vidtheque.

read-only demo

Videos FWMJQDH3iK0

Local Models: Trust, Control, Optimization — Carter Abdallah, NVIDIA

index_state ready data_status ok

AI Engineer· published 2026-08-07· 0:43:20· en-US· indexed 2026-08-10 19:34

Open on YouTube

Scene timeline

  1. Shot 0, 0:00 to 0:03, 1 of 1 keyframes kept
  2. Shot 1, 0:03 to 0:05, 1 of 1 keyframes kept
  3. Shot 2, 0:05 to 0:12, 1 of 1 keyframes kept
  4. Shot 3, 0:12 to 0:35, 1 of 1 keyframes kept
  5. Shot 4, 0:35 to 0:46, 1 of 1 keyframes kept
  6. Shot 5, 0:46 to 1:08, 1 of 1 keyframes kept
  7. Shot 6, 1:08 to 1:21, 1 of 1 keyframes kept
  8. Shot 7, 1:21 to 1:28, 1 of 1 keyframes kept
  9. Shot 8, 1:28 to 2:14, 1 of 1 keyframes kept
  10. Shot 9, 2:14 to 2:21, 1 of 1 keyframes kept
  11. Shot 10, 2:21 to 2:52, 1 of 1 keyframes kept
  12. Shot 11, 2:52 to 3:23, 1 of 1 keyframes kept
  13. Shot 12, 3:23 to 3:53, 1 of 1 keyframes kept
  14. Shot 13, 3:53 to 3:55, 1 of 1 keyframes kept
  15. Shot 14, 3:55 to 4:18, 1 of 1 keyframes kept
  16. Shot 15, 4:18 to 4:23, 1 of 1 keyframes kept
  17. Shot 16, 4:23 to 4:55, 1 of 1 keyframes kept
  18. Shot 17, 4:55 to 5:27, 0 of 1 keyframes kept
  19. Shot 18, 5:27 to 6:00, 1 of 1 keyframes kept
  20. Shot 19, 6:00 to 6:04, 1 of 1 keyframes kept
  21. Shot 20, 6:04 to 6:10, 1 of 1 keyframes kept
  22. Shot 21, 6:10 to 6:15, 1 of 1 keyframes kept
  23. Shot 22, 6:15 to 6:22, 1 of 1 keyframes kept
  24. Shot 23, 6:22 to 6:52, 1 of 1 keyframes kept
  25. Shot 24, 6:52 to 7:22, 1 of 1 keyframes kept
  26. Shot 25, 7:22 to 7:51, 1 of 1 keyframes kept
  27. Shot 26, 7:51 to 8:20, 1 of 1 keyframes kept
  28. Shot 27, 8:20 to 8:49, 1 of 1 keyframes kept
  29. Shot 28, 8:49 to 9:18, 1 of 1 keyframes kept
  30. Shot 29, 9:18 to 9:47, 1 of 1 keyframes kept
  31. Shot 30, 9:47 to 10:16, 1 of 1 keyframes kept
  32. Shot 31, 10:16 to 10:21, 1 of 1 keyframes kept
  33. Shot 32, 10:21 to 10:50, 1 of 1 keyframes kept
  34. Shot 33, 10:50 to 11:20, 1 of 1 keyframes kept
  35. Shot 34, 11:20 to 11:29, 1 of 1 keyframes kept
  36. Shot 35, 11:29 to 11:40, 1 of 1 keyframes kept
  37. Shot 36, 11:40 to 11:44, 1 of 1 keyframes kept
  38. Shot 37, 11:44 to 12:10, 1 of 1 keyframes kept
  39. Shot 38, 12:10 to 12:29, 1 of 1 keyframes kept
  40. Shot 39, 12:29 to 13:02, 1 of 1 keyframes kept
  41. Shot 40, 13:02 to 13:35, 1 of 1 keyframes kept
  42. Shot 41, 13:35 to 14:08, 1 of 1 keyframes kept
  43. Shot 42, 14:08 to 14:17, 0 of 1 keyframes kept
  44. Shot 43, 14:17 to 14:50, 1 of 1 keyframes kept
  45. Shot 44, 14:50 to 15:24, 1 of 1 keyframes kept
  46. Shot 45, 15:24 to 15:27, 1 of 1 keyframes kept
  47. Shot 46, 15:27 to 15:57, 1 of 1 keyframes kept
  48. Shot 47, 15:57 to 16:01, 1 of 1 keyframes kept
  49. Shot 48, 16:01 to 16:29, 1 of 1 keyframes kept
  50. Shot 49, 16:29 to 16:56, 1 of 1 keyframes kept
  51. Shot 50, 16:56 to 17:24, 1 of 1 keyframes kept
  52. Shot 51, 17:24 to 17:30, 1 of 1 keyframes kept
  53. Shot 52, 17:30 to 17:55, 1 of 1 keyframes kept
  54. Shot 53, 17:55 to 18:20, 0 of 1 keyframes kept
  55. Shot 54, 18:20 to 18:45, 1 of 1 keyframes kept
  56. Shot 55, 18:45 to 19:10, 1 of 1 keyframes kept
  57. Shot 56, 19:10 to 19:14, 1 of 1 keyframes kept
  58. Shot 57, 19:14 to 19:36, 1 of 1 keyframes kept
  59. Shot 58, 19:36 to 19:40, 1 of 1 keyframes kept
  60. Shot 59, 19:40 to 20:07, 1 of 1 keyframes kept
  61. Shot 60, 20:07 to 20:35, 1 of 1 keyframes kept
