Episode 44
Meta’s Shipping Spree: From Robotics to Video
Keywords Meta, CoTracker 3, MovieGen, robotics, open source, technology, entrepreneurship, AI, innovation, podcast Summary In this episode of Vichy Step Podcasts, host Dalton Anderson…
Keywords Meta, CoTracker 3, MovieGen, robotics, open source, technology, entrepreneurship, AI, innovation, podcast Summary In this episode of Vichy Step Podcasts, host Dalton Anderson discusses the latest innovations from Meta, including the CoTracker 3, a groundbreaking computer vision model, and MovieGen, a text-to-video generation tool. He also explores Meta's advancements in robotics and the significance of open-source technology in shaping the future. Despite battling COVID, Dalton shares insights on the potential applications of these technologies in various fields, emphasizing the importance of innovation and collaboration in the tech industry. Takeaways Meta has recently launched several innovative technologies. CoTracker 3 offers advanced tracking capabilities that outperform traditional methods. MovieGen allows users to create videos from text descriptions. Open-source technology is becoming increasingly important in the tech landscape. Meta's advancements in robotics could revolutionize various industries. The future of technology is promising, with many exciting developments on the horizon. Collaboration and community feedback are essential for technological progress. Sparsh, meaning touch, highlights the focus on sensory technology in robotics. The integration of AI and robotics will enhance user experiences in VR and beyond. Dalton emphasizes a positive outlook on the future of technology and innovation.
Episode content
Explore every layer of this episode.
Each article, guide, analysis, and field note has its own focused page and stays linked to this source conversation.
Articles & stories
Narrative and editorial pieces that carry the conversation forward.
Venture Step E044: Meta's 2024 AI Release Run
Revisit Venture Step E044 and see how CoTracker3, Movie Gen, Spirit LM, and Sparsh represented four very different kinds of Meta AI release.
Meta AI Releases 2024: An Artifact-Level Map
Compare CoTracker3, Movie Gen, Spirit LM, and Sparsh by paper, code, weights, license, access, evidence, and current maintenance state.
Research & analysis
Evidence-led work that tests and expands the claims in the conversation.
Spirit LM Paper Code and License Record
The [Spirit LM paper](https://arxiv.org/abs/2402.05755) describes a 7 billion parameter pretrained text language model extended to speech through continued training on te
Open Source AI Artifact and License Record
The [Open Source AI Definition 1.0](https://opensource.org/ai/open-source-ai-definition) was announced by the Open Source Initiative on October 28, 2024. It requires free
Movie Gen Paper Access and Rights Record
Meta's [Movie Gen publication](https://ai.meta.com/research/publications/movie-gen-a-cast-of-media-foundation-models/) presents a family of media foundation models for te
Meta AI 2024 Release Artifact Map
The four projects discussed in E044 were research releases, but they did not expose the same artifacts or rights.
AI Research Release Evaluation Framework
A review starts with a decision, not an announcement. The reviewer should state the task, current baseline, acceptable failure cost, protected data, required latency, int
AI Research Ecosystem Strategy Record
In [Open Source AI Is the Path Forward](https://about.fb.com/news/2024/07/open-source-ai-is-the-path-forward/), Mark Zuckerberg argues that Meta benefits when an ecosyste
Field notes
Focused observations and durable ideas worth carrying into other work.
Spirit LM Explained: How Speech and Text Interleave
Understand how Meta's Spirit LM mixes speech and text tokens, what Base and Expressive add, how the model was evaluated, and what its license permits.
What Open Source AI Means: An Artifact Test
Understand open source AI by checking code, weights, data information, licenses, services, and maintenance instead of trusting one broad label.
How AI Research Releases Build an Ecosystem
Evaluate whether AI papers, models, tools, and pilots create a durable ecosystem through complements, interfaces, distribution, feedback, and value capture.
Meta Movie Gen Explained: Research, Access, and Limits
Learn what Meta Movie Gen demonstrated across video, personalization, editing, and audio, plus what its 2024 research release did not make public.
How to Evaluate an AI Research Release
Use a practical evidence path to review an AI paper, demo, code release, weights, license, benchmark, compute needs, maintenance, and field fit.
Full episode
Read the complete record.
The show notes, transcript, and source trail remain on this canonical episode page.
TranscriptRead the full conversation.
Meta's Shipping Spree: From Robotics to Video
Source fidelity
The timestamped transcript below is the exact English auto-generated SRT caption file recovered from the preserved YouTube episode on 2026-07-28. No spelling, punctuation, speaker, or wording corrections were applied. The original SRT is preserved beside this note as E44 - YouTube automatic captions.en.srt.
Timestamped transcript
1 00:00:01,360 --> 00:00:05,600 welcome to vep podcast where we discuss
2 00:00:03,159 --> 00:00:08,280 entrepreneurship industry Trends and the
3 00:00:05,600 --> 00:00:11,120 occasion book review fair play or
4 00:00:08,280 --> 00:00:13,920 playing to win meta has been on fire
5 00:00:11,120 --> 00:00:15,360 lately but not in a bad way in this
6 00:00:13,920 --> 00:00:17,480 episode we're going to discuss some of
7 00:00:15,360 --> 00:00:20,279 the things that they have shipped in the
8 00:00:17,480 --> 00:00:24,320 last five or so weeks this is going to
9 00:00:20,279 --> 00:00:27,160 include some of their robotics releases
10 00:00:24,320 --> 00:00:30,080 some of their tracking releases and
11 00:00:27,160 --> 00:00:33,440 we'll have a actual demo of a video that
12 00:00:30,080 --> 00:00:35,079 I took in Japan of a random man walking
13 00:00:33,440 --> 00:00:37,480 down a sidewalk and we're going to be
14 00:00:35,079 --> 00:00:40,840 tracking him which is pretty cool and
15 00:00:37,480 --> 00:00:40,840 then there's also some
16 00:00:40,960 --> 00:00:46,039 hollywoodish not publicly released yet
17 00:00:43,399 --> 00:00:48,680 but madz movie J but of course before we
18 00:00:46,039 --> 00:00:51,199 dive in I'm your host Dalton
19 00:00:48,680 --> 00:00:54,239 Anderson my background is a bit of mix
20 00:00:51,199 --> 00:00:57,399 of programming data science or and
21 00:00:54,239 --> 00:00:59,359 insurance offline you can find me
22 00:00:57,399 --> 00:01:00,960 building my side business lost in a good
23 00:00:59,359 --> 00:01:02,879 book
24 00:01:00,960 --> 00:01:05,720 you can listen to his podcast in both
25 00:01:02,879 --> 00:01:07,159 video and audio format on YouTube and if
26 00:01:05,720 --> 00:01:09,479 audio is more thing you can find the
27 00:01:07,159 --> 00:01:11,280 podcast on Apple podcast Spotify or
28 00:01:09,479 --> 00:01:14,159 wherever else you get your
29 00:01:11,280 --> 00:01:18,280 podcast and another thing I should State
30 00:01:14,159 --> 00:01:22,240 before we start this episode is I am in
31 00:01:18,280 --> 00:01:24,520 a state of delirium I have covid and
32 00:01:22,240 --> 00:01:28,000 this is not the co that I want to meet
33 00:01:24,520 --> 00:01:30,759 again I would like the kind Co that
34 00:01:28,000 --> 00:01:35,759 treats me nicely
35 00:01:30,759 --> 00:01:42,680 and this Co is not so friendly so I am
36 00:01:35,759 --> 00:01:46,479 throwing that out there that I may
37 00:01:42,680 --> 00:01:49,320 be mentally a little jump starting like
38 00:01:46,479 --> 00:01:52,920 just jumping about in the episode trains
39 00:01:49,320 --> 00:01:54,920 of thoughts not being complete just in
40 00:01:52,920 --> 00:01:57,039 general so I'm just throwing that out
41 00:01:54,920 --> 00:02:00,840 there I'm in a state of delirium I would
42 00:01:57,039 --> 00:02:01,840 describe myself as feeling weightless
43 00:02:00,840 --> 00:02:04,680 but
44 00:02:01,840 --> 00:02:07,320 cloudy that's how I would feel at the
45 00:02:04,680 --> 00:02:09,560 moment so just throwing that out there
46 00:02:07,320 --> 00:02:12,440 so when you are listening to this
47 00:02:09,560 --> 00:02:14,480 episode you are not like what the heck
48 00:02:12,440 --> 00:02:18,120 this is crazy this doesn't make any
49 00:02:14,480 --> 00:02:20,840 sense and it may or may not make sense
50 00:02:18,120 --> 00:02:22,680 but this is the issues that I find my
51 00:02:20,840 --> 00:02:26,440 myself in
52 00:02:22,680 --> 00:02:29,519 because I have fallen victim to not
53 00:02:26,440 --> 00:02:31,160 having episodes in the bank and I've
