vidtheque.

read-only demo

Videos KB41dTlX1Uc

State of the Union: Why Local, Why Now — NVIDIA, Osmantic, Roboflow, EXO Labs, @matthew_berman

index_state ready data_status ok

AI Engineer· published 2026-07-11· 0:44:29· en-US· indexed 2026-08-10 19:53

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 0:43, 1 of 1 keyframes kept
  5. Shot 4, 0:43 to 1:12, 1 of 1 keyframes kept
  6. Shot 5, 1:12 to 1:41, 1 of 1 keyframes kept
  7. Shot 6, 1:41 to 2:09, 1 of 1 keyframes kept
  8. Shot 7, 2:09 to 2:38, 1 of 1 keyframes kept
  9. Shot 8, 2:38 to 2:48, 1 of 1 keyframes kept
  10. Shot 9, 2:48 to 3:25, 1 of 1 keyframes kept
  11. Shot 10, 3:25 to 3:52, 1 of 1 keyframes kept
  12. Shot 11, 3:52 to 4:31, 1 of 1 keyframes kept
  13. Shot 12, 4:31 to 4:40, 1 of 1 keyframes kept
  14. Shot 13, 4:40 to 4:49, 1 of 1 keyframes kept
  15. Shot 14, 4:49 to 4:53, 1 of 1 keyframes kept
  16. Shot 15, 4:53 to 5:00, 0 of 1 keyframes kept
  17. Shot 16, 5:00 to 5:32, 1 of 1 keyframes kept
  18. Shot 17, 5:32 to 6:16, 1 of 1 keyframes kept
  19. Shot 18, 6:16 to 6:27, 0 of 1 keyframes kept
  20. Shot 19, 6:27 to 6:35, 1 of 1 keyframes kept
  21. Shot 20, 6:35 to 7:13, 1 of 1 keyframes kept
  22. Shot 21, 7:13 to 7:39, 1 of 1 keyframes kept
  23. Shot 22, 7:39 to 8:05, 1 of 1 keyframes kept
  24. Shot 23, 8:05 to 8:32, 1 of 1 keyframes kept
  25. Shot 24, 8:32 to 8:37, 1 of 1 keyframes kept
  26. Shot 25, 8:37 to 9:10, 1 of 1 keyframes kept
  27. Shot 26, 9:10 to 9:42, 1 of 1 keyframes kept
  28. Shot 27, 9:42 to 10:15, 1 of 1 keyframes kept
  29. Shot 28, 10:15 to 10:26, 1 of 1 keyframes kept
  30. Shot 29, 10:26 to 10:28, 1 of 1 keyframes kept
  31. Shot 30, 10:28 to 10:32, 1 of 1 keyframes kept
  32. Shot 31, 10:32 to 11:16, 1 of 1 keyframes kept
  33. Shot 32, 11:16 to 11:35, 1 of 1 keyframes kept
  34. Shot 33, 11:35 to 12:02, 1 of 1 keyframes kept
  35. Shot 34, 12:02 to 12:30, 1 of 1 keyframes kept
  36. Shot 35, 12:30 to 12:58, 1 of 1 keyframes kept
  37. Shot 36, 12:58 to 13:30, 0 of 1 keyframes kept
  38. Shot 37, 13:30 to 14:02, 1 of 1 keyframes kept
  39. Shot 38, 14:02 to 14:33, 1 of 1 keyframes kept
  40. Shot 39, 14:33 to 14:36, 0 of 1 keyframes kept
  41. Shot 40, 14:36 to 15:05, 1 of 1 keyframes kept
  42. Shot 41, 15:05 to 15:34, 1 of 1 keyframes kept
  43. Shot 42, 15:34 to 16:04, 1 of 1 keyframes kept
  44. Shot 43, 16:04 to 16:16, 1 of 1 keyframes kept
  45. Shot 44, 16:16 to 16:26, 1 of 1 keyframes kept
  46. Shot 45, 16:26 to 17:04, 0 of 1 keyframes kept
  47. Shot 46, 17:04 to 17:32, 1 of 1 keyframes kept
  48. Shot 47, 17:32 to 18:00, 1 of 1 keyframes kept
  49. Shot 48, 18:00 to 18:28, 1 of 1 keyframes kept
  50. Shot 49, 18:28 to 18:56, 1 of 1 keyframes kept
  51. Shot 50, 18:56 to 19:15, 0 of 1 keyframes kept
  52. Shot 51, 19:15 to 19:19, 1 of 1 keyframes kept
  53. Shot 52, 19:19 to 19:44, 1 of 1 keyframes kept
  54. Shot 53, 19:44 to 20:10, 0 of 1 keyframes kept
  55. Shot 54, 20:10 to 20:40, 1 of 1 keyframes kept
  56. Shot 55, 20:40 to 21:11, 1 of 1 keyframes kept
  57. Shot 56, 21:11 to 21:41, 1 of 1 keyframes kept
  58. Shot 57, 21:41 to 21:57, 0 of 1 keyframes kept
  59. Shot 58, 21:57 to 22:11, 1 of 1 keyframes kept
  60. Shot 59, 22:11 to 22:21, 1 of 1 keyframes kept