  62. Shot 61, 20:35 to 21:03, 1 of 1 keyframes kept
  63. Shot 62, 21:03 to 21:30, 1 of 1 keyframes kept
  64. Shot 63, 21:30 to 21:37, 1 of 1 keyframes kept
  65. Shot 64, 21:37 to 22:05, 1 of 1 keyframes kept
  66. Shot 65, 22:05 to 22:33, 1 of 1 keyframes kept
  67. Shot 66, 22:33 to 22:41, 1 of 1 keyframes kept
  68. Shot 67, 22:41 to 23:10, 1 of 1 keyframes kept
  69. Shot 68, 23:10 to 23:38, 1 of 1 keyframes kept
  70. Shot 69, 23:38 to 23:55, 1 of 1 keyframes kept
  71. Shot 70, 23:55 to 24:07, 1 of 1 keyframes kept
  72. Shot 71, 24:07 to 24:09, 0 of 1 keyframes kept
  73. Shot 72, 24:09 to 24:35, 1 of 1 keyframes kept
  74. Shot 73, 24:35 to 25:00, 1 of 1 keyframes kept
  75. Shot 74, 25:00 to 25:25, 1 of 1 keyframes kept
  76. Shot 75, 25:25 to 25:51, 1 of 1 keyframes kept
  77. Shot 76, 25:51 to 26:16, 1 of 1 keyframes kept
  78. Shot 77, 26:16 to 26:41, 1 of 1 keyframes kept
  79. Shot 78, 26:41 to 27:07, 1 of 1 keyframes kept
  80. Shot 79, 27:07 to 27:11, 1 of 1 keyframes kept
  81. Shot 80, 27:11 to 27:49, 1 of 1 keyframes kept
  82. Shot 81, 27:49 to 28:26, 1 of 1 keyframes kept
  83. Shot 82, 28:26 to 28:31, 1 of 1 keyframes kept
  84. Shot 83, 28:31 to 28:57, 1 of 1 keyframes kept
  85. Shot 84, 28:57 to 29:22, 0 of 1 keyframes kept
  86. Shot 85, 29:22 to 29:48, 1 of 1 keyframes kept
  87. Shot 86, 29:48 to 29:52, 0 of 1 keyframes kept
  88. Shot 87, 29:52 to 30:23, 1 of 1 keyframes kept
  89. Shot 88, 30:23 to 30:53, 1 of 1 keyframes kept
  90. Shot 89, 30:53 to 31:24, 1 of 1 keyframes kept
  91. Shot 90, 31:24 to 31:54, 1 of 1 keyframes kept
  92. Shot 91, 31:54 to 31:59, 1 of 1 keyframes kept
  93. Shot 92, 31:59 to 32:24, 1 of 1 keyframes kept
  94. Shot 93, 32:24 to 32:50, 0 of 1 keyframes kept
  95. Shot 94, 32:50 to 33:16, 1 of 1 keyframes kept
  96. Shot 95, 33:16 to 33:18, 1 of 1 keyframes kept
  97. Shot 96, 33:18 to 33:43, 1 of 1 keyframes kept
  98. Shot 97, 33:43 to 34:09, 1 of 1 keyframes kept
  99. Shot 98, 34:09 to 34:34, 1 of 1 keyframes kept
  100. Shot 99, 34:34 to 34:59, 1 of 1 keyframes kept
  101. Shot 100, 34:59 to 35:24, 1 of 1 keyframes kept
  102. Shot 101, 35:24 to 35:49, 1 of 1 keyframes kept
  103. Shot 102, 35:49 to 36:15, 1 of 1 keyframes kept
  104. Shot 103, 36:15 to 36:40, 0 of 1 keyframes kept
  105. Shot 104, 36:40 to 37:05, 1 of 1 keyframes kept
  106. Shot 105, 37:05 to 37:30, 1 of 1 keyframes kept
  107. Shot 106, 37:30 to 37:56, 1 of 1 keyframes kept
  108. Shot 107, 37:56 to 38:23, 1 of 1 keyframes kept
  109. Shot 108, 38:23 to 38:50, 0 of 1 keyframes kept
  110. Shot 109, 38:50 to 39:17, 0 of 1 keyframes kept
  111. Shot 110, 39:17 to 39:44, 1 of 1 keyframes kept
  112. Shot 111, 39:44 to 40:04, 1 of 1 keyframes kept
  113. Shot 112, 40:04 to 40:07, 1 of 1 keyframes kept
  114. Shot 113, 40:07 to 40:10, 1 of 1 keyframes kept
  115. Shot 114, 40:10 to 40:58, 1 of 1 keyframes kept
  116. Shot 115, 40:58 to 41:01, 1 of 1 keyframes kept
  117. Shot 116, 41:01 to 41:25, 1 of 1 keyframes kept
  118. Shot 117, 41:25 to 41:29, 1 of 1 keyframes kept
  119. Shot 118, 41:29 to 41:59, 1 of 1 keyframes kept
  120. Shot 119, 41:59 to 42:28, 1 of 1 keyframes kept
  121. Shot 120, 42:28 to 42:32, 1 of 1 keyframes kept
  122. Shot 121, 42:32 to 42:52, 1 of 1 keyframes kept
  123. Shot 122, 42:52 to 43:03, 1 of 1 keyframes kept
  124. Shot 123, 43:03 to 43:20, 0 of 1 keyframes kept

