Upload 9 files

KimmyK2Post/KimmyK2.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58702bb765a0e9f6c22fd3735b9906784310cdc715d4852a77b28e3b30be1db3
+size 741015924

KimmyK2Post/KimmyK2Transcript.txt

@@ -0,0 +1,483 @@
+00:00:00:50 - 00:00:01:24
+Hi everyone.
+
+00:00:01:24 - 00:00:05:20
+Welcome to our latest installment
+
+00:00:05:54 - 00:00:10:22
+This week's subject is moonshot
+
+00:00:10:22 - 00:00:13:28
+as the best open source
+
+00:00:14:09 - 00:00:18:30
+In this episode, we will go a little bit
+
+00:00:18:48 - 00:00:22:57
+We'll try it out in action and see what we
+
+00:00:22:57 - 00:00:26:43
+on some concluding remarks
+
+00:00:27:05 - 00:00:30:05
+So without further ado, join me.
+
+00:00:31:02 - 00:00:34:45
+Okay, so we're here in the moonshot
+
+00:00:34:45 - 00:00:37:45
+K2 instruct,
+
+00:00:37:45 - 00:00:39:33
+And this is the one stop shop
+
+00:00:39:33 - 00:00:41:50
+or you can download the model
+
+00:00:41:50 - 00:00:44:08
+And kudos for the moonshot team,
+
+00:00:44:08 - 00:00:48:11
+for writing a really good
+
+00:00:48:11 - 00:00:52:38
+And I think this key features section here
+
+00:00:52:38 - 00:00:56:23
+all the information, that we need to get
+
+00:00:56:50 - 00:01:01:18
+So the first key innovation
+
+00:01:01:18 - 00:01:04:18
+So it's a 1 trillion parameter
+
+00:01:04:43 - 00:01:06:09
+So let's unpack that.
+
+00:01:06:09 - 00:01:09:20
+So the total parameter account
+
+00:01:09:36 - 00:01:10:47
+So that's huge
+
+00:01:10:47 - 00:01:14:33
+I think that's the biggest open
+
+00:01:14:51 - 00:01:17:17
+And it's
+
+00:01:17:17 - 00:01:21:04
+So that is the same architecture
+
+00:01:21:38 - 00:01:26:24
+And they claim that their architecture
+
+00:01:26:24 - 00:01:29:41
+So here we see an example of open source
+
+00:01:29:59 - 00:01:32:44
+to get better, results.
+
+00:01:32:44 - 00:01:35:44
+So the second thing is
+
+00:01:36:31 - 00:01:38:48
+So what they claim is
+
+00:01:38:48 - 00:01:42:35
+that basically there's a fixed set of
+
+00:01:42:35 - 00:01:45:35
+tokens out there,
+
+00:01:45:49 - 00:01:46:56
+So it's kind of like oil.
+
+00:01:46:56 - 00:01:49:56
+We're running out of oil,
+
+00:01:50:19 - 00:01:53:21
+So what they claim
+
+00:01:53:46 - 00:01:55:56
+is be more efficient during training.
+
+00:01:55:56 - 00:01:59:50
+So for a fixed amount of tokens,
+
+00:01:59:50 - 00:02:02:50
+The model is just able to learn better
+
+00:02:03:31 - 00:02:07:29
+I don't know that much about
+
+00:02:07:29 - 00:02:10:09
+If you do, please post in the comments.
+
+00:02:10:09 - 00:02:13:09
+And we can, curious to learn
+
+00:02:13:39 - 00:02:17:24
+And basically they're claiming
+
+00:02:17:35 - 00:02:19:07
+agentic coding.
+
+00:02:19:07 - 00:02:22:56
+And it’s really good
+
+00:02:22:56 - 00:02:27:08
+So it's if you plug it into a lots of MCP
+
+00:02:27:08 - 00:02:31:59
+or, or whatever, it should use those
+
+00:02:32:26 - 00:02:36:08
+So over here they put a little small file
+
+00:02:36:08 - 00:02:39:51
+with some, results,
+
+00:02:39:51 - 00:02:42:20
+So I've opened that up here
+
+00:02:42:20 - 00:02:46:09
+Basically, long story short,
+
+00:02:46:42 - 00:02:51:02
+against Deepseek, Qwen, 
+
+00:02:51:27 - 00:02:54:24
+and, basically
+
+00:02:54:24 - 00:02:58:07
+Kimmy K2's better than all
+
+00:02:59:14 - 00:03:02:37
+It's they're also
+
+00:03:02:52 - 00:03:05:44
+And it's almost as good as Claude 4
+
+00:03:05:44 - 00:03:08:44
+So here
+
+00:03:08:51 - 00:03:11:53
+They're claiming
+
+00:03:12:13 - 00:03:15:13
+So basically long story short,
+
+00:03:15:22 - 00:03:18:20
+agentic uses use cases and coding.