54 00:02:29,519 --> 00:02:33,440 never episodes in the bank so I've been
55 00:02:31,160 --> 00:02:36,000 recording every episode every week
56 00:02:33,440 --> 00:02:38,959 instead of having many episodes in the
57 00:02:36,000 --> 00:02:42,640 bank and
58 00:02:38,959 --> 00:02:45,159 potentially having a Keepsake
59 00:02:42,640 --> 00:02:46,599 and yeah I couldn't have that many
60 00:02:45,159 --> 00:02:48,800 episodes in the bank because of the
61 00:02:46,599 --> 00:02:50,400 nature of the content I create but like
62 00:02:48,800 --> 00:02:52,280 two episodes would be
63 00:02:50,400 --> 00:02:55,959 grand
64 00:02:52,280 --> 00:02:57,400 so just wanted to let you know if you're
65 00:02:55,959 --> 00:02:58,599 a little confused about the stuff I'm
66 00:02:57,400 --> 00:03:01,879 talking about and the way that I'm
67 00:02:58,599 --> 00:03:03,599 explaining it that's probably why okay
68 00:03:01,879 --> 00:03:05,000 so now we've got that straightened out
69 00:03:03,599 --> 00:03:07,000 we are going to share a screen and we're
70 00:03:05,000 --> 00:03:11,200 going to talk about cracker
71 00:03:07,000 --> 00:03:14,640 3 so cracker 3 is a initiative that was
72 00:03:11,200 --> 00:03:16,879 released as a computer a general purpose
73 00:03:14,640 --> 00:03:19,599 computer vision model it's got many
74 00:03:16,879 --> 00:03:22,680 applications which are quite
75 00:03:19,599 --> 00:03:24,560 interesting but I'm going to talk about
76 00:03:22,680 --> 00:03:27,120 that a little bit later but I would like
77 00:03:24,560 --> 00:03:29,560 to share my screen and if you can't see
78 00:03:27,120 --> 00:03:32,439 this and you're listening I'll talk
79 00:03:29,560 --> 00:03:35,480 about it but if you are viewing it you
80 00:03:32,439 --> 00:03:38,760 can see what it is
81 00:03:35,480 --> 00:03:40,519 so basically Cod tracker
82 00:03:38,760 --> 00:03:43,360 3
83 00:03:40,519 --> 00:03:44,400 is sorry I accidentally hit my elbow on
84 00:03:43,360 --> 00:03:47,879 the on the
85 00:03:44,400 --> 00:03:54,879 mic hope that wasn't too
86 00:03:47,879 --> 00:03:58,120 disturbing so the craer 3 is this way of
87 00:03:54,879 --> 00:04:01,000 tracking and it basically pix pixelized
88 00:03:58,120 --> 00:04:03,640 or puts a point on each picked pixel and
89 00:04:01,000 --> 00:04:06,159 then identifies the object and then it
90 00:04:03,640 --> 00:04:08,599 tracks and predicts where the object is
91 00:04:06,159 --> 00:04:11,000 or will go which I think is pretty cool
92 00:04:08,599 --> 00:04:13,519 so you it will track but if if its
93 00:04:11,000 --> 00:04:18,440 vision gets obscured it will still be
94 00:04:13,519 --> 00:04:21,360 able to pick up on it because it knows
95 00:04:18,440 --> 00:04:24,960 or it would already have
96 00:04:21,360 --> 00:04:27,360 identified and matched the pixels of the
97 00:04:24,960 --> 00:04:30,880 object so when the object comes back
98 00:04:27,360 --> 00:04:32,600 into frame the object is instantly
99 00:04:30,880 --> 00:04:34,440 picked back up which isn't something
100 00:04:32,600 --> 00:04:38,360 that other
101 00:04:34,440 --> 00:04:40,800 tracking methodologies can do so it's
102 00:04:38,360 --> 00:04:43,360 it's better it's better and here's some
103 00:04:40,800 --> 00:04:44,919 examples um and it it Compares it
104 00:04:43,360 --> 00:04:49,160 Compares some
105 00:04:44,919 --> 00:04:50,360 others so this this biker goes up the
106 00:04:49,160 --> 00:04:53,120 hill and then
107 00:04:50,360 --> 00:04:55,720 he's his vision the vision of the video
108 00:04:53,120 --> 00:04:58,720 of the object which is a kid biking
109 00:04:55,720 --> 00:05:00,960 doing some biking it gets obscured a
110 00:04:58,720 --> 00:05:03,280 little bit and it messes up the other
111 00:05:00,960 --> 00:05:04,639 trackers and they get all out of whack
112 00:05:03,280 --> 00:05:07,560 while as
113 00:05:04,639 --> 00:05:09,919 cracker it stays right on target the
114 00:05:07,560 --> 00:05:11,919 whole time because it predicts where the
115 00:05:09,919 --> 00:05:15,000 the
116 00:05:11,919 --> 00:05:16,080 next movement of the object will go
117 00:05:15,000 --> 00:05:18,440 depending on
118 00:05:16,080 --> 00:05:21,639 the on the flight I would say like the
119 00:05:18,440 --> 00:05:23,479 flight path of like the pixels and so it
120 00:05:21,639 --> 00:05:25,680 picks picks right back up where where it
121 00:05:23,479 --> 00:05:28,160 needs to go which I think is pretty cool
122 00:05:25,680 --> 00:05:31,400 and here's one of like a rhino almost
123 00:05:28,160 --> 00:05:34,440 said hippopotamus I was like what
124 00:05:31,400 --> 00:05:38,080 it's got horns on it on its on its
125 00:05:34,440 --> 00:05:42,440 face that's I was talking about state of
126 00:05:38,080 --> 00:05:44,360 delirium uh that's pretty funny okay so
127 00:05:42,440 --> 00:05:46,440 there's this skateboarder and you can
128 00:05:44,360 --> 00:05:47,919 see the skateboarder is being tracked by
129 00:05:46,440 --> 00:05:49,919 by this little point
130 00:05:47,919 --> 00:05:52,319 tracker
131 00:05:49,919 --> 00:05:54,240 and it's kind of blocking out the other
132 00:05:52,319 --> 00:05:57,000 objects it'll pick them up but it's
133 00:05:54,240 --> 00:05:59,000 really focusing on the focal point and
134 00:05:57,000 --> 00:06:02,000 the focal point is this guy
135 00:05:59,000 --> 00:06:04,919 skateboarding or this rhino or this
136 00:06:02,000 --> 00:06:07,080 person walking a baby I don't know if
137 00:06:04,919 --> 00:06:08,080 that's a man or a woman with the shorter
138 00:06:07,080 --> 00:06:11,919 hair
139 00:06:08,080 --> 00:06:14,440 could could be man I don't know CU my
140 00:06:11,919 --> 00:06:17,720 face are blocked
141 00:06:14,440 --> 00:06:19,759 out but okay so there's a paper on this
142 00:06:17,720 --> 00:06:23,280 I'm going to review the paper but we
143 00:06:19,759 --> 00:06:25,759 also have a actual demo of a video so I
144 00:06:23,280 --> 00:06:27,960 am going to share this
145 00:06:25,759 --> 00:06:31,759 instead so let's watch the video
146 00:06:27,960 --> 00:06:33,599 together and this is a Rand man that I I
147 00:06:31,759 --> 00:06:35,120 guess I took a video of I was really
148 00:06:33,599 --> 00:06:37,160 trying to get a video of the city but I
149 00:06:35,120 --> 00:06:39,520 thought it was a pretty good example
150 00:06:37,160 --> 00:06:41,440 guys and it's it's pretty short it's
151 00:06:39,520 --> 00:06:43,360 pretty short but it's basically walking
152 00:06:41,440 --> 00:06:46,400 in the city of Japan there's a man in
153 00:06:43,360 --> 00:06:49,080 front of me and the vision gets obscured
154 00:06:46,400 --> 00:06:50,639 a couple times so I've already loaded it
155 00:06:49,080 --> 00:06:53,120 up so I'm submitting
156 00:06:50,639 --> 00:06:54,319 it and what I'm doing here so I'm
157 00:06:53,120 --> 00:06:57,479 loading in the
158 00:06:54,319 --> 00:06:59,599 cloud hold let me cough all right sorry
159 00:06:57,479 --> 00:07:04,039 yeah as I said I'm kind of sick so
160 00:06:59,599 --> 00:07:08,400 anyways so I'm loading it in it's at 20%
161 00:07:04,039 --> 00:07:10,400 or so so we've got another 80 and I had
162 00:07:08,400 --> 00:07:14,120 this loaded
163 00:07:10,400 --> 00:07:15,639 before and okay now it's done okay
164 00:07:14,120 --> 00:07:18,199 pretty cool so then I'm just going to
165 00:07:15,639 --> 00:07:20,160 hit track you could identify which Which
166 00:07:18,199 --> 00:07:22,720 object you want to track like by each
167 00:07:20,160 --> 00:07:25,120 frame but I'm not doing that because
168 00:07:22,720 --> 00:07:27,800 there's only one object that it it'll be
169 00:07:25,120 --> 00:07:30,720 pick it up so it's loading
170 00:07:27,800 --> 00:07:31,879 in so we saw the original video was that
171 00:07:30,720 --> 00:07:33,919 man
172 00:07:31,879 --> 00:07:35,599 walking and then it's going to put these
173 00:07:33,919 --> 00:07:39,960 points on the
174 00:07:35,599 --> 00:07:44,080 map for you and when I've tried this
175 00:07:39,960 --> 00:07:47,120 before it identified what I was trying
176 00:07:44,080 --> 00:07:48,759 to look at so it was fine so I didn't
177 00:07:47,120 --> 00:07:53,599 really see any edits that needed to be
178 00:07:48,759 --> 00:07:55,159 done it did a pretty good job in my
179 00:07:53,599 --> 00:07:58,919 opinion
180 00:07:55,159 --> 00:08:00,560 but hey I'm not I'm not looking I'm not
181 00:07:58,919 --> 00:08:04,879 looking for
182 00:08:00,560 --> 00:08:06,520 some scientific exact like oh wow like
183 00:08:04,879 --> 00:08:08,400 it missed the it missed the little
184 00:08:06,520 --> 00:08:11,039 shadow on the shoulder or something like
185 00:08:08,400 --> 00:08:12,919 I'm not I'm not zooming in and looking
