Transcription: OpenAgents Episode 046 - Hello EpsteinSleuth

Source: https://twitter.com/OpenAgentsInc/status/1745521898824356193 Wiki source: https://raw.githubusercontent.com/wiki/OpenAgentsInc/openagents/Video-Series.md Media title: OpenAgents - Episode 046: Hello EpsteinSleuth We can now upload multiple PDFs dire... Upload date: 20240111 Transcription model: mlx-whisper/mlx-community/whisper-tiny Generated at: 2026-06-01T20:20:03Z

Machine-generated transcript. Review speaker labels and wording before using this as quote-grade source material.

[00:00] Christopher David: Happy Bitcoin ETF Day. Let's take a look. So we have this knowledge tab on our agent page now.

[00:08] Christopher David: So I can upload a PDF and I just a second ago got this to really work for the first time with one document.

[00:17] Christopher David: So let's kind of evaluate this together. So I created a new agent and here's what that process looks like.

[00:25] Christopher David: I've made a name description instructions. Welcome message.

[00:29] Christopher David: I didn't really give it anything good.

[00:35] Christopher David: Anything else? I'm duplication going on here. But.

[00:39] Christopher David: I said, who was telling who was opinion in the deposition?

[00:42] Christopher David: This was the first message that I sent and this would have no way of answering that question just by consulting.

[00:48] Christopher David: It's knowledge based.

[00:51] Christopher David: So it, you know, ran a similarity search over the document that I uploaded and I uploaded this 1320-8, which was one of the documents from that.

[01:08] Christopher David: That's the first time I've been in the first place.

[01:16] Christopher David: I've been in the first place.

[01:20] Christopher David: I've been in the first place.

[01:24] Christopher David: I've been in the first place.

[01:28] Christopher David: I've been in the first place.

[01:33] Christopher David: I'll try asking it another thing.

[01:40] Christopher David: I don't know what this is.

[01:46] Christopher David: Where do.

[01:50] Christopher David: What do does the document say about?

[01:55] Christopher David: One thing we need to add is one of those little like spinning loading indicators for what's searching.

[02:01] Christopher David: I'll add a little bit of the information.

[02:09] Christopher David: I'll add a little bit of the information.

[02:13] Christopher David: I'll add a little bit of the information.

[02:17] Christopher David: I'll add a little bit of the information.

[02:21] Christopher David: I'll add a little bit of the information.

[02:24] Christopher David: I'll add a little bit of the information.

[02:27] Christopher David: I'll add a little bit of the information.

[02:35] Christopher David: I'll add a little bit of the information.

[02:41] Christopher David: I'll add a little bit of the information.

[02:45] Christopher David: I'll add a little bit of the information.

[02:51] Christopher David: I'll add a little bit to process the embeddings and a lot of progress in the cater to the interface in a minute.

[03:01] Christopher David: Let's take a look at this.

[03:03] Christopher David: This is the database.

[03:05] Christopher David: I'm still refreshing.

[03:07] Christopher David: This is our database of embeddings along with the content.

[03:13] Christopher David: I'll write an image of the page.

[03:23] Christopher David: I'll add a little bit of the information.

[03:29] Christopher David: I'll add a little bit of the information.

[03:33] Christopher David: This is a 10-page document.

[03:37] Christopher David: I'll add a little bit of the information.

[03:43] Christopher David: I'll add a little bit of the information.

[03:49] Christopher David: I'll add a little bit of the information.

[03:55] Christopher David: I'll add a little bit of the information.

[03:59] Christopher David: I'll add a little bit of the information.

[04:07] Christopher David: I'll add a little bit of the information.

[04:13] Christopher David: I'll add a little bit of the information.

[04:17] Christopher David: I'll add a little bit of the information.

[04:21] Christopher David: I'll add a little bit of the information.

[04:25] Christopher David: Let's try what did Jane do number one.

[04:29] Christopher David: Tell me what Jane do number one.

[04:41] Christopher David: Set.

[04:49] Christopher David: If you want to add about letting you see embedding progress and adding some retries and

[04:55] Christopher David: make it zoom down.

[04:59] Christopher David: This specific statement our test may provide is not mentioned.

[05:07] Christopher David: I'll add number one.

[05:11] Christopher David: Page one of 10.

[05:17] Christopher David: Okay.

[05:21] Christopher David: This is a disgusting material.

[05:33] Christopher David: I'll add a little bit of the information.

[05:41] Christopher David: I'll add a little bit of the information.

[05:47] Christopher David: Let me know what you think about that.

[06:01] Christopher David: See.

[06:07] Christopher David: Jane do number four.

[06:17] Christopher David: That might be true.

[06:31] Christopher David: Tell me what it says about Jane do number four.

[06:57] Christopher David: I'll add a little bit of the information.

[07:07] Christopher David: I'll add a little bit of the information.

[07:11] Christopher David: I'll add a little bit of the information.

[07:21] Christopher David: I'll add a little bit of the information.

[07:27] Christopher David: I'll add a little bit of the information.

[07:37] Christopher David: I'll add a little bit of the information.

[07:41] Christopher David: I'll add a little bit of the information.

[07:45] Christopher David: I'll add a little bit of the information.

[07:49] Christopher David: There's probably a whole lot more that's possible.

[07:55] Christopher David: Rather than just this basic coastline similarity surge in feeding raw material and then like passing the raw material to the language model.

[08:05] Christopher David: What this probably would not be good at doing is generating like summaries or extrapolations based on like the entire document.

[08:18] Christopher David: There's probably more that needs to be done.

[08:20] Christopher David: We need to kind of alter our retrieval algorithms are kind of analysis and maybe do more different passes.

[08:28] Christopher David: For example, if you have 10 different pages, maybe you could have it generate a bunch of like more condensed insights per page.

[08:40] Christopher David: Where you then be able to fit multiple of those condensed insights into prompts where you wouldn't be able to fit like the entire 10 page document into one.

[08:47] Christopher David: Like call to a language model or if some model could eat or understand a 10 page document.

[08:55] Christopher David: There's probably like a hundred pages or a thousand page document that would not be able to.

[08:59] Christopher David: So there's probably some additional reasoning that we're going to want to put in here.

[09:03] Christopher David: What I found is that there's a lot of companies that like try to solve this for you and they say like, oh we use our super model that lets our super algorithm lets us.

[09:14] Christopher David: You know fill in the blank.

[09:16] Christopher David: But what I'm thinking is more powerful is to have all this be very open and let other people kind of play with other people's prompts and kind of collectively improve each other's prompts.

[09:28] Christopher David: So we're going to be doing that.

[09:30] Christopher David: It's kind of like what we've been building towards and we're going to be you know expanding beyond just this kind of Epstein example.

[09:38] Christopher David: I'm happy it for it to be a case study and we're going to keep revisiting this to like.

[09:43] Christopher David: Have it be better and better.

[09:45] Christopher David: But we want to start providing incentives for other Epstein slutes or other people who want to help with this or any other.

[09:56] Christopher David: Subject.

[09:58] Christopher David: Have a financial incentive in coming into improve these algorithms.

[10:03] Christopher David: So we'll keep revisiting these themes as we move forward.

[10:07] Christopher David: If you want to play around with this a little bit all the turn on the ability to create agents yourself and upload your own knowledge.

[10:15] Christopher David: Note that everything is going to be public for now or just assume that everything is public.

[10:19] Christopher David: But you're welcome to play around with this.

[10:23] Christopher David: Let us know your feedback.