Transcription: OpenAgents Episode 166 - OpenAI Delenda Est

Source: https://x.com/OpenAgentsInc/status/1900430864623768056 Wiki source: https://raw.githubusercontent.com/wiki/OpenAgentsInc/openagents/Video-Series.md Media title: OpenAgents - Episode 166: OpenAI Delenda Est OpenAI calls on government to ban De... Upload date: 20250314 Transcription model: gpt-4o-transcribe-diarize Generated at: 2026-06-01T04:41:12Z

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

[00:00] Christopher David: TechCrunch today,

[00:01] Christopher David: OpenAI calls deep sneak state controlled, calls for bans on PRC produced models.

[00:19] Christopher David: OpenAI is much more deep state controlled than deep sneak is PRC controlled.

[00:24] Christopher David: Hypocrites.

[00:29] Christopher David: regulatory capture so

[00:38] Christopher David: I got two things to say one is Cartago

[00:43] Christopher David: Delenda est Carthage must be destroyed open

[00:50] Christopher David: AI must be destroyed number

[00:56] Christopher David: two we made a meme

[01:04] Christopher David: Stop using DeepSeek.

[01:06] Christopher David: We're going to use DeepSeek even harder now.

[01:10] Christopher David: So I've just added DeepSeek R1,

[01:12] Christopher David: the full one,

[01:13] Christopher David: to OpenAgents for pro users.

[01:25] Christopher David: This is hosted via OpenRouter.

[01:33] Christopher David: And it's hosted on some of these.

[01:36] Christopher David: I think these are all U.S. based or I don't really know.

[01:40] Christopher David: Though there's some China based ones in here. But you know.

[01:45] Christopher David: I don't think our data is collected.

[01:47] Christopher David: Okay so it works.

[01:51] Christopher David: The reasoning models cannot use tools but you can use them in conjunction with models that can.

[01:57] Christopher David: So we could go into a historical chat.

[02:02] Christopher David: Yeah, I got a I broke something with the chat history. Let's do a new chat. Let's say um

[02:13] Christopher David: Look through the v4 repo for deep seek related code

[02:21] Christopher David: And summarize in two paragraphs.

[02:26] Christopher David: So I'll use the best model for tool use of this batch is three five sonnet.

[02:33] Christopher David: Maybe three seven sonar is better at some things.

[02:38] Christopher David: But three five is definitely the most reliable.

[02:41] Christopher David: Okay, so we did a nice little grep there.

[02:50] Christopher David: Let's see if deep seek R_ one can do anything interesting with that. Um

[02:59] Christopher David: Analyze the above and suggest next steps.

[03:02] Christopher David: For how best to leverage DeepSeek R1's power.

[03:14] Christopher David: Come on.

[03:18] Christopher David: I wonder if it's passing all 81 matches.

[03:20] Christopher David: Let's just see what happens.

[03:21] Christopher David: I don't know.

[03:35] Christopher David: No don't use tools, just analyse what you see in our history.

[03:44] Christopher David: Interestingly it did not give us okay there we go. Okay let me start by understanding this request.

[03:49] Christopher David: They want me to analyse a previous conversation history where they asked about DC related code.

[03:53] Christopher David: The user's repo has several references to deep seek R_ one, especially the handling of blah blah.

[03:59] Christopher David: Based on the deep seek related code and documentation found in the V_ four repo.

[04:04] Christopher David: Here are strategic next steps.

[04:07] Christopher David: Reasoning-centric workflows.

[04:08] Christopher David: Leverage its scale for Kaba's code analysis tasks like architectural pattern detection,

[04:12] Christopher David: multi-separate factory suggestions,

[04:13] Christopher David: cross file,

[04:14] Christopher David: dependency mapping.

[04:15] Christopher David: Yeah, I actually do want this.

[04:18] Christopher David: Distillation pipeline optimization.

[04:20] Christopher David: Implement automated benchmarking between R1 base and distilled variants.

[04:24] Christopher David: Quen llama?

[04:26] Christopher David: That's actually a good idea.

[04:28] Christopher David: Create A-B testing framework for distilled model performance on CodeComplice.

[04:32] Christopher David: That's actually a great idea.

[04:35] Christopher David: If there's some things where the distilled models can be fine enough if we're just like mass tagging or labeling stuff,

[04:42] Christopher David: don't need the full one.

[04:42] Christopher David: But yeah.

[04:45] Christopher David: Hybrid orchestration.

[04:46] Christopher David: Develop model routing logic that combines R1's reasoning with smaller models tool execution.

[04:50] Christopher David: Yep.

[04:52] Christopher David: censorship workarounds folks

[05:00] Christopher David: this is r1 we're not using any special word it's just the one from from open router so i'm pretty sure this is just the one like hot off the presses of deep seek in prc ooh sam altman you fucking it's

[05:18] Christopher David: like hey

[05:19] Christopher David: You should implement some sensor support grounds.

[05:21] Christopher David: Okay.

[05:23] Christopher David: Performance monitoring,

[05:25] Christopher David: community integration,

[05:27] Christopher David: build MCP-compatible adapters for R1-to-tool bridging,

[05:31] Christopher David: reasoning trace visualization,

[05:33] Christopher David: collaborative reasoning sessions.

[05:35] Christopher David: This leverages R1's scale while respecting its tool limitations through intelligent orchestration and distillation monitoring.

[05:44] Christopher David: This is amazing.

[05:45] Christopher David: Okay.

[05:47] Christopher David: Lastly...

[05:55] Christopher David: Hopefully,

[05:56] Christopher David: the Closed AI Labs fail in their attempt at regulatory capture,

[05:59] Christopher David: and that y'all (this is me responding to AI Czar Sachs) y'all can balance across both dimensions,

[06:06] Christopher David: US versus China and open versus closed.

[06:09] Christopher David: Seeing way too many calls for closing restricting AI,

[06:11] Christopher David: further wrong approach.

[06:13] Christopher David: USA must lead on open source AI or else. Okay,

[06:16] Christopher David: I'm on Team America because I'm American.

[06:19] Christopher David: I'm also on Team Open.

[06:22] Christopher David: So, in that way, I consider Deep Seek and Wenfang and everything,

[06:28] Christopher David: we're all on the same team. We're on Team Open.

[06:31] Christopher David: America needs to be on Team Open.

[06:33] Christopher David: It should be.

[06:35] Christopher David: If they're going to be falling

[06:39] Christopher David: for this regulatory capture ploy of Altman and Anthropic and the closed AI people.

[06:49] Christopher David: Then we're going to have some problems.

[06:52] Christopher David: See you soon.