186 00:08:11,039 --> 00:08:17,680 at like every pixel like oh it's three
187 00:08:12,919 --> 00:08:21,400 pixels off oh well but it's almost done
188 00:08:17,680 --> 00:08:23,000 it's at 80 or so percent oh here it is
189 00:08:21,400 --> 00:08:24,360 wow I I kind of jinx it every time
190 00:08:23,000 --> 00:08:26,120 because I'm telling it it's almost done
191 00:08:24,360 --> 00:08:30,120 so let me make this full
192 00:08:26,120 --> 00:08:32,519 screen so when they video originally
193 00:08:30,120 --> 00:08:35,000 starts see how it has all the little
194 00:08:32,519 --> 00:08:37,039 little dots on the map and then it
195 00:08:35,000 --> 00:08:39,120 finally it picks up to the guy and then
196 00:08:37,039 --> 00:08:42,200 it moves the dots off and then now it
197 00:08:39,120 --> 00:08:43,640 starts tracking him see them see how
198 00:08:42,200 --> 00:08:45,519 there aren't dots everywhere and the
199 00:08:43,640 --> 00:08:48,640 dots are being picked up by the object
200 00:08:45,519 --> 00:08:50,640 that it's tracking the man he's just I
201 00:08:48,640 --> 00:08:52,480 mean maybe he's working out his left arm
202 00:08:50,640 --> 00:08:55,320 but he's just carrying looks like he's
203 00:08:52,480 --> 00:08:57,360 carrying three bags in his left arm and
204 00:08:55,320 --> 00:09:01,200 then no bags in his right
205 00:08:57,360 --> 00:09:05,040 arm and life's all about balance you
206 00:09:01,200 --> 00:09:07,519 know but see how it's tracking again see
207 00:09:05,040 --> 00:09:10,160 it's got all these points and so the
208 00:09:07,519 --> 00:09:12,640 object left the focal point of the video
209 00:09:10,160 --> 00:09:14,640 twice and then right when it flipped
210 00:09:12,640 --> 00:09:17,040 back it was right there it was right
211 00:09:14,640 --> 00:09:19,720 there every time it's pretty
212 00:09:17,040 --> 00:09:21,920 cool and then it flips back boom it's
213 00:09:19,720 --> 00:09:26,279 it's already there it's already there
214 00:09:21,920 --> 00:09:29,839 it's crazy it's crazy I I don't know I
215 00:09:26,279 --> 00:09:29,839 think it's pretty sick
216 00:09:30,000 --> 00:09:34,880 all right stop sharing and then I'll go
217 00:09:32,120 --> 00:09:37,760 back to my actual
218 00:09:34,880 --> 00:09:42,160 video so okay
219 00:09:37,760 --> 00:09:45,440 so it does tracking like why is that why
220 00:09:42,160 --> 00:09:48,040 is that useful you know well there's a
221 00:09:45,440 --> 00:09:50,880 couple things that it could be doing and
222 00:09:48,040 --> 00:09:52,839 and maybe to get some interest and some
223 00:09:50,880 --> 00:09:55,800 Intrigue in the episode we could start
224 00:09:52,839 --> 00:10:00,000 with sports so you could use it for
225 00:09:55,800 --> 00:10:02,200 sports tracking for players or
226 00:10:00,000 --> 00:10:04,120 tracking on who is doing what and
227 00:10:02,200 --> 00:10:06,839 identifying them on the map and see what
228 00:10:04,120 --> 00:10:09,120 they're doing and then track them so it
229 00:10:06,839 --> 00:10:12,079 be pretty cool for sports analysis
230 00:10:09,120 --> 00:10:14,040 something that people may not like and a
231 00:10:12,079 --> 00:10:16,240 a lot of folks wouldn't like it is
232 00:10:14,040 --> 00:10:18,519 surveillance and security so government
233 00:10:16,240 --> 00:10:20,440 surveillance really comes to mind where
234 00:10:18,519 --> 00:10:22,160 you're tracking everybody and you're
235 00:10:20,440 --> 00:10:24,480 identifying them and then tracking where
236 00:10:22,160 --> 00:10:27,120 they go and then making sure they don't
237 00:10:24,480 --> 00:10:30,040 go to certain places or doing X things
238 00:10:27,120 --> 00:10:32,040 or whatever they're doing
239 00:10:30,040 --> 00:10:34,560 autonomous vehicles is a good a good
240 00:10:32,040 --> 00:10:36,680 shout so with autonomous vehicles you
241 00:10:34,560 --> 00:10:38,760 could use something like this to improve
242 00:10:36,680 --> 00:10:41,399 their tracking capabilities of tracking
243 00:10:38,760 --> 00:10:44,959 objects that go out of
244 00:10:41,399 --> 00:10:46,839 frame and then come in frame later like
245 00:10:44,959 --> 00:10:50,560 it goes out of the frame of vision of
246 00:10:46,839 --> 00:10:53,399 the vehicle but then where does it
247 00:10:50,560 --> 00:10:55,920 go like where would it go and then it
248 00:10:53,399 --> 00:10:59,880 picks it right back up so I think those
249 00:10:55,920 --> 00:11:02,839 are those are particular
250 00:10:59,880 --> 00:11:05,880 L interesting and then there's some for
251 00:11:02,839 --> 00:11:08,360 manufacturing so manufacturing like
252 00:11:05,880 --> 00:11:10,519 understanding where
253 00:11:08,360 --> 00:11:14,560 a I want say
254 00:11:10,519 --> 00:11:18,320 where like a robot
255 00:11:14,560 --> 00:11:20,959 arm needs to pick up an item and then
256 00:11:18,320 --> 00:11:25,360 pick it up and put it on
257 00:11:20,959 --> 00:11:28,040 something this tracking
258 00:11:25,360 --> 00:11:30,519 mechanism or improved tracking mechanism
259 00:11:28,040 --> 00:11:32,040 would be able to help
260 00:11:30,519 --> 00:11:34,800 pick up these items and know where
261 00:11:32,040 --> 00:11:37,519 they're at or identify items that
262 00:11:34,800 --> 00:11:38,680 potentially are defective and and track
263 00:11:37,519 --> 00:11:41,600 where they're
264 00:11:38,680 --> 00:11:43,040 going those kind of things and then
265 00:11:41,600 --> 00:11:44,279 there's there's healthc care I mean
266 00:11:43,040 --> 00:11:48,440 there there's there's probably
267 00:11:44,279 --> 00:11:53,480 definitely some ability to use this
268 00:11:48,440 --> 00:11:55,639 for like some sort of I would think like
269 00:11:53,480 --> 00:11:57,639 healthc Care on the on the lines of
270 00:11:55,639 --> 00:12:01,880 Rehabilitation like
271 00:11:57,639 --> 00:12:04,279 tracking the movement of of a body like
272 00:12:01,880 --> 00:12:05,959 so I would think of it if if you were
273 00:12:04,279 --> 00:12:07,839 rehabilitating your shoulder and you
274 00:12:05,959 --> 00:12:11,000 were doing shoulder rotations or
275 00:12:07,839 --> 00:12:12,800 something like that if you had a Tracker
276 00:12:11,000 --> 00:12:15,000 and it was tracking your arm and like
277 00:12:12,800 --> 00:12:17,519 where it's supposed to go and then what
278 00:12:15,000 --> 00:12:21,560 you're doing and seeing on like how you
279 00:12:17,519 --> 00:12:21,560 correct it and
280 00:12:21,720 --> 00:12:26,360 providing I think it'd be pretty cool to
281 00:12:24,199 --> 00:12:29,399 provide like if you were showing someone
282 00:12:26,360 --> 00:12:32,000 like your range of movement and then
283 00:12:29,399 --> 00:12:34,240 you're with a doctor and then you're
284 00:12:32,000 --> 00:12:36,639 tracking the movement of what you have
285 00:12:34,240 --> 00:12:38,920 your your Mobility on the mobility and
286 00:12:36,639 --> 00:12:41,720 then it it will show you what you should
287 00:12:38,920 --> 00:12:45,760 should look like with your tracking and
288 00:12:41,720 --> 00:12:49,199 how you would would look if you had the
289 00:12:45,760 --> 00:12:50,279 right not access but maybe the right
290 00:12:49,199 --> 00:12:52,480 features
291 00:12:50,279 --> 00:12:55,680 unlocked like you're some kind of
292 00:12:52,480 --> 00:12:59,560 character where maybe you had
293 00:12:55,680 --> 00:13:01,560 your all your Rehabilitation done or you
294 00:12:59,560 --> 00:13:05,560 had your surgery or whatever like this
295 00:13:01,560 --> 00:13:08,040 is what it would look like and or
296 00:13:05,560 --> 00:13:10,440 correcting some movement like hey you're
297 00:13:08,040 --> 00:13:12,600 walking this way really should be
298 00:13:10,440 --> 00:13:14,519 walking this way or like
299 00:13:12,600 --> 00:13:18,399 running yeah all that stuff would be
300 00:13:14,519 --> 00:13:20,399 pretty cool in my
301 00:13:18,399 --> 00:13:22,800 opinion but next thing that we're going
302 00:13:20,399 --> 00:13:24,519 to talk about here is this movie genin
303 00:13:22,800 --> 00:13:28,720 or transitioning and this is just going
304 00:13:24,519 --> 00:13:31,600 to be a a list of of things that meta
305 00:13:28,720 --> 00:13:34,480 shipped they've shipped so much stuff so
306 00:13:31,600 --> 00:13:37,120 we're transitioning over to movie gen
307 00:13:34,480 --> 00:13:39,320 movie gen is pretty cool so if you are
308 00:13:37,120 --> 00:13:43,199 watching you'll see what I'm doing and
309 00:13:39,320 --> 00:13:46,440 if you're on audio then you won't
310 00:13:43,199 --> 00:13:48,959 but I'll explain it as always so metam
311 00:13:46,440 --> 00:13:51,199 movi Jin is
312 00:13:48,959 --> 00:13:55,000 a they're saying one of the most
313 00:13:51,199 --> 00:13:57,279 advanced foundational media models in
314 00:13:55,000 --> 00:14:00,519 existence they haven't launched it yet
