쉐도잉 연습: I Tried PewDiePie's Odysseus AI So You Don't Have To - 영상으로 영어 말하기 배우기

- Today, I'm going to be testing PewDiePie's AI system called Odysseus so that you don't have to.

Today, we're going to look at everything that Odysseus has to offer, whether this is actually valuable for the average person and for people in the AI space.

And then at the end, I'll talk about how to install it and kind of give my final thoughts.

So, Odysseus is created by The PewDiePie.

Bro is doing side quests in retirement.

And so, I've provided the GitHub link to this in the description.

This is honestly pretty straightforward to install.

I will give it a point right out of the gate for that.

So, here are some of the features and PewDiePie himself describes this as really directly competing with big tech.

He wanted to create a platform very similar to OpenWebUI or OpenClaw that basically lets you use local models, platform.

So you really have all these different models in one place.

Now you do need Ollama if you want to run these models locally.

And there are also others like OpenRouter and these two that I haven't even heard of.

Ollama or OpenRouter are probably going to be your best bet.

Ollama is very easy to set up.

And again, we'll kind of talk about that at the end during the installation section.

He has created agents in here as well, which is pretty interesting.

I think one thing that's pretty unique is the cookbook.

Now one big problem with AI and especially people who aren't all the different types of quantizations out there,

FP8 versus FP16 versus GGUFs.

There's definitely a lot to understand and unpack before you even download your first model, whether that's locally or even sometimes via API.

I think it's still important understanding the distinction between models, for example, Sonnet versus Opus 4 .6 versus 4 .7, etc. So there's definitely a lot to unpack.

And I think this cookbook is really interesting in terms of letting people get in benefits to that.

The first being that your data is staying within your own PC.

It's not being monitored by some sort of AI tech overlord, right?

And the second being that, you know, you're saving water.

I think especially in the US, there's a big conversation about data centers right now.

And I'm personally on the side of this technology is great.

I think it's certainly helpful, but we need to make it a bit more efficient.

And I do think that using open source models is a really great first step to that.

Now, one note on open source models as well is that you do need to have

Now that typically means some sort of system that has a GPU.

I would say at least eight gigabytes of VRAM

and at least maybe 64 gigs of system RAM should get you by on some models.

I don't want to spend too much time talking about open source models.

There are a ton of different videos out there about it.

So make sure you understand a little bit of what that space looks like before you start diving into this.

He's also created deep research in here, the ability to compare against different models, which I think is interesting.

He's also created a group chat feature, which I think is pretty cool too. to write together.

Also memory and skills, I believe MCP is enabled in here.

There's some native email support, notes and tasks, calendar.

This works on mobile which is not something I'm testing here

but I think is really interesting as well and there's kind of some extra stuff.

So overall, there is definitely a lot to unpack here.

So let's go ahead and dive in.

So when you get in, you'll be greeted with something like this.

And now again, this first part of the video is going to be literally us just kind of testing this

and looking at the different features that this has to offer.

And then at the end, we'll talk about how to actually get to this point.

So right out of the gate, you have a few different options.

So I am using open source models.

I'm using the GPT OSS model through OLAMA.

Now, like I said, there are a bunch of different models you can pick.

Then you have two different modes here.

You'll notice that

when we flip to agent these two options toggle on here

first one being web search then we also have shell access now I believe

that this is like a sandbox so I'm not sure if it actually is you know impacting your device but either way, you know Just be careful with this especially when you're giving models,

you know really access to your computer, right?

It's basically the same type of problem that openclaw had

and why people were saying just be careful then We also have just a couple of other options here.

We can attach files.

We have documents, which I think is interesting

must be where he was saying that you can kind of work with the model

or you start typing and it can edit.

And then you can either full screen this by going like this, or you can just close it by going like this.

So we can see that the document is attached to the message, but I don't want that right now.

And we also have this prompt section, which is really interesting.

So when we're doing kind of big AI systems, especially if you're using the API, for example, regardless of what model it is, typically you have some sort of like prefix in your prompt.