+
+00:03:18:20 - 00:03:20:41
+So with that being said,
+
+00:03:20:41 - 00:03:23:37
+let's take a look at two
+
+00:03:23:37 - 00:03:27:26
+websites where you can use
+
+00:03:27:35 - 00:03:30:35
+and agent encoding task, and let's see how
+
+00:03:33:02 - 00:03:33:22
+Cool.
+
+00:03:33:22 - 00:03:34:40
+So we're here on Deepsite,
+
+00:03:34:40 - 00:03:36:58
+which is an application
+
+00:03:36:58 - 00:03:40:11
+where you can vibe code
+
+00:03:40:34 - 00:03:43:18
+So here I'll go to the settings
+
+00:03:43:18 - 00:03:46:58
+and I'll select communicate
+
+00:03:47:17 - 00:03:48:46
+And then yep.
+
+00:03:50:08 - 00:03:51:10
+And then for the
+
+00:03:51:10 - 00:03:54:58
+prompt, I told it to create an interactive
+
+00:03:55:16 - 00:03:56:25
+And then hovering on a country
+
+00:03:56:25 - 00:03:59:25
+should pop up some basic facts
+
+00:03:59:35 - 00:04:02:34
+you know, the capital, the flag,
+
+00:04:02:34 - 00:04:06:17
+And then I told it to me to make it
+
+00:04:06:53 - 00:04:09:35
+Okay. So let's just see it in action.
+
+00:04:09:35 - 00:04:10:40
+Let's see how well Kimmy K2
+
+00:04:10:40 - 00:04:13:40
+does on this task
+
+00:04:16:01 - 00:04:16:13
+Cool.
+
+00:04:16:13 - 00:04:19:13
+So I'm not speeding anything up.
+
+00:04:20:26 - 00:04:22:40
+You can kind of see how fast is going.
+
+00:04:22:40 - 00:04:24:26
+Whoa. So, yeah, within a minute.
+
+00:04:24:26 - 00:04:26:23
+We already have an answer.
+
+00:04:26:23 - 00:04:28:04
+And this is actually pretty cool.
+
+00:04:28:04 - 00:04:31:54
+So obviously there are some,
+
+00:04:31:54 - 00:04:34:29
+So some things are showing up as unknown.
+
+00:04:34:29 - 00:04:35:50
+No flags or anything like that.
+
+00:04:35:50 - 00:04:38:46
+But the map is pretty accurate.
+
+00:04:38:46 - 00:04:42:20
+The style guides, pretty pretty closely.
+
+00:04:42:52 - 00:04:45:42
+And then, you know,
+
+00:04:45:42 - 00:04:47:21
+whatever data issue is happening here.
+
+00:04:47:21 - 00:04:49:05
+I'm sure it's just like a small bug.
+
+00:04:49:05 - 00:04:51:04
+But, yeah, that's very impressive.
+
+00:04:51:04 - 00:04:52:59
+The inference was really fast.
+
+00:04:52:59 - 00:04:57:22
+So that's one of the benefits
+
+00:04:57:42 - 00:04:59:17
+sort of inference providers
+
+00:04:59:17 - 00:05:01:59
+have access to the model,
+
+00:05:01:59 - 00:05:06:15
+So you see how fast that was despite
+
+00:05:06:29 - 00:05:07:40
+So that's super cool.
+
+00:05:09:43 - 00:05:11:57
+So I wanted to try and make it to again.
+
+00:05:11:57 - 00:05:13:19
+With a different app.
+
+00:05:13:19 - 00:05:18:25
+So now I'm using AnyCoder also hosted
+
+00:05:18:55 - 00:05:23:22
+Websites really easily and it's
+
+00:05:23:35 - 00:05:27:25
+So you can upload an image and it will
+
+00:05:27:42 - 00:05:29:03
+It also has web search.
+
+00:05:29:03 - 00:05:32:03
+And here you can see
+
+00:05:32:24 - 00:05:36:22
+So I just told it to redesign my website
+
+00:05:36:22 - 00:05:39:22
+So we'll see what comes up.