315 00:13:57,279 --> 00:14:02,800 they're still in
316 00:14:00,519 --> 00:14:05,000 sorry I had to cough they're still in
317 00:14:02,800 --> 00:14:07,360 like Creator Creator Partnerships what
318 00:14:05,000 --> 00:14:11,360 they call them but basically movie Jin
319 00:14:07,360 --> 00:14:15,959 allows you to not only generate movies
320 00:14:11,360 --> 00:14:19,639 or videos via to text it also allows you
321 00:14:15,959 --> 00:14:22,639 to this is an example so this one
322 00:14:19,639 --> 00:14:24,160 is uh text input a girl running across
323 00:14:22,639 --> 00:14:26,240 the beach and holding a kite she's
324 00:14:24,160 --> 00:14:30,800 wearing jean shorts a yellow shirt and
325 00:14:26,240 --> 00:14:32,800 the the sun is shining down
326 00:14:30,800 --> 00:14:34,920 I mean it's pretty it's a pretty good
327 00:14:32,800 --> 00:14:38,279 video like you could see the like the
328 00:14:34,920 --> 00:14:41,120 steps in the sand and the sand is
329 00:14:38,279 --> 00:14:42,639 impactful like not impactful but like
330 00:14:41,120 --> 00:14:45,000 when she's running in the
331 00:14:42,639 --> 00:14:49,079 sand
332 00:14:45,000 --> 00:14:51,759 her her feet are hitting the sand and
333 00:14:49,079 --> 00:14:55,480 they're leaving imprints but the closer
334 00:14:51,759 --> 00:14:59,560 she gets to the water the more
335 00:14:55,480 --> 00:15:02,240 damp the sand is so the not only as the
336 00:14:59,560 --> 00:15:03,880 sand start like sticking up when the
337 00:15:02,240 --> 00:15:05,880 sand gets it when it when you're
338 00:15:03,880 --> 00:15:10,759 approaching the sand when it's more wet
339 00:15:05,880 --> 00:15:13,920 the the imprints of her foot are more
340 00:15:10,759 --> 00:15:15,519 pronounced and the kite is flying in the
341 00:15:13,920 --> 00:15:17,800 sky and she's running and her hands
342 00:15:15,519 --> 00:15:19,680 don't look weird and you can't really
343 00:15:17,800 --> 00:15:22,160 see her hands as well but her her skin
344 00:15:19,680 --> 00:15:24,279 complexion on her leg looks a little
345 00:15:22,160 --> 00:15:27,120 like odd but that might be just cuz like
346 00:15:24,279 --> 00:15:30,920 baby's legs look weird you know I don't
347 00:15:27,120 --> 00:15:34,639 know but this is something that's pretty
348 00:15:30,920 --> 00:15:36,480 cool text of a woman's in the grass
349 00:15:34,639 --> 00:15:38,519 pumpkin patch wearing a scarf and
350 00:15:36,480 --> 00:15:39,639 holding a cup and the backgrounds filled
351 00:15:38,519 --> 00:15:42,360 with
352 00:15:39,639 --> 00:15:44,720 Republicans but the interesting here
353 00:15:42,360 --> 00:15:46,720 thing here is that you could not only
354 00:15:44,720 --> 00:15:49,440 put yourself personalized videos of
355 00:15:46,720 --> 00:15:51,199 yourself or things you can you can
356 00:15:49,440 --> 00:15:54,959 provide a face
357 00:15:51,199 --> 00:15:57,639 shot of your friend or of
358 00:15:54,959 --> 00:16:02,440 yourself and you can make a video from
359 00:15:57,639 --> 00:16:06,040 it which I think is pretty neat so
360 00:16:02,440 --> 00:16:08,160 look these are more videos from text
361 00:16:06,040 --> 00:16:11,880 generation this is
362 00:16:08,160 --> 00:16:13,880 some sloth this guy's got fire and then
363 00:16:11,880 --> 00:16:15,800 this is pretty cool so you could have an
364 00:16:13,880 --> 00:16:17,519 original video that you have so maybe
365 00:16:15,800 --> 00:16:19,399 you're doing some this example is like
366 00:16:17,519 --> 00:16:23,240 running or maybe you're with your
367 00:16:19,399 --> 00:16:25,880 friends and you're you're at a
368 00:16:23,240 --> 00:16:29,319 party like a house
369 00:16:25,880 --> 00:16:32,600 party and maybe you want to change it to
370 00:16:29,319 --> 00:16:35,240 a nightclub or some kind of Sky Rooftop
371 00:16:32,600 --> 00:16:38,600 Bar you can do that so in this original
372 00:16:35,240 --> 00:16:42,079 video the this person's running and then
373 00:16:38,600 --> 00:16:45,199 it says add blue pom poms to his hands
374 00:16:42,079 --> 00:16:48,519 and now he's running with pom poms and
375 00:16:45,199 --> 00:16:52,600 then this other one says
376 00:16:48,519 --> 00:16:54,319 turn turn it into a cais desert and it
377 00:16:52,600 --> 00:16:56,040 does and then the next one says replace
378 00:16:54,319 --> 00:16:58,600 the running clothes with an inflatable
379 00:16:56,040 --> 00:16:59,920 dinosaur costume and it looks pretty
380 00:16:58,600 --> 00:17:03,560 real
381 00:16:59,920 --> 00:17:04,760 like I don't know it's crazy this one
382 00:17:03,560 --> 00:17:06,240 I'm sorry I don't know if this coughs
383 00:17:04,760 --> 00:17:08,039 are getting picked up but I've been
384 00:17:06,240 --> 00:17:12,439 having a cough this one has this
385 00:17:08,039 --> 00:17:15,360 original video is some chick in like a
386 00:17:12,439 --> 00:17:16,919 room doing some weird hand movement
387 00:17:15,360 --> 00:17:18,880 stuff with her hands I think she's doing
388 00:17:16,919 --> 00:17:23,199 like VR oh she's doing VR
389 00:17:18,880 --> 00:17:25,360 stuff and then it changes the VR headset
390 00:17:23,199 --> 00:17:27,880 to change the VR headset to
391 00:17:25,360 --> 00:17:30,120 binoculars and then the next one says uh
392 00:17:27,880 --> 00:17:31,840 remove the VR headset I don't know how
393 00:17:30,120 --> 00:17:33,559 it generates the eyes and the rest of
394 00:17:31,840 --> 00:17:36,000 her face cuz all that stuff's covered up
395 00:17:33,559 --> 00:17:38,200 but that's pretty
396 00:17:36,000 --> 00:17:41,200 incredible the next one says add Bubbles
397 00:17:38,200 --> 00:17:42,600 and futuristic effect pretty sick and
398 00:17:41,200 --> 00:17:45,360 then this one's changing the color of
399 00:17:42,600 --> 00:17:47,799 this guy's t-shirt and the
400 00:17:45,360 --> 00:17:50,200 background stuff's incredible I I don't
401 00:17:47,799 --> 00:17:53,120 I don't know how they're doing it it it
402 00:17:50,200 --> 00:17:54,679 looks so real like I don't I don't know
403 00:17:53,120 --> 00:17:57,000 what's going on here this is some black
404 00:17:54,679 --> 00:17:59,240 magic stuff like I've seen some cool
405 00:17:57,000 --> 00:18:02,960 stuff before but when I saw this I was
406 00:17:59,240 --> 00:18:06,400 really wild really wild I was like
407 00:18:02,960 --> 00:18:07,799 wow that's pretty neat and this is going
408 00:18:06,400 --> 00:18:11,240 to be you know what what is the
409 00:18:07,799 --> 00:18:12,360 practical application here well maybe
410 00:18:11,240 --> 00:18:14,799 you don't like your shirt that you're
411 00:18:12,360 --> 00:18:16,400 wearing in that video change the shirt
412 00:18:14,799 --> 00:18:18,080 or change the background like this is
413 00:18:16,400 --> 00:18:23,400 original video warming up for a run it
414 00:18:18,080 --> 00:18:25,120 looks like and this man has this changed
415 00:18:23,400 --> 00:18:27,520 the shirt in the background changed the
416 00:18:25,120 --> 00:18:30,200 background to an outdoor stadium and
417 00:18:27,520 --> 00:18:31,480 he's warming up with a blue shirt
418 00:18:30,200 --> 00:18:32,880 I guess they didn't tell to change the
419 00:18:31,480 --> 00:18:34,400 shirt change the background to an
420 00:18:32,880 --> 00:18:37,720 outdoor stadium I guess it Chang the
421 00:18:34,400 --> 00:18:39,600 shirt on its own weird that's a bit odd
422 00:18:37,720 --> 00:18:42,600 I don't know why I did that
423 00:18:39,600 --> 00:18:44,000 but I don't know so then the next thing
424 00:18:42,600 --> 00:18:45,520 that's pretty cool is you can make
425 00:18:44,000 --> 00:18:48,159 personalized video so you can take a
426 00:18:45,520 --> 00:18:51,080 selfie of yourself and then it'll make a
427 00:18:48,159 --> 00:18:53,000 video of you doing something like this
428 00:18:51,080 --> 00:18:55,039 guy took a video of himself and then he
429 00:18:53,000 --> 00:18:59,039 made a he became a
430 00:18:55,039 --> 00:19:02,240 scientist um this woman
431 00:18:59,039 --> 00:19:06,159 who just had a selfie with I don't know
432 00:19:02,240 --> 00:19:06,159 I would say bad lighting it's
433 00:19:06,320 --> 00:19:09,720 Overexposed has herself in a painting
434 00:19:08,440 --> 00:19:13,120 video and she's
435 00:19:09,720 --> 00:19:16,840 painting and the hands look great lots
436 00:19:13,120 --> 00:19:18,760 of dexterity there oh man this smile is
437 00:19:16,840 --> 00:19:22,559 odd but
438 00:19:18,760 --> 00:19:24,799 it I mean it looks a little exaggerated
439 00:19:22,559 --> 00:19:29,640 but I think it's the angle that the guy
440 00:19:24,799 --> 00:19:32,039 like his smile to what he gave is pretty
441 00:19:29,640 --> 00:19:33,440 close in certain spot like in the