124 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
348
whisperx 348
chunks
74
from 348 cues
keyframes
113
kept of 124 captured
frames with text
43
80 lines read
chapters
27
from the source metadata
keyframe bytes
11.8 MB
word timings on 348 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-09 21:12 0s
stt done — 2026-08-08 23:41 49s
chunk done — 2026-08-08 23:42 0s
text_embed done — 2026-08-10 19:34 1s
keyframe done — 2026-08-08 23:42 7m 30s
ocr done — 2026-08-08 23:50 16s
frame_embed done — 2026-08-10 19:34 19s

Frames, and what the machine read

  • 0:02 #0 done2 line(s)

    shot 0·sharpness 453.7

    1. AIEngineer0.95
    2. World's Fair0.96

  • 0:03 #1 done2 line(s)

    shot 1·sharpness 659.6

    1. AlEngineer0.95
    2. World's Fair0.99

  • 0:10 #2 done24 line(s)

    shot 2·sharpness 2734.1

    1. LAB & PLATINUM SPONSORS0.99
    2. Amazon AGI Lab0.98
    3. ANTHROP\C1.00
    4. Google DeepMind1.00
    5. MINIMAX0.94
    6. OpenAI0.92
    7. Akamai1.00
    8. arize1.00
    9. aws1.00
    10. Braintrust bright data0.99
    11. B1.00
    12. Browserbase1.00
    13. docker1.00
    14. :neo4j0.91
    15. ORACLE1.00
    16. PayPal1.00
    17. qodo1.00
    18. reducto1.00
    19. Sonar1.00
    20. Makers of0.99
    21. togetherai1.00
    22. Unblocked1.00
    23. WorkOS1.00
    24. SonarQube1.00

