:55.000 You've probably learned that this morning, but now you'll see it in detail. 00:55.000 --> 00:58.000 So, what do we mean when we talk about first brain versus second brain? 00:58.000 --> 01:03.000 Your first brain is great at things like creativity, logic, reasoning, and identifying facts. 01:03.000 --> 01:06.000 Your mileage may vary, but don't use an AI for that. 01:06.000 --> 01:12.000 Your second brain, memory aid, combining concepts, summarizing info into new formats. 01:12.000 --> 01:16.000 Those are the things that we can use our local second brain for. 01:16.000 --> 01:22.000 Our big goal interaction here is, can you break down the big project and to several possible steps 01:22.000 --> 01:25.000 for me and add reminders to my to-do list. 01:25.000 --> 01:32.000 You want it really using the data that belongs to you in your own way. 01:32.000 --> 01:35.000 What does that look like when we break it down into a stack? 01:35.000 --> 01:38.000 We need something private, local, low energy. 01:38.000 --> 01:41.000 We need a clever model and curated data sources. 01:41.000 --> 01:45.000 We need a chat interface that's just like chatGbT that uses your notes. 01:45.000 --> 01:49.000 Process the information the way you need it to do and can act on your behalf. 01:49.000 --> 01:51.000 So, let's break that down. 01:51.000 --> 01:56.000 How are we going to get to our private, low energy, clever model with curated data sources? 01:56.000 --> 01:59.000 We're going to be using Olamma and Granite. 01:59.000 --> 02:03.000 Olamma, you've heard this already today if you've been here for a while, 02:03.000 --> 02:07.000 but it's got an easy to use command line interface, a great model registry. 02:07.000 --> 02:12.000 It's open AI API compatible and useslamma.cpp for model inference. 02:12.000 --> 02:15.000 Granite, I am extremely biased here. 02:15.000 --> 02:19.000 My team created this, but the Granite model is great because it's small. 02:19.000 --> 02:25.000 It's highly performant across all of the benchmarks and on 95% open data. 02:25.000 --> 02:30.000 And it's designed for enterprise tasks. 02:30.000 --> 02:34.000 So, now we can go ahead and pull our model down, run it on our laptops, 02:34.000 --> 02:37.000 and get to a quick response. 02:37.000 --> 02:39.000 Okay, so now we want to add a chat interface. 02:39.000 --> 02:41.000 So, it looks like chatGbT. 02:41.000 --> 02:45.000 Our choices are going to be open web UI or anything at all. 02:45.000 --> 02:50.000 Anything at all in an open web UI, both awesome options in the open source space. 02:50.000 --> 02:56.000 Anything at all in an MIT license, it's got a standalone app and cross-operating system support. 02:56.000 --> 03:03.000 Open web UI, BSD3 clause license, web interface, it's multi-user enterprise hostable. 03:03.000 --> 03:06.000 Both of them can use our Olamma that we've already put together. 03:06.000 --> 03:09.000 It's got a familiar chat interface, much like chatGbT, 03:09.000 --> 03:11.000 the ability to do collections of data. 03:11.000 --> 03:13.000 So, here we are. 03:13.000 --> 03:17.000 Now we can talk in our little open web UI instance that's running on our laptop, 03:17.000 --> 03:21.000 and hey, now we can get the same response in a chat interface. 03:21.000 --> 03:23.000 Okay, great. 03:23.000 --> 03:25.000 Now we wanted to actually use our notes. 03:25.000 --> 03:28.000 So, we can add this in two ways. 03:28.000 --> 03:30.000 Langchain or Lama Index. 03:30.000 --> 03:35.000 Langchain, Lama Index, both awesome ways to add new things to your models. 03:36.000 --> 03:38.000 Langchain is more of a general purpose. 03:38.000 --> 03:43.000 LLem pipeline framework enables rag, but also does things like tool use and all kinds of stuff. 03:43.000 --> 03:47.000 Lama Index is a little bit more tuned towards actually doing rag in particular. 03:47.000 --> 03:52.000 It's got a lot of data indexing specialty, and it's more document-specific. 03:52.000 --> 03:53.000 But cool thing. 03:53.000 --> 03:57.000 Open web UI and anything else both use Langchain under the hood. 03:57.000 --> 04:01.000 So, we're just going to use that for a moment and use their collections in our face. 04:01.000 --> 04:02.000 So, we're going to sit here. 04:02.000 --> 04:05.000 We're going to create an knowledge base in our open web UI interface. 04:05.000 --> 04:09.000 We're going to pull in our Obsidian notes, which are already sitting on our laptop and 04:09.000 --> 04:15.000 Markdown format, and that it'll go ahead and start syncing that directory into our thing. 04:15.000 --> 04:20.000 So, now, when we actually ask our chat, you can see instance questions. 04:20.000 --> 04:28.000 Hey, now we can reference our Obsidian collection, and it's using real actual meeting notes that are on our laptop. 04:28.000 --> 04:29.000 Okay. 04:29.000 --> 04:31.000 Now, we need to process information like we would. 04:31.000 --> 04:33.000 That means we need to add agents. 04:33.000 --> 04:34.000 So, we're looking at auto-gen. 04:34.000 --> 04:38.000 What's an agent, by the way, LLem prompt designed to operate autonomously. 04:38.000 --> 04:40.000 Don't have time to tell you more. 04:40.000 --> 04:46.000 A agent framework, auto-gen 2, abstract complex agent interaction away and allows multi-agent 04:46.000 --> 04:48.000 interaction for easy tool use. 04:48.000 --> 04:49.000 Okay. 04:49.000 --> 04:54.000 Now, we get a whole bunch of agents running in the background, running a big execution plan, 04:54.000 --> 04:58.000 searching the internet to bring more information into our project. 04:58.000 --> 05:02.000 And the last thing we need to do is be able to actually modify our things. 05:02.000 --> 05:08.000 So, we want to add tools tool calling uses JSON generated by the LLem to make a function call. 05:08.000 --> 05:10.000 So, we can do just about anything. 05:10.000 --> 05:14.000 We want it to go ahead and store these tasks back in our task manager. 05:14.000 --> 05:15.000 What it's in sprocket down. 05:15.000 --> 05:20.000 And now, we have something that actually gives us that second brain goal interaction. 05:20.000 --> 05:21.000 That was easy, right? 05:21.000 --> 05:25.000 Fantastic. 05:25.000 --> 05:27.000 This is great. 05:27.000 --> 05:29.000 Thank you. 05:29.000 --> 05:31.000 Like speed dating almost. 05:31.000 --> 05:32.000 There you go. 05:32.000 --> 05:33.000 Yep.