442 00:19:32,039 --> 00:19:34,960 beginning part of the video it's like
443 00:19:33,440 --> 00:19:38,559 it's really extended but I think it's
444 00:19:34,960 --> 00:19:41,919 the angle of which like he's like s his
445 00:19:38,559 --> 00:19:43,720 photo but yeah this is pretty cool stuff
446 00:19:41,919 --> 00:19:45,159 pretty cool stuff and then you can add a
447 00:19:43,720 --> 00:19:48,200 sound like create a effects and
448 00:19:45,159 --> 00:19:51,240 soundtracks for your videos
449 00:19:48,200 --> 00:19:56,240 so not only
450 00:19:51,240 --> 00:20:00,120 can you create a video from text you
451 00:19:56,240 --> 00:20:04,440 could also personalize the video and put
452 00:20:00,120 --> 00:20:07,360 objects or Andor things you can edit the
453 00:20:04,440 --> 00:20:10,120 video with text completely change the
454 00:20:07,360 --> 00:20:12,240 background or objects in the video you
455 00:20:10,120 --> 00:20:15,039 can personalize it putting putting in
456 00:20:12,240 --> 00:20:17,039 yourself and then you can add sound like
457 00:20:15,039 --> 00:20:18,500 then you can create soundtracks and
458 00:20:17,039 --> 00:20:20,159 effects for your
459 00:20:18,500 --> 00:20:23,799 [Music]
460 00:20:20,159 --> 00:20:26,280 video crazy this isn't released yet
461 00:20:23,799 --> 00:20:26,280 they're
462 00:20:26,360 --> 00:20:33,000 only they are only
463 00:20:30,200 --> 00:20:36,120 allowing Creator Partners to use these
464 00:20:33,000 --> 00:20:37,760 videos or this movie genin thing which
465 00:20:36,120 --> 00:20:42,280 is a bummer because I really wanted to
466 00:20:37,760 --> 00:20:43,880 use it but whatever it is what it is so
467 00:20:42,280 --> 00:20:48,520 that that's movie
468 00:20:43,880 --> 00:20:51,120 gen so movie J's pretty cool um I don't
469 00:20:48,520 --> 00:20:52,480 know wow I said um that's probably the
470 00:20:51,120 --> 00:20:56,159 first time in a long
471 00:20:52,480 --> 00:20:58,919 time anyways
472 00:20:56,159 --> 00:21:00,880 so there's this other object that that
473 00:20:58,919 --> 00:21:04,520 they released called The
474 00:21:00,880 --> 00:21:06,840 Meta Spirit LM which is originally I
475 00:21:04,520 --> 00:21:09,799 thought and was explained to me as like
476 00:21:06,840 --> 00:21:11,760 a Google notebook LM competitor but open
477 00:21:09,799 --> 00:21:13,320 source but really it's a different
478 00:21:11,760 --> 00:21:14,640 approach where it's it's less about
479 00:21:13,320 --> 00:21:17,360 standardizing
480 00:21:14,640 --> 00:21:20,960 information and more about just having
481 00:21:17,360 --> 00:21:23,640 like an assistant nearby that you could
482 00:21:20,960 --> 00:21:28,039 proc it's multimodal it could
483 00:21:23,640 --> 00:21:31,480 process text audio and just having an
484 00:21:28,039 --> 00:21:34,679 assistant nearby like on your computer
485 00:21:31,480 --> 00:21:35,919 locally which is you know a different
486 00:21:34,679 --> 00:21:39,520 approach
487 00:21:35,919 --> 00:21:42,159 than notebook LM by Google because
488 00:21:39,520 --> 00:21:44,840 that's clearly a summarization and
489 00:21:42,159 --> 00:21:48,200 educational tool
490 00:21:44,840 --> 00:21:51,200 whereas meta Spirit
491 00:21:48,200 --> 00:21:52,640 LM is Ian it's just overall a different
492 00:21:51,200 --> 00:21:54,760 approach like it's like having something
493 00:21:52,640 --> 00:21:57,520 on your computer that you know can
494 00:21:54,760 --> 00:22:00,559 process text and
495 00:21:57,520 --> 00:22:03,120 audio notebook Elum can do that and then
496 00:22:00,559 --> 00:22:05,720 it could respond back in audio the voice
497 00:22:03,120 --> 00:22:09,080 is a little odd and
498 00:22:05,720 --> 00:22:11,840 very I would say somebody said in the
499 00:22:09,080 --> 00:22:15,400 comments in the video that meta posted
500 00:22:11,840 --> 00:22:18,279 like this is what my sleep paralysis
501 00:22:15,400 --> 00:22:22,039 demon would sound like so the like it
502 00:22:18,279 --> 00:22:24,480 it's not the best but I think overall
503 00:22:22,039 --> 00:22:30,520 it's something that meta ship that
504 00:22:24,480 --> 00:22:30,520 people can use and it is
505 00:22:30,559 --> 00:22:36,919 interesting in its own
506 00:22:32,440 --> 00:22:39,200 regard and it is open
507 00:22:36,919 --> 00:22:42,159 sourced this one it doesn't allow
508 00:22:39,200 --> 00:22:44,320 commercial use but that's good enough I
509 00:22:42,159 --> 00:22:45,919 mean open sourcing the community and and
510 00:22:44,320 --> 00:22:48,480 letting people play with it and figure
511 00:22:45,919 --> 00:22:51,840 out the best way to do stuff you get a
512 00:22:48,480 --> 00:22:58,600 lot more feedback when you do
513 00:22:51,840 --> 00:23:01,320 so okay So Meta robotics advancements
514 00:22:58,600 --> 00:23:03,520 I think before I talk about stuff
515 00:23:01,320 --> 00:23:06,880 relating to this I think the best thing
516 00:23:03,520 --> 00:23:10,799 to do is to share my screen and watch a
517 00:23:06,880 --> 00:23:12,600 two-minute video this is it right here
518 00:23:10,799 --> 00:23:14,200 cool so I've already watched it because
519 00:23:12,600 --> 00:23:17,240 I wanted to test the video and thank
520 00:23:14,200 --> 00:23:19,919 goodness I did because originally it was
521 00:23:17,240 --> 00:23:21,400 coming in so loud like I wanted to make
522 00:23:19,919 --> 00:23:22,720 sure when I played this video that the
523 00:23:21,400 --> 00:23:26,120 audio was coming
524 00:23:22,720 --> 00:23:28,520 through and pretty sure it did but I
525 00:23:26,120 --> 00:23:31,240 couldn't hurt the test so when I was
526 00:23:28,520 --> 00:23:33,279 doing it I was like listening to the
527 00:23:31,240 --> 00:23:37,600 after the video after the the video
528 00:23:33,279 --> 00:23:40,200 recording test and oh my it was so loud
529 00:23:37,600 --> 00:23:43,760 so I turned it down and made it a little
530 00:23:40,200 --> 00:23:46,320 bit less loud so hopefully that is not
531 00:23:43,760 --> 00:23:48,200 too loud but let's watch the video and
532 00:23:46,320 --> 00:23:50,559 it will give us a pretty good
533 00:23:48,200 --> 00:23:52,039 understanding of what this is and then I
534 00:23:50,559 --> 00:23:56,679 could talk about it maybe make it full
535 00:23:52,039 --> 00:23:59,559 screen yeah insane right I was think the
536 00:23:56,679 --> 00:24:01,919 same thing crazy
537 00:23:59,559 --> 00:24:04,679 one thing I think is
538 00:24:01,919 --> 00:24:08,240 more I would
539 00:24:04,679 --> 00:24:10,520 say I think it's hard to understand that
540 00:24:08,240 --> 00:24:14,000 these independent teams launched all
541 00:24:10,520 --> 00:24:18,559 these things in the last five weeks or
542 00:24:14,000 --> 00:24:20,200 so crazy that they oh maybe not five
543 00:24:18,559 --> 00:24:24,400 eight weeks say eight
544 00:24:20,200 --> 00:24:29,919 weeks they've launched all this stuff so
545 00:24:24,400 --> 00:24:32,360 soon within each other and they're all
546 00:24:29,919 --> 00:24:32,360 I would
547 00:24:32,679 --> 00:24:37,679 say incredible
548 00:24:35,200 --> 00:24:40,039 projects and they're all coming with a
549 00:24:37,679 --> 00:24:41,960 90 page research paper which I'll need
550 00:24:40,039 --> 00:24:44,799 to read them and I'll talk about them in
551 00:24:41,960 --> 00:24:47,320 the episode so the LA next couple weeks
552 00:24:44,799 --> 00:24:49,039 while I'm working on some special
553 00:24:47,320 --> 00:24:53,480 episodes with guests to get that all
554 00:24:49,039 --> 00:24:56,159 squared away I'll be talking about meta
555 00:24:53,480 --> 00:24:59,240 research papers and I'll probably start
556 00:24:56,159 --> 00:25:00,559 with the robotics one cuz digitizing
557 00:24:59,240 --> 00:25:05,320 seems very
558 00:25:00,559 --> 00:25:09,120 interesting and overall super cool super
559 00:25:05,320 --> 00:25:12,760 cool uh stop sharing like I I don't know
560 00:25:09,120 --> 00:25:15,559 like how how are they shipping this much
561 00:25:12,760 --> 00:25:18,279 stuff is is what I'm having a hard time
562 00:25:15,559 --> 00:25:19,880 understanding like this whole like meta
563 00:25:18,279 --> 00:25:23,080 robotics
564 00:25:19,880 --> 00:25:26,440 advancement and the just different
565 00:25:23,080 --> 00:25:29,000 Integrations with the digitizing the
566 00:25:26,440 --> 00:25:31,039 touch with the sensors and and having
567 00:25:29,000 --> 00:25:35,200 the 460,000
568 00:25:31,039 --> 00:25:40,120 images and then being 90 like having a
569 00:25:35,200 --> 00:25:44,559 95% Improvement to or did they say 95 or
570 00:25:40,120 --> 00:25:46,840 90% um I think that they
571 00:25:44,559 --> 00:25:49,840 said we literally just watched it
572 00:25:46,840 --> 00:25:53,320 together let's see I think it's
573 00:25:49,840 --> 00:25:56,120 90 90 that's sad I just we literally