  • 0:15 #3 done1 line(s)

    shot 3·sharpness 433.3

    1. Fair1.00

  • 0:36 #4 done2 line(s)

    shot 4·sharpness 77.0

    1. TSAT0.76
    2. 小0.97

  • 0:48 #5 done1 line(s)

    shot 5·sharpness 408.7

    1. 'sFaii0.88

  • 1:11 #6 done1 line(s)

    shot 6·sharpness 74.1

    1. 香0.78

  • 1:27 #7 done1 line(s)

    shot 7·sharpness 351.0

    1. haring1.00

  • 1:42 #8 empty

    shot 8·sharpness 43.5

  • 2:15 #9 empty

    shot 9·sharpness 366.9

  • 2:28 #10 done1 line(s)

    shot 10·sharpness 104.1

    1. mle0.58

  • 3:13 #11 empty

    shot 11·sharpness 89.2

  • 3:32 #12 done2 line(s)

    shot 12·sharpness 82.4

    1. Chris0.84
    2. 國0.55

  • 3:55 #13 empty

    shot 13·sharpness 101.8

  • 4:00 #14 done3 line(s)

    shot 14·sharpness 94.1

    1. 养业0.73
    2. Speaker0.98
    3. Cheris0.80

  • 4:21 #15 empty

    shot 15·sharpness 370.3

  • 4:48 #16 done1 line(s)

    shot 16·sharpness 106.9

    1. Speaker1.00

  • 5:20 #17 skipped

    shot 17·duplicate of #16

  • 5:40 #18 done2 line(s)

    shot 18·sharpness 101.5

    1. a0.58
    2. Speaker1.00

  • 6:03 #19 empty

    shot 19·sharpness 371.8

  • 6:09 #20 done1 line(s)

    shot 20·sharpness 105.5

    1. Speaker1.00

  • 6:15 #21 done1 line(s)