So I think this is really interesting

that you can kind of standardize your prompt in some way ah okay

so this must be a prefix to your user message

because here we have the persona

so this is where you put your system prompt you can

put a name i love how there are some different kind of presets here then we also have temperature

and max tokens which is pretty cool as well i kind of wish

that we had a little bit more in here um sometimes

you do want to get really granular with the settings depending

on what you're doing for example the local music player

that we made tape has an ai dj feature exactly like spotify

and in my opinion it's arguably a little bit better super important.

So it would be nice to have like top P in here as well.

And then here is the group chat section, which I think is super fun.

So you can also set it whether you want it sequential or parallel, which is interesting.

Let's go ahead and do sequential.

Let's just do kind of three of these different philosophers.

I don't, I don't actually don't know if Odysseus was a philosopher, but, but I believe these two are well, Socrates, obviously.

One thing I wish that we had was the

Close out some of these like I don't want four even

three feels a little bit much little bug here Every time I go back and forth.

It seems to reset what happens if I refresh the little refresh we'll go to prompt group There we go.

So now we're back.

All right, so let's do again all four here

So actually I'm gonna do two I'm going to do Odysseus as the GPT OSS model

and then Socrates as Gemma 3 4b to go ahead and start this group chat what happens?

that Let's just start with something very simple.

We'll say, tell me a funny joke.

So it looks like we actually didn't get a response from both of them.

Let's see if we just, hmm.

100

Interesting.

101

I'm actually going to go ahead and disable the group message.

102

And let's just start a new chat, I think.

103

Let's just start simple.

104

So to make sure everything is connected, we'll just ask our GPT -OSS model.

105

And so there we go.

106

We can see that everything is connected via Ollama and we are getting our message coming through here.

107

Now, again, since this is a local model, it's going to use the resources on your system.

108

So obviously, the more that you have running, like, for example, right now I'm recording, the more difficult and longer this response is going to take.

109

But even that is pretty fast, I guess.

110

And so we can see here's a quick one from the fetch list, which is interesting.

111

We'll take a look at that.

112

Why did Beethoven get rid of his chickens?

113

All they ever said was, bah, bah, bah.

114

Okay, that's kind of funny.

115

also search the web because we do have that enabled in our settings so pretty cool stuff here.

116

We also have the thinking process and I also like that it gives you kind of some different stats here.

117

You have the token count, the time to generate, you have some additional options here which is really cool as well.

118

You can copy the message to your clipboard, you can also delete the message and then it tells you how much context window you used so really cool stuff.

119

You also have the search here so the more conversations you have, It'll make it easy to kind of find the different ones, especially if you are renaming them from the dropdown here.

120

Again, we can rename, copy chat, create a PDF or save to documents.

121

So again, some really cool options here.

122

Now, I think this is really interesting because we talk about the Kaparthi second brain method, right?

123

With obsidian, I wonder how much better this could be with some sort of obsidian connection as well.

124

But cool though, that it's still made the dock for us.

125

But I'm going to go ahead and close that again.

126

We have our dropdown here for our chats.

127

You can also set up email or write a new email from here.

128

Interesting.

129

So this, you know, is definitely very useful.

130

You can kind of see how PewDiePie has approached this with

131

more of a human first philosophy supported by AI rather than the other way around.

132

And then we also have this tool section here.

133

So I like that this is collapsible as well.

134

So I don't need to see that the whole time.

135

Also have the brain.

136

So here we have our memories.

137

So this is pretty cool.

138

I believe this is what he was saying over time.

139

It will get better at understanding you.

140

You can also import memories here from the add tab.

141

So pretty cool.

142

We also have skills.

143

Now this isn't something that I'm planning on exploring here.

144

Skills have obviously been super popular lately with Claude, for example.

145

Here's where you can actually add skills.

146

And then finally we have settings to auto extract memories.

147

Yep, auto extract skills, which is really interesting as well.

148

You can also set a max skill count per request.

149

And I like this too, to auto approve skills.