574 00:25:53,320 --> 00:25:58,960 just watched it
575 00:25:56,120 --> 00:26:00,919 95 all right all right right so yeah I
576 00:25:58,960 --> 00:26:02,399 was right I was doubting myself it just
577 00:26:00,919 --> 00:26:05,399 I sometimes I doubt myself with these
578 00:26:02,399 --> 00:26:07,559 numbers because it it's such a high
579 00:26:05,399 --> 00:26:10,080 number to say to stay out but we found
580 00:26:07,559 --> 00:26:13,360 that Spar outperforms task and sensor
581 00:26:10,080 --> 00:26:15,559 specific models by an average of over
582 00:26:13,360 --> 00:26:18,559 95% by this
583 00:26:15,559 --> 00:26:22,279 Benchmark
584 00:26:18,559 --> 00:26:25,120 so not only are they shipping all this
585 00:26:22,279 --> 00:26:26,399 stuff they are absolutely just
586 00:26:25,120 --> 00:26:27,960 dominating in the things that they're
587 00:26:26,399 --> 00:26:30,039 releasing like they're dominating with
588 00:26:27,960 --> 00:26:31,720 the co tracker 3 they're dominating with
589 00:26:30,039 --> 00:26:34,039 Spar and these other
590 00:26:31,720 --> 00:26:38,960 things and it it's just not even close
591 00:26:34,039 --> 00:26:41,399 any 95% better than the other models
592 00:26:38,960 --> 00:26:43,200 like and then the movie genin thing like
593 00:26:41,399 --> 00:26:46,080 the movie genin thing
594 00:26:43,200 --> 00:26:47,279 looks just flabber I'm flabbergasted
595 00:26:46,080 --> 00:26:48,799 with all the stuff that they're that
596 00:26:47,279 --> 00:26:52,880 they're launching I'm really excited to
597 00:26:48,799 --> 00:26:55,240 read these papers and see what knowledge
598 00:26:52,880 --> 00:26:56,640 I can gleam and share with the audience
599 00:26:55,240 --> 00:26:59,120 and everyone that listens to the
600 00:26:56,640 --> 00:27:00,279 episodes every week appreciate you and I
601 00:26:59,120 --> 00:27:02,480 hope that you find this stuff
602 00:27:00,279 --> 00:27:04,279 interesting because it improves me quite
603 00:27:02,480 --> 00:27:08,720 a bit and
604 00:27:04,279 --> 00:27:10,919 my understanding of technology and
605 00:27:08,720 --> 00:27:13,960 hopefully a mix of insurance and finance
606 00:27:10,919 --> 00:27:16,600 and hopefully in the future all these
607 00:27:13,960 --> 00:27:20,679 things intertwine and become one and
608 00:27:16,600 --> 00:27:25,080 allow me to elevate myself and
609 00:27:20,679 --> 00:27:29,000 others but anyways so that was meta Spar
610 00:27:25,080 --> 00:27:32,799 Spar and Spar means touch in a language
611 00:27:29,000 --> 00:27:36,320 called like s sanset sanset which is a
612 00:27:32,799 --> 00:27:39,360 language back in the Bronze Age
613 00:27:36,320 --> 00:27:42,720 and I don't think many people speak it
614 00:27:39,360 --> 00:27:45,600 today but that's what that Spar means it
615 00:27:42,720 --> 00:27:47,159 means touch but it sounds pretty cool
616 00:27:45,600 --> 00:27:49,440 though it's like wow that's all Nifty
617 00:27:47,159 --> 00:27:50,960 but really it literally means touch so
618 00:27:49,440 --> 00:27:55,080 maybe that's something to pull out of
619 00:27:50,960 --> 00:27:56,720 the product rule book or or road map is
620 00:27:55,080 --> 00:27:59,600 when next time I come out with a product
621 00:27:56,720 --> 00:28:03,039 or XYZ
622 00:27:59,600 --> 00:28:05,200 you just pull up your your foreign
623 00:28:03,039 --> 00:28:09,120 language index and just okay what is
624 00:28:05,200 --> 00:28:10,960 this all right perfect okay that's what
625 00:28:09,120 --> 00:28:13,880 that means all right let's call it that
626 00:28:10,960 --> 00:28:15,559 it's easy to say it's easy to spell
627 00:28:13,880 --> 00:28:17,600 let's go let's do it and I didn't
628 00:28:15,559 --> 00:28:21,399 mention it earlier but the combination
629 00:28:17,600 --> 00:28:23,720 of this sparsh and the stuff that
630 00:28:21,399 --> 00:28:27,000 they're doing with that the digit
631 00:28:23,720 --> 00:28:29,720 360 the Digiplex all these things and
632 00:28:27,000 --> 00:28:32,559 creating not only the technology open
633 00:28:29,720 --> 00:28:35,919 sourcing to the community but then also
634 00:28:32,559 --> 00:28:39,519 creating a way for it to
635 00:28:35,919 --> 00:28:42,360 be easily integrated with having a
636 00:28:39,519 --> 00:28:45,760 platform for these sensor connections
637 00:28:42,360 --> 00:28:48,080 and also partnering with these other
638 00:28:45,760 --> 00:28:49,640 companies to help build out these little
639 00:28:48,080 --> 00:28:52,840 digits that they're going to be used for
640 00:28:49,640 --> 00:28:56,720 the fingers and the sensors and all that
641 00:28:52,840 --> 00:28:59,240 stuff I think that
642 00:28:56,720 --> 00:29:00,559 it's it's going to change the game like
643 00:28:59,240 --> 00:29:03,279 5 years from
644 00:29:00,559 --> 00:29:06,440 now it might be the
645 00:29:03,279 --> 00:29:09,440 standard like I I don't I don't know
646 00:29:06,440 --> 00:29:11,960 other how how else someone else would
647 00:29:09,440 --> 00:29:14,320 compete with unless they came out with a
648 00:29:11,960 --> 00:29:16,880 superior product but is it is it open
649 00:29:14,320 --> 00:29:20,080 source like when you open source a
650 00:29:16,880 --> 00:29:21,240 superior product than everyone else I
651 00:29:20,080 --> 00:29:24,440 don't know
652 00:29:21,240 --> 00:29:28,880 how someone can get away with not open
653 00:29:24,440 --> 00:29:31,159 sourcing like that's like saying
654 00:29:28,880 --> 00:29:35,960 okay you can
655 00:29:31,159 --> 00:29:38,360 buy my hot dogs right and you could also
656 00:29:35,960 --> 00:29:41,080 make them yourself here's the recipe if
657 00:29:38,360 --> 00:29:42,919 you like it a lot and then there's
658 00:29:41,080 --> 00:29:44,399 someone else selling hot dogs but not
659 00:29:42,919 --> 00:29:46,720 only do you have to sell you have to buy
660 00:29:44,399 --> 00:29:50,559 the hot dog you also have some kind of
661 00:29:46,720 --> 00:29:52,480 membership fee like to get to get your
662 00:29:50,559 --> 00:29:55,760 your hot dog you've got to
663 00:29:52,480 --> 00:29:58,240 have Sam's Hot Dog membership and then
664 00:29:55,760 --> 00:30:01,000 you got to buy the hot dog
665 00:29:58,240 --> 00:30:03,159 when this open source hot dog is so much
666 00:30:01,000 --> 00:30:05,720 better and you can make it yourself at
667 00:30:03,159 --> 00:30:08,000 home and you have all the knowledge of
668 00:30:05,720 --> 00:30:09,440 what goes in it like why would you use
669 00:30:08,000 --> 00:30:12,919 the other
670 00:30:09,440 --> 00:30:16,440 one so it really burns the alternative
671 00:30:12,919 --> 00:30:19,360 the alternative is just done and and
672 00:30:16,440 --> 00:30:22,640 that's really the goal of Mark
673 00:30:19,360 --> 00:30:25,519 Zuckerberg and meta he is really pushing
674 00:30:22,640 --> 00:30:28,399 for these generational technologies that
675 00:30:25,519 --> 00:30:30,840 are going to transform
676 00:30:28,399 --> 00:30:34,640 this not
677 00:30:30,840 --> 00:30:37,159 only the industry but Society in
678 00:30:34,640 --> 00:30:38,720 general the demand is open source he
679 00:30:37,159 --> 00:30:41,440 wants it to be open
680 00:30:38,720 --> 00:30:44,480 source he doesn't want to have a closed
681 00:30:41,440 --> 00:30:46,880 closed system model like phones or
682 00:30:44,480 --> 00:30:48,880 computers he wants it to be open source
683 00:30:46,880 --> 00:30:50,080 and he wants open source to win so he's
684 00:30:48,880 --> 00:30:51,519 like yeah I'm fine with just open
685 00:30:50,080 --> 00:30:54,080 sourcing everything like who cares if it
686 00:30:51,519 --> 00:30:56,559 makes any money we don't need we don't
687 00:30:54,080 --> 00:30:59,159 need the money we don't need the product
688 00:30:56,559 --> 00:31:03,279 we need to and invest in these things to
689 00:30:59,159 --> 00:31:06,200 become Advanced and to protect ourselves
690 00:31:03,279 --> 00:31:08,000 against you know sophisticated Nations
691 00:31:06,200 --> 00:31:09,159 attacking our systems so we're going to
692 00:31:08,000 --> 00:31:11,480 have to have a really advanced
693 00:31:09,159 --> 00:31:13,960 technology and this in this instance I
694 00:31:11,480 --> 00:31:15,720 was talking about llms and Ai and then
695 00:31:13,960 --> 00:31:18,120 this other stuff with Advanced Robotics
696 00:31:15,720 --> 00:31:20,840 and the co- tracker the
697 00:31:18,120 --> 00:31:24,000 original adaptation of these things
698 00:31:20,840 --> 00:31:27,679 possibly wasn't from VR but a lot of
699 00:31:24,000 --> 00:31:30,760 these applications are related to VR and
700 00:31:27,679 --> 00:31:33,240 it was something that Mark talked about
701 00:31:30,760 --> 00:31:36,399 before with when he was getting a lot of