    shot 21·sharpness 374.0

    1. S0.65

  • 6:18 #22 empty

    shot 22·sharpness 78.3

  • 6:25 #23 empty

    shot 23·sharpness 373.7

Transcript

348 cues· 7,661 words· 40,961 chars

  1. 0:12 So I hope everybody had a great lunch and you got to check out some of the amazing demos that we have.
  2. 0:18 We're going to begin the panel, the first panel of the afternoon here where we're going to be talking about, of course, the engines that are actually powering the stuff that could remotely be used for things like local sovereign, any kind of ownership over your own artificial intelligence.
  3. 0:34 And of course, the engine powering those in addition to the hardware is the models themselves.
  4. 0:38 And so for this panel, we have excellent guests.
  5. 0:41 We have
  6. 0:42 Vincent, who is the CEO and founder of Prime Intellect.
  7. 0:45 We've got Lucas, the CTO of RCAI, and we've got Chris, who is the senior product research engineer on the Nematron family of models at NVIDIA.
  8. 0:53 Now, what's really cool about working in this industry is really cool companies like this.
  9. 0:58 We all get to work together.
  10. 0:59 And so this is one panel where we all directly get to work together on both models, infrastructure, some of the ways that we think that the direction of the industry should go.
  11. 1:08 And each of us kind of play a different role in that stack.
  12. 1:11 But I wanna leave it to you guys to introduce yourself and be able to talk about sort of the charter that you see, the problem of the stack that you guys are working on.
  13. 1:20 Awesome, should I kick it off?
  14. 1:22 Kick it off.
  15. 1:22 Yeah, so I'm Vincent, as you mentioned.
  16. 1:24 And then really the goal with Prime Intellect from the beginning was like to ensure that basically frontier intelligence will be open and accessible, not just the models, but also the full stack to train the models.
  17. 1:36 So kind of like this was like our motivation from the beginning.
  18. 1:39 And we've worked also together with a lot of gentlemen on stage.
  19. 1:44 Like on the one side, it's like we work with folks like Lucas and Arcee to help them train frontier open models.
  20. 1:51 We help also...
  21. 1:53 like NVIDIA on the Nemotron coalition help out there from the open malls.
  22. 1:56 And I think I'm actually think both like Nemotron and Trinity might be the best like two open malls right now outside of China.
  23. 2:05 So I think it's actually like we need to fact check that, you know, but this is actually for my, I think they might be.
  24. 2:12 Our marketing says that, yeah.
  25. 2:15 But yeah, so that's the high level.
  26. 2:19 My name is Lucas Atkins.
  27. 2:20 I'm happy to be here, and thank you for joining.
  28. 2:23 Very similar to Vincent, RC was founded with the idea of domain-specific owned models are going to be needed.
  29. 2:34 We were founded early 2023, jumping on the custom model train quite early.
  30. 2:41 You have all these people who are excited about AI and all the things these new generation of LLMs can do, but they're using these monolithic, very expensive closed APIs for at the time and still very narrow tasks that don't require
  31. 2:57 You know, at the time, it was $100 per million tokens out.
  32. 3:01 And through doing that, we were building on top of open models, and we were releasing a lot of our tooling in the open.
  33. 3:07 And we noticed that in the United States and in the West, you know, in general, we were starting to lose
  34. 3:15 leadership in the open model space.
  35. 3:17 A lot of it was coming out of China, and that's amazing.
  36. 3:19 I love those models.
  37. 3:20 We learn a lot from them.
  38. 3:21 We're close with a lot of the people building those.
  39. 3:24 But when you're working with large enterprises and companies and geopolitics gets involved, whether you like it or not, you have people that become concerned about where those models are coming from.
  40. 3:37 decided that we had a good group of people and we had a good group of partners like NVIDIA and like Prime Intellect where we could probably try to pre-train ourselves.
  41. 3:46 So last year we did that.
  42. 3:47 We kind of reoriented the entire company towards let's figure out how to pre-train a 400 billion parameter model in six months.
  43. 3:56 A lot of people said it was impossible and in many ways it was, but we figured it out and now we are an open model lab working with our wonderful partners
  44. 4:05 AND OUR CUSTOMERS TO BUILD WESTERN OPEN MODELS THAT ARE PERMISSIVE AND YOU CAN OWN THOSE AND CUSTOMIZE THEM OR RUN THEM WHEREVER YOU WANT.
  45. 4:16 AND THAT'S KIND OF WHERE WE'RE AT RIGHT NOW.
  46. 4:17 SO THANKS FOR HAVING ME.
  47. 4:19 YEAH.
  48. 4:19 SO I'M CHRIS.
  49. 4:21 I WORK AT NVIDIA AS A PRODUCT RESEARCH ENGINEER AND I SUPPORT THE NEMOTRON FAMILY OF MODELS.
  50. 4:26 I THINK IT'S, YOU KNOW, WE'VE TALKED A LOT ABOUT WHY WE DO NEMOTRON BUT JUST TO SAY IT A FEW MORE TIMES.

Chapters

  1. 0:00 Welcome and why this panel is the whole stack
  2. 1:14 Prime Intellect: keeping the training stack open
  3. 2:15 Arcee: why the west was losing the open model lead
  4. 3:22 Pretraining a 400 billion parameter model in six months
  5. 4:23 Nemotron: faster models are smarter models
  6. 6:35 Is open source actually less trustworthy
  7. 7:40 Trust is not safety
  8. 9:48 When access stopped being guaranteed
  9. 10:56 Releasing the data sets alongside the weights
  10. 11:59 The open superintelligence stack
  11. 13:02 Post training as the accessible layer
  12. 14:07 Beating frontier models on a specific use case
  13. 15:11 Making your costs predictable
  14. 16:11 When the model and the harness blend together
  15. 18:15 The mismanaged genius
  16. 19:19 A call to action for builders
  17. 20:24 Who owns the data you generate
  18. 21:27 Owning your outputs, not just your weights
  19. 22:35 The open MDW license
  20. 23:36 Where post training unlocks new use cases
  21. 27:46 You do not need frontier intelligence for most tasks
  22. 28:53 Why efficiency has to happen in the open
  23. 29:55 Closed models are not the enemy
  24. 33:05 Predictions for the next year
  25. 38:22 Running everything on your laptop
  26. 39:29 Agent operating systems and the next Siri moment
  27. 41:36 Closing

Open at this second