+
+00:05:41:46 - 00:05:43:50
+You can see here
+
+00:05:43:50 - 00:05:46:50
+pretty fast as well.
+
+00:05:49:59 - 00:05:53:14
+So, the model, finished generating now.
+
+00:05:53:38 - 00:05:55:05
+And it's actually pretty cool.
+
+00:05:55:05 - 00:05:58:31
+This is, You know,
+
+00:05:58:31 - 00:06:01:38
+when I said the cyberpunk esthetic,
+
+00:06:02:13 - 00:06:05:37
+This is my,
+
+00:06:05:37 - 00:06:08:48
+So you can see it's
+
+00:06:08:48 - 00:06:11:10
+So, I use this instead.
+
+00:06:11:10 - 00:06:12:55
+So, you know, as we can see in these two
+
+00:06:12:55 - 00:06:16:45
+can we get two is actually pretty good at,
+
+00:06:16:45 - 00:06:17:55
+you know, very simple prompts.
+
+00:06:17:55 - 00:06:19:00
+I didn't try anything crazy.
+
+00:06:19:00 - 00:06:22:49
+And you can always reiterate on them, 
+
+00:06:22:49 - 00:06:25:47
+So I think it's yeah,
+
+00:06:28:43 - 00:06:30:06
+So we've seen how good the
+
+00:06:30:06 - 00:06:33:51
+benchmarks can be, and we've seen
+
+00:06:34:13 - 00:06:39:17
+Has Kimi K2 become the undisputed
+
+00:06:39:50 - 00:06:41:50
+So far the answer is no.
+
+00:06:41:50 - 00:06:42:24
+And in fact,
+
+00:06:42:24 - 00:06:45:24
+the reception has been a little bit
+
+00:06:45:33 - 00:06:49:02
+So here for comparison, I'm
+
+00:06:49:02 - 00:06:54:04
+of DeepSeek V3 and Kimi K2,
+
+00:06:54:04 - 00:06:57:31
+is enjoying more than 400,000 
+
+00:06:58:06 - 00:07:00:46
+Kimi K2 is less than 100,000.
+
+00:07:00:46 - 00:07:02:19
+So there's a big gap there.
+
+00:07:02:19 - 00:07:05:55
+And I think, I know
+
+00:07:05:55 - 00:07:08:55
+Kimi K2 hasn't
+
+00:07:09:08 - 00:07:12:08
+Is that the most popular
+
+00:07:12:10 - 00:07:15:48
+Haven't added
+
+00:07:16:09 - 00:07:21:46
+So, vs code cursor reply
+
+00:07:21:53 - 00:07:25:29
+Haven't added support yet for Kimi k2.
+
+00:07:25:45 - 00:07:29:09
+Obviously you can configure these tools
+
+00:07:29:18 - 00:07:31:36
+so it is possible to use Kimmy K2
+
+00:07:31:36 - 00:07:34:02
+But I think, you know,
+
+00:07:34:02 - 00:07:37:16
+support means that, you know, less
+
+00:07:37:16 - 00:07:38:07
+And, you know,
+
+00:07:38:07 - 00:07:41:55
+maybe that there's just less appetite
+
+00:07:41:57 - 00:07:45:03
+So not exactly sure what causes that.
+
+00:07:45:03 - 00:07:48:44
+Maybe the model's like really big
+
+00:07:49:09 - 00:07:51:06
+inference support.
+
+00:07:51:06 - 00:07:54:19
+Is, you know, it's
+
+00:07:54:21 - 00:07:56:28
+I think it started like late last week.
+
+00:07:56:28 - 00:07:58:46
+So maybe it's just a matter of time.
+
+00:07:58:46 - 00:08:02:55
+Or or not,
+
+00:08:03:28 - 00:08:05:57
+But in short,
+
+00:08:05:57 - 00:08:06:48
+Really impressive.
+
+00:08:06:48 - 00:08:09:44
+It's a really impressive
+
+00:08:09:44 - 00:08:13:33
+So I'm excited to see what model comes
+
+00:08:13:33 - 00:08:17:43
+what the next, you know, leaderboard
+
+00:08:17:52 - 00:08:21:40
+So we'll just we'll just have to, 
+
+00:08:21:49 - 00:08:24:06
+So anyways, thanks so much for joining me.
+
+00:08:24:06 - 00:08:24:44
+Until next time.
+