150

This is really interesting.

151

Then we have this calendar option here, which yeah, I think that's pretty cool too.

152

I feel like, unfortunately, this probably isn't too useful for most people

153

because I don't see a world where I'm going to you know, transition all my calendar events from Google or Teams or something into here.

154

It would definitely be nice if this could like auto sync or something.

155

Well, it looks like maybe we are able to.

156

It could be in the setting.

157

So, I don't want to speak too soon.

158

Again, I'm exploring this for the first time with you guys.

159

So, there totally could be things that I'm missing here.

160

But yeah, I think just overall looking at this, assuming that you can, it would be better if I could attach my own calendar,

161

which you could potentially do through MCP, for example, or API.

162

So there are a bunch of different ways you could keep building on this, and I'm sure people will, but just a thought kind of while looking at this.

163

But it is cool though, you know, if you did want to fully transition here, you could.

164

You can create different calendars, export them, import them.

165

So there's definitely a lot that you can do here.

166

Here is the compare section.

167

And I think this could definitely be interesting for comparing all the different models that we have out there.

168

I think it was NatDev way back in the day used to have something like this.

169

I'm sure that there are other platforms that do.

170

But for those of you who are like very active in the open source model space, this could be really interesting for you.

171

You can also do different types here, which I think is pretty cool as well.

172

So you can test against specific things.

173

In this case, we're just going to do chat.

174

Go ahead and start.

175

Okay.

176

And so now we have model A and model B.

177

covered is because that's what we selected in the settings.

178

Now let's come up with an interesting prompt.

179

Tell me the story of Steve Jobs and Apple, but make it an action comedy.

180

Now let's see what we get.

181

and so we can see the first one is you know really moving fast so

182

if our goal here was to verify speed this would be a clear winner the model a is already done

183

and model b is just starting the model b is still going but you know obviously let's say that i'm

184

going for speed.

185

Obviously we're also told fastest here, I'm gonna vote A.

186

I love that little animation, that's fun.

187

So Gemma 3 4B was actually the winner.

188

I was convinced that this was GPT OSS.

189

So a really cool way to kind of blind test models.

190

A little thought for PewDiePie, potentially doing this with image or video could be really insane.

191

I don't know how difficult that would be to integrate.

192

That could just be like a comfy UI MCP integration, but really interesting stuff here for sure.

193

We can also reset and go back to the blind test.

194

So we also have different preset prompts.

195

Let's actually go ahead with one of these and see what we get.

196

Explain the one cringe.

197

I don't know how to pronounce that.

198

So this must be GPT OSS since we're seeing the same kind of thought process code here.

199

And then Gemma 4B hasn't started yet.

200

Now this is interesting though.

201

So it did the first prompt, but then it went back to the action comedy Apple story, which is very bizarre.

202

So it looks like maybe there's some lingering prompt stuff being happening here.

203

Maybe something to check in on.

204

And then it goes back to the butterfly life cycle

205

with the prompt but you get the idea you can kind of blind test between models

206

and there are a bunch of different ways to do it

207

so here we go here's gemma 4b

208

and overall this seems to be a lot more detailed again i'll vote a

209

and sure enough gemma 4b so then we also have the cookbook

210

and this is what i think is really interesting

211

so you can download specific models from here

212

and you also have the ability to actually scan your device

213

and see what models that you can download we talked a

214

run and this solves the problem of trying to figure it out yourself

215

and go to a bunch of different YouTube videos, right?

216

So they also have this drop down here, which is super cool of trending models that fit your hardware.

217

So it literally tells you exactly what you can and can't run, how many downloads it has, what the size is.

218

And so the famous llama three is down here.

219

We have a bunch of different models as well.

220

So this is pretty much the obvious way to go about this.

221

And so if I click on any of these and then I hit download, it'll start downloading it for me.

222

And so automatically these are all also scanned.

223

are the models that would work best on my device.

224

You can also filter by what type of model you want.

225

So again, if I do vision, we can see Gemma 4 jumps to the top.