702 00:31:33,240 --> 00:31:39,080 crap for pushing into VR and really
703 00:31:36,399 --> 00:31:40,679 leaning into the metaverse and all this
704 00:31:39,080 --> 00:31:44,120 stuff
705 00:31:40,679 --> 00:31:48,480 was the technologies that we will
706 00:31:44,120 --> 00:31:49,360 create from investing and optimizing
707 00:31:48,480 --> 00:31:52,360 this
708 00:31:49,360 --> 00:31:54,360 technology are going to be incredible
709 00:31:52,360 --> 00:31:57,399 and if you guys can't see that then I
710 00:31:54,360 --> 00:31:59,240 can't help you but otherwise we're going
711 00:31:57,399 --> 00:32:01,799 to be moving forward with
712 00:31:59,240 --> 00:32:04,080 this and you can just see you can just
713 00:32:01,799 --> 00:32:08,120 see some of these things that a lot of
714 00:32:04,080 --> 00:32:09,639 these research projects fold over to VR
715 00:32:08,120 --> 00:32:12,639 fold over to the other things that
716 00:32:09,639 --> 00:32:14,360 they're doing like the cracker 3 I
717 00:32:12,639 --> 00:32:16,480 didn't talk about it I think I forgot to
718 00:32:14,360 --> 00:32:19,960 but gaming like you're
719 00:32:16,480 --> 00:32:23,200 gaming like for VR Gaming tracking your
720 00:32:19,960 --> 00:32:27,080 hands maybe other players in the room
721 00:32:23,200 --> 00:32:30,960 other things like that huge huge and for
722 00:32:27,080 --> 00:32:34,679 this thing right here The Spar and the
723 00:32:30,960 --> 00:32:36,880 the Digiplex and the the touch 360 or
724 00:32:34,679 --> 00:32:40,880 meta digit
725 00:32:36,880 --> 00:32:44,960 360 those items would help in VR like
726 00:32:40,880 --> 00:32:48,240 you could you could scan in textures and
727 00:32:44,960 --> 00:32:50,399 then have these sensors you could scan
728 00:32:48,240 --> 00:32:52,320 in the the textures in
729 00:32:50,399 --> 00:32:54,080 VR
730 00:32:52,320 --> 00:32:57,080 and
731 00:32:54,080 --> 00:32:58,559 then Som simulate these textures back to
732 00:32:57,080 --> 00:33:02,159 the human
733 00:32:58,559 --> 00:33:04,240 with sensors I don't know I mean that
734 00:33:02,159 --> 00:33:05,799 would get you get pretty get pretty wild
735 00:33:04,240 --> 00:33:07,039 pretty quick with all that stuff like
736 00:33:05,799 --> 00:33:10,639 scanning in
737 00:33:07,039 --> 00:33:12,840 textures and then allowing people to
738 00:33:10,639 --> 00:33:15,720 feel the
739 00:33:12,840 --> 00:33:18,360 textures in
740 00:33:15,720 --> 00:33:21,760 NVR before you know it the skynets here
741 00:33:18,360 --> 00:33:24,720 I don't know pretty cool and
742 00:33:21,760 --> 00:33:27,840 I'm always excited when people are
743 00:33:24,720 --> 00:33:29,240 shipping stuff and it's open source
744 00:33:27,840 --> 00:33:30,679 really like meta's approach I think it's
745 00:33:29,240 --> 00:33:33,360 very
746 00:33:30,679 --> 00:33:37,120 adorable and
747 00:33:33,360 --> 00:33:39,720 overall their stuff is great
748 00:33:37,120 --> 00:33:41,720 like and they ship it with a paper like
749 00:33:39,720 --> 00:33:43,760 everything they're shipping comes with a
750 00:33:41,720 --> 00:33:46,159 paper
751 00:33:43,760 --> 00:33:48,960 and I think
752 00:33:46,159 --> 00:33:52,399 that it's such a different approach than
753 00:33:48,960 --> 00:33:53,799 Google or open AI Google Google is more
754 00:33:52,399 --> 00:33:58,200 open source
755 00:33:53,799 --> 00:33:59,279 but like open AI isn't really open
756 00:33:58,200 --> 00:34:02,440 right
757 00:33:59,279 --> 00:34:07,679 like they're not very open so the only
758 00:34:02,440 --> 00:34:08,599 two big open source folks is X with XI
759 00:34:07,679 --> 00:34:13,119 which
760 00:34:08,599 --> 00:34:14,560 is AI on or I guess groon is what their
761 00:34:13,119 --> 00:34:15,679 product is called but the company's
762 00:34:14,560 --> 00:34:18,359 called
763 00:34:15,679 --> 00:34:20,440 xai and it's worth like 50 billion
764 00:34:18,359 --> 00:34:23,960 recently it was evaluated or something
765 00:34:20,440 --> 00:34:29,560 like that and
766 00:34:23,960 --> 00:34:33,119 then there is meta
767 00:34:29,560 --> 00:34:37,520 open AI doesn't open
768 00:34:33,119 --> 00:34:41,800 source anthropic isn't open
769 00:34:37,520 --> 00:34:45,119 source Microsoft AI is an open source so
770 00:34:41,800 --> 00:34:49,919 there's very few folks that are open
771 00:34:45,119 --> 00:34:51,520 sourcing but meta stuff is really good
772 00:34:49,919 --> 00:34:54,440 not only is their AI but their other
773 00:34:51,520 --> 00:34:56,560 research and it comes with a paper and
774 00:34:54,440 --> 00:34:58,720 they and they disclose so much stuff in
775 00:34:56,560 --> 00:35:00,079 the papers that that have read from them
776 00:34:58,720 --> 00:35:05,400 they're they're very
777 00:35:00,079 --> 00:35:07,440 interesting like I really don't know if
778 00:35:05,400 --> 00:35:09,880 if they keep shipping at this rate I'm
779 00:35:07,440 --> 00:35:11,480 not sure that there is a closed Source
780 00:35:09,880 --> 00:35:12,880 available I think everything just
781 00:35:11,480 --> 00:35:14,800 eventually becomes open source which
782 00:35:12,880 --> 00:35:18,320 would be great for everyone I mean it's
783 00:35:14,800 --> 00:35:21,720 a negative for closed Source but if open
784 00:35:18,320 --> 00:35:24,560 source and free is maybe
785 00:35:21,720 --> 00:35:27,680 90% there like why would you use
786 00:35:24,560 --> 00:35:31,200 something else if you get 90% of your
787 00:35:27,680 --> 00:35:34,040 use cases for free and maybe paying 20
788 00:35:31,200 --> 00:35:37,520 bucks or 40 bucks a month gives you the
789 00:35:34,040 --> 00:35:39,160 other 10% like you're those people the
790 00:35:37,520 --> 00:35:40,839 majority of the population are not going
791 00:35:39,160 --> 00:35:44,760 to search for the extra 10 they're going
792 00:35:40,839 --> 00:35:46,680 to be file with a 90 which is
793 00:35:44,760 --> 00:35:49,200 fine
794 00:35:46,680 --> 00:35:52,079 so let me know what you thought about
795 00:35:49,200 --> 00:35:54,760 this episode maybe maybe you felt my
796 00:35:52,079 --> 00:35:56,560 state of delirium maybe not so much felt
797 00:35:54,760 --> 00:35:59,520 like sometimes I was all over the place
798 00:35:56,560 --> 00:36:02,880 but as it's just the nature of of having
799 00:35:59,520 --> 00:36:05,720 a an eccentric host I
800 00:36:02,880 --> 00:36:07,920 guess what do you think about some of
801 00:36:05,720 --> 00:36:11,000 the things I talked about I thought the
802 00:36:07,920 --> 00:36:15,560 cracker three demo was pretty
803 00:36:11,000 --> 00:36:19,920 cool meta movie gen was something else
804 00:36:15,560 --> 00:36:22,400 entirely and this sparse
805 00:36:19,920 --> 00:36:24,800 sparse Robotics
806 00:36:22,400 --> 00:36:28,280 and all those things that come with it
807 00:36:24,800 --> 00:36:33,880 are incredible and be really cool for or
808 00:36:28,280 --> 00:36:36,520 robotic surgeries or manufacturing
809 00:36:33,880 --> 00:36:38,880 or what is it called and blanking on it
810 00:36:36,520 --> 00:36:40,760 Prosthetics I was like what yeah
811 00:36:38,880 --> 00:36:43,240 Prosthetics Prosthetics would be really
812 00:36:40,760 --> 00:36:45,160 cool for Prosthetics like maybe Advanced
813 00:36:43,240 --> 00:36:48,119 hand sensing for Prosthetics and be able
814 00:36:45,160 --> 00:36:50,400 to grip things with I don't know it's
815 00:36:48,119 --> 00:36:52,720 just going to be cool everything's going
816 00:36:50,400 --> 00:36:54,839 to be really cool in like 10 five years
817 00:36:52,720 --> 00:36:57,160 like the future is bright it's brighter
818 00:36:54,839 --> 00:37:00,319 than people people believe and and I'd
819 00:36:57,160 --> 00:37:03,480 always always lean into the favorable
820 00:37:00,319 --> 00:37:06,119 outcome and being positive and and and
821 00:37:03,480 --> 00:37:08,839 looking looking for the best in people
822 00:37:06,119 --> 00:37:11,240 and not searching for the bad there's a
823 00:37:08,839 --> 00:37:12,319 lot of things that happen to you but a
824 00:37:11,240 --> 00:37:14,480 lot of times those things aren't
825 00:37:12,319 --> 00:37:16,119 intentional from others so when
826 00:37:14,480 --> 00:37:20,119 something happens and you're like oh
827 00:37:16,119 --> 00:37:22,280 that person's not so nice or blah blah
828 00:37:20,119 --> 00:37:25,920 blah they probably didn't intentionally
829 00:37:22,280 --> 00:37:29,160 do that just throwing it out there just
830 00:37:25,920 --> 00:37:32,040 lean into the brightness of the world
831 00:37:29,160 --> 00:37:35,880 and feel the
832 00:37:32,040 --> 00:37:38,560 delirium but in the next coming weeks
833 00:37:35,880 --> 00:37:42,319 I've been trying to gather some guests
834 00:37:38,560 --> 00:37:46,119 on the show and I had some guests lined