226

If I do coding, then it's Quan 3 Coder.

227

If I do reasoning, then it's GPT OSS 20B, which is the model that we were using earlier.

228

So super cool, super easy.

229

I think that this here definitely solves a massive problem in the AI space.

230

So great job to Pewds on this one.

231

Now we also have this dependency section, and this is described as capabilities.

232

Now, I think this is interesting because PewDiePie talked about his kind of in -house Photoshop.

233

He's trying to create his own.

234

He's kind of started that with Odysseus.

235

So, he has removed background here, real Essergan, or Esgran as I like to pronounce it, for denoise and upscale.

236

Now, I think these are interesting for those of you who haven't tried ComfyUI, but, you know, for subscribers, you guys definitely have.

237

Now, if this does support MCP, which my understanding is that it does, I think it would be a better move to just connect your ComfyUI MCP here and just have.

238

Again, assuming that's possible.

239

This has only been out for a couple days.

240

You know, maybe people haven't figured that out yet, so...

241

You also have a slot here for your hugging face token to download private models.

242

So yeah, overall, the cookbook is definitely super useful.

243

I have a deep research section here.

244

You can also select the search engine here, which I believe is based on your selection in settings.

245

So this is super cool for sure.

246

I will definitely be trying this.

247

You also have your gallery here, which I believe is what he was talking about with kind of image editing.

248

So I've gone ahead and just uploaded a pictures so this is super interesting.

249

Before we do that, let's look at the settings.

250

I'm not sure what the tagging would be for.

251

That's particularly interesting though.

252

You have a couple of different options here.

253

So what if we do new canvas?

254

Oh, interesting.

255

Wow.

256

So he's got like a whole editor in here.

257

So in a lot of ways, he does have Photoshop.

258

So here's my question right out of the gate is does the model or agent rather actually have access to this?

259

Because I could see this being huge for an agent that can actually connect to this and make edits independently.

260

I mean, otherwise it's cool, but I'm not sure that I'd actually spend a lot of time trying to use this.

261

It reminds me a lot of like the old paint software, the Microsoft Paint.

262

So again, pretty cool, but I'm not sure if it's something I'd actually end up using.

263

We also have our notes section.

264

So again, this looks like pretty much just a to do list.

265

I think it'd be a lot better if people could connect it to their own systems, come back again to Obsidian, for example.

266

And we also have different tasks here.

267

So overall, there's definitely quite a bit.

268

I mean, like I said, this was my first time kind of looking through this.

269

I wanted to do it with you guys and give my honest thoughts as I went through it.

270

I think it's definitely something that I'll play with a little bit more to really see what you can do with this.

271

It definitely seems like a reliable competitor potentially to like open web UI.

272

an open claw or Hermes agent competitor of any kind.

273

Definitely a lot of respect to Pewds for, you know, just putting this together and trying to do something for the people.

274

He's the people's man.

275

He always has been.

276

This is definitely a super cool project, but in terms of practical application in kind of the modern state of AI

277

and how people are using it at a very complex level and at a very kind of mid level, I do have kind of some doubts.

278

But, you know, as the community reported in 469 pull requests.

279

I think this thing will definitely get better with it being

280

driven by PewDiePie who definitely has like a near infinite amount of resources

281

and a huge very loyal following of people that literally grew up with this man like me.

282

I can definitely see it becoming something really big and really useful.

283

So again really cool stuff.

284

Congrats to Pewds on putting this together and I will definitely be reporting kind of anything that I think is really

285

Patreon and also our Discord so you can become a free member there or you can become a paid member.

286

We also have a lot of really cool comfy UI workflows that we've put together and tape, which I discussed earlier in the video, is also available on our Patreon.

287

That is completely free if you're interested in that.

288

So yeah, super cool and let's talk about how to install this.

289

Alright, so installing this is super easy.

290

I'm going to be showing how to do this for Windows, but there's also instructions on Apple Silicon.

291

There's a native Windows installer.