835 00:37:42,319 --> 00:37:47,960 up and then I was out of country for
836 00:37:46,119 --> 00:37:50,480 five or so weeks and that kind of fell
837 00:37:47,960 --> 00:37:52,920 off and they're a bit annoyed with me so
838 00:37:50,480 --> 00:37:55,680 try to get them back on the show and
839 00:37:52,920 --> 00:37:57,880 then I have a special series an
840 00:37:55,680 --> 00:37:59,960 insurance series I want to have some
841 00:37:57,880 --> 00:38:03,240 insurance vendors on the show when I was
842 00:37:59,960 --> 00:38:04,640 at ITC I met some Founders that would be
843 00:38:03,240 --> 00:38:06,160 interested in coming on the show for
844 00:38:04,640 --> 00:38:08,960 some of the problems that they're
845 00:38:06,160 --> 00:38:12,200 solving and with that comes the
846 00:38:08,960 --> 00:38:13,760 responsibility of explaining what really
847 00:38:12,200 --> 00:38:16,200 is going on in
848 00:38:13,760 --> 00:38:18,240 Insurance what's happening in these
849 00:38:16,200 --> 00:38:20,240 catastrophic areas and why is it so
850 00:38:18,240 --> 00:38:24,160 difficult to get insurance and then in
851 00:38:20,240 --> 00:38:26,960 general explaining on like what what is
852 00:38:24,160 --> 00:38:29,880 Insurance like in and outs like how is
853 00:38:26,960 --> 00:38:32,560 insurance created what is reinsurance
854 00:38:29,880 --> 00:38:35,119 how do you price Insurance all those
855 00:38:32,560 --> 00:38:37,160 things and more and then I want to have
856 00:38:35,119 --> 00:38:38,280 some venders on the show talking about
857 00:38:37,160 --> 00:38:40,440 some of the problems that they're
858 00:38:38,280 --> 00:38:42,359 solving for either governmental agencies
859 00:38:40,440 --> 00:38:44,480 homeowners or
860 00:38:42,359 --> 00:38:45,599 companies but I want to do those things
861 00:38:44,480 --> 00:38:48,680 and do it
862 00:38:45,599 --> 00:38:50,960 right so you have the proper background
863 00:38:48,680 --> 00:38:53,079 you're like wow this is pretty
864 00:38:50,960 --> 00:38:56,160 incredible without that it's going to
865 00:38:53,079 --> 00:38:58,040 fall flat so bear with me I'm trying my
866 00:38:56,160 --> 00:39:01,960 best
867 00:38:58,040 --> 00:39:04,640 Co doesn't help so but as always I
868 00:39:01,960 --> 00:39:08,560 appreciate you and I hope that you tune
869 00:39:04,640 --> 00:39:10,760 in next week and I I hope I'm around
870 00:39:08,560 --> 00:39:13,040 next week as well I'm sure I will be
871 00:39:10,760 --> 00:39:15,599 just being dramatic but of course
872 00:39:13,040 --> 00:39:18,440 wherever you are in this world have a
873 00:39:15,599 --> 00:39:23,920 good afternoon good evening good morning
874 00:39:18,440 --> 00:39:23,920 and overall have a great day goodbye
SourcesFollow the source trail.
E044 Sources
Preserved episode evidence
[[E044 Production Transcript]] is the preferred author-supplied production transcript. Its untouched TXT source is preserved under Raw Sources, and its SHA-256 is D8DE3C3D053C693D693FAA364E76E4BA4FD672A11F597D8743A48A3770B8358E.
[[E44 - Transcript - YouTube automatic captions]] is the byte-preserved English automatic-caption file recovered from the public YouTube recording. Its SHA-256 is C913E0946FEE47DD37C3E6984891EC315D084629C58E20D34B7381B127E8F85E.
[[E44 - Transcript - YouTube automatic captions]] wraps the exact SRT body in an immutable Markdown record. Its SHA-256 is 8C9F7EEF3061E4A84178EE60097403698EB630D0EAE592907C7B745044C0B0FE.
The production transcript controls the spoken sequence and wording for drafting. The captions remain useful independent timing evidence, but they are machine generated and contain technical recognition errors. Exact quotations still require editorial review against the recording when precision matters.
[[E44 - Meta AI Shipping Spree - Legacy Episode Note]] is a retained legacy episode article dated November 26, 2024. It supports editorial comparison but is not an immutable transcript.
Existing public identity
daltonanderson.ghost.io/metas-tech-spree-robotics-video-and-ai-releases
This is the existing Ghost identity.
open.spotify.com/episode/5OwJfB19t12yKJs4QayHy0
This is the preserved Spotify identity.
This is the preserved YouTube identity and the best available recovery source if the raw transcript cannot be located.
CoTracker3
The paper defines CoTracker3 as a point-tracking model trained with pseudo-labels on real video. Use the paper for architecture, training, benchmark, and performance claims.
github.com/facebookresearch/co-tracker
The official repository provides code, checkpoints, setup requirements, licenses, and implementation notes. Repository availability does not prove suitability, support, security, or production readiness.
The project page is useful for visual examples and paper links. Demonstrations are selected examples, not a representative field evaluation.
Movie Gen
ai.meta.com/research/movie-gen
Meta's research page describes text-to-video, personalized video, video editing, and audio generation research. It is the controlling source for what Meta demonstrated at the time.
ai.meta.com/research/publications/movie-gen-a-cast-of-media-foundation-models
The research publication supports model, evaluation, dataset, and capability descriptions. It does not establish general public availability or rights to use every input or output.
ai.meta.com/blog/movie-gen-video-sound-generation-blumhouse
Meta's creator-partnership announcement documents a limited development setting. Do not turn it into a claim that the research models were an unrestricted public product.
about.fb.com/news/2025/06/edit-videos-with-meta-ai
Meta's June 2025 product announcement says its preset-prompt video editing feature was inspired by Movie Gen. Use this for later product context, not as proof that the original Movie Gen research models became public.
about.fb.com/news/2025/09/introducing-vibes-ai-videos
The September 2025 Vibes announcement documents a later early preview for AI video creation and remixing. Do not conflate that product surface or later Muse models with the original Movie Gen research release.
Spirit LM
ai.meta.com/blog/fair-news-segment-anything-2-1-meta-spirit-lm-layer-skip-salsa-lingua
Meta describes Spirit LM as a research model for expressive speech and text using interleaved sequences. It is not properly described as a general local assistant or a direct NotebookLM replacement.
ai.meta.com/resources/models-and-libraries/spiritlm-license
The license is required reading before any reuse. Availability under a research license is different from open source under a standard permissive software license.
The paper controls the model architecture, Base and Expressive variants, training data, interleaving method, evaluations, limitations, and harmful-output warning.
github.com/facebookresearch/spiritlm
The official repository contains weights, inference code, evaluation scripts, and a model card. Public artifacts do not override the FAIR Noncommercial Research License.
Sparsh
The primary publication describes self-supervised representations for vision-based tactile sensors, more than 460,000 tactile images, TacBench, and the reported evaluation.
ai.meta.com/blog/fair-robotics-open-source
Meta's announcement connects Sparsh with Digit 360 and Digit Plexus. Product, commercialization, and healthcare language remains attributed to Meta.
github.com/facebookresearch/sparsh
The official repository was archived on April 1, 2026 and is read-only. That current maintenance state must be visible anywhere readers are directed to the code.
Open source and ecosystem references
opensource.org/ai/open-source-ai-definition
The Open Source AI Definition 1.0 requires freedom to use, study, modify, and share an AI system for any purpose and describes the preferred form for modification. Use it as one current standard, not a substitute for reviewing a specific license.
The October 28, 2024 announcement establishes the version 1.0 release date and process.
about.fb.com/news/2024/07/open-source-ai-is-the-path-forward
Mark Zuckerberg's letter states Meta's ecosystem argument for Llama. It can support attributed claims about that stated strategy. It does not prove that every Meta research project follows one coordinated plan.
Editorial boundary
The production transcript supports the episode sequence and author wording. Official sources establish what their publishers disclosed. Neither source class establishes independent model quality, production fitness, safety, privacy, copyright clearance, commercial availability, or a universal legal conclusion.
The E045 Sparsh package and E046 CoTracker3 package own the deep technical records. E044 links to [[Sparsh Paper Code and Artifact Record]] and [[CoTracker3 Paper and Code Record]] rather than duplicating their detailed findings.