292

I wouldn't do that. with Docker.

293

So to use this method, you're going to need to install Docker and create an account.

294

Docker is free and you can find the link to that in the description below.

295

So once you have Docker installed and you have your account signed in, go ahead and make a folder somewhere on your system where it's safe and you have kind of a lot of space.

296

So I've created a folder just called Odysseus YT that I'm going to click on the search bar here and type PowerShell.

297

That's going to open up your PowerShell window and all we're going to do is go line by line here.

298

We're going to copy and paste it.

299

And so there we go.

300

We can see that the repo was cloned.

301

And if we go back into the YouTube folder, we'll see it here.

302

Now from our PowerShell, we're going to type CD, and then the name of the folder, which is Odysseus.

303

And there we go.

304

Now you can see that we've changed directories.

305

Now you're going to take this third line here, come back to PowerShell, paste it, and run it.

306

And so if you don't get any errors back, that means it worked successfully, by the way.

307

And now very simple.

308

All we're going to do is Docker compose up dash D build.

309

and get everything up and running.

310

And now we're going to type Docker compose logs Odysseus.

311

And so the first thing that you're going to look for is your kind of host.

312

So this will always be on your local host, which is this address right here.

313

So you can just copy this, go to your Explorer of choice and paste it in.

314

And you also want to add colon 7000 here.

315

Now when we hit enter, we can see that we will be brought to our page.

316

And so you're going to type admin, that is the initial username, password.

317

So you'll copy this, paste it in for your password, and sign in.

318

So now you can see that we are officially in.

319

And the first thing you want to do is go to your settings.

320

Then you want to go to your account and change the password.

321

So I'll just type in the password that I want and click update.

322

And now you can see we're good to go.

323

So that's how you can pretty quickly get set up with Odysseus.

324

It's very straightforward.

325

Now as for local models, go back to your settings here and you're going to click on add models.

326

Now to do this, like I said, you are going to want Olama Now, I use Olama, I think it's the easiest.

327

And you can download Olama by going to the link in the description, which is going to take you here.

328

So again, depending on where you are running, whether that's Mac, Linux, or Windows, you'll obviously want to select which system you're on.

329

In this case, you're going to just take this line and again, paste it into PowerShell, the same window that we have here.

330

Once that's done, you can go ahead and click scan for servers.

331

And so we can see that Olama was automatically detected.

332

It found one server with seven models that I've already installed in the past.

333

Now you do also have API here

334

so you have a bunch of different options you can pick from

335

if you want to use your OpenAI API you can

336

so definitely a lot of options and then again

337

if you are starting fresh with Olama go to the cookbook click rescan just to double check

338

and you can actually just start downloading models from here.

339

So that's everything.

340

That is Odysseus and how you get started with it.

341

Thank you guys, and I will see you in the next one.

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법

쉐도잉(Shadowing)은 원래 전문 통역사 훈련을 위해 개발된 언어 학습 기법으로, 다언어 학자인 Dr. Alexander Arguelles에 의해 대중화된 방법입니다. 핵심 원리는 간단하지만 매우 강력합니다: 원어민의 영어를 들으면서 1~2초의 짧은 지연으로 즉시 소리 내어 따라 말하는 것——마치 '그림자(shadow)'처럼 화자를 따라가는 것입니다. 문법 공부나 수동적인 청취와 달리, 쉐도잉은 뇌와 입 근육이 동시에 실시간으로 영어를 처리하고 재현하도록 훈련합니다. 연구에 따르면 이 방법은 발음 정확도, 억양, 리듬, 연음, 청취력, 말하기 유창성을 크게 향상시킵니다. IELTS 스피킹 준비와 자연스러운 영어 소통을 원하는 분들에게 특히 효과적입니다.

쉐도잉 연습: I Tried PewDiePie's Odysseus AI So You Don't Have To - 영상으로 영어 말하기 배우기

맥락 및 배경

일상 커뮤니케이션을 위한 상위 5개 문구

단계별 섀도잉 가이드

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법