  61. Shot 60, 22:21 to 22:50, 1 of 1 keyframes kept
  62. Shot 61, 22:50 to 23:20, 1 of 1 keyframes kept
  63. Shot 62, 23:20 to 23:25, 1 of 1 keyframes kept
  64. Shot 63, 23:25 to 23:56, 1 of 1 keyframes kept
  65. Shot 64, 23:56 to 24:26, 1 of 1 keyframes kept
  66. Shot 65, 24:26 to 24:57, 1 of 1 keyframes kept
  67. Shot 66, 24:57 to 25:05, 1 of 1 keyframes kept
  68. Shot 67, 25:05 to 25:48, 1 of 1 keyframes kept
  69. Shot 68, 25:48 to 25:50, 1 of 1 keyframes kept
  70. Shot 69, 25:50 to 26:15, 1 of 1 keyframes kept
  71. Shot 70, 26:15 to 26:40, 1 of 1 keyframes kept
  72. Shot 71, 26:40 to 27:06, 1 of 1 keyframes kept
  73. Shot 72, 27:06 to 27:22, 1 of 1 keyframes kept
  74. Shot 73, 27:22 to 27:51, 1 of 1 keyframes kept
  75. Shot 74, 27:51 to 28:19, 1 of 1 keyframes kept
  76. Shot 75, 28:19 to 28:47, 1 of 1 keyframes kept
  77. Shot 76, 28:47 to 29:16, 1 of 1 keyframes kept
  78. Shot 77, 29:16 to 29:21, 0 of 1 keyframes kept
  79. Shot 78, 29:21 to 29:30, 1 of 1 keyframes kept
  80. Shot 79, 29:30 to 29:35, 0 of 1 keyframes kept
  81. Shot 80, 29:35 to 29:40, 1 of 1 keyframes kept
  82. Shot 81, 29:40 to 29:46, 1 of 1 keyframes kept
  83. Shot 82, 29:46 to 30:34, 1 of 1 keyframes kept
  84. Shot 83, 30:34 to 30:38, 0 of 1 keyframes kept
  85. Shot 84, 30:38 to 31:00, 1 of 1 keyframes kept
  86. Shot 85, 31:00 to 31:04, 1 of 1 keyframes kept
  87. Shot 86, 31:04 to 31:33, 1 of 1 keyframes kept
  88. Shot 87, 31:33 to 32:02, 1 of 1 keyframes kept
  89. Shot 88, 32:02 to 32:31, 1 of 1 keyframes kept
  90. Shot 89, 32:31 to 32:34, 0 of 1 keyframes kept
  91. Shot 90, 32:34 to 32:58, 1 of 1 keyframes kept
  92. Shot 91, 32:58 to 33:06, 1 of 1 keyframes kept
  93. Shot 92, 33:06 to 33:33, 1 of 1 keyframes kept
  94. Shot 93, 33:33 to 33:37, 0 of 1 keyframes kept
  95. Shot 94, 33:37 to 34:07, 1 of 1 keyframes kept
  96. Shot 95, 34:07 to 34:36, 1 of 1 keyframes kept
  97. Shot 96, 34:36 to 35:03, 1 of 1 keyframes kept
  98. Shot 97, 35:03 to 35:29, 1 of 1 keyframes kept
  99. Shot 98, 35:29 to 35:33, 0 of 1 keyframes kept
  100. Shot 99, 35:33 to 36:00, 1 of 1 keyframes kept
  101. Shot 100, 36:00 to 36:27, 1 of 1 keyframes kept
  102. Shot 101, 36:27 to 36:54, 1 of 1 keyframes kept
  103. Shot 102, 36:54 to 37:21, 1 of 1 keyframes kept
  104. Shot 103, 37:21 to 37:34, 1 of 1 keyframes kept
  105. Shot 104, 37:34 to 37:50, 1 of 1 keyframes kept
  106. Shot 105, 37:50 to 38:31, 1 of 1 keyframes kept
  107. Shot 106, 38:31 to 38:34, 1 of 1 keyframes kept
  108. Shot 107, 38:34 to 39:06, 1 of 1 keyframes kept
  109. Shot 108, 39:06 to 39:29, 1 of 1 keyframes kept
  110. Shot 109, 39:29 to 39:36, 1 of 1 keyframes kept
  111. Shot 110, 39:36 to 40:12, 1 of 1 keyframes kept
  112. Shot 111, 40:12 to 40:49, 1 of 1 keyframes kept
  113. Shot 112, 40:49 to 41:26, 1 of 1 keyframes kept
  114. Shot 113, 41:26 to 41:31, 1 of 1 keyframes kept
  115. Shot 114, 41:31 to 41:57, 0 of 1 keyframes kept
  116. Shot 115, 41:57 to 42:00, 1 of 1 keyframes kept
  117. Shot 116, 42:00 to 42:41, 1 of 1 keyframes kept
  118. Shot 117, 42:41 to 42:57, 0 of 1 keyframes kept
  119. Shot 118, 42:57 to 43:22, 1 of 1 keyframes kept
  120. Shot 119, 43:22 to 43:31, 0 of 1 keyframes kept
  121. Shot 120, 43:31 to 43:49, 1 of 1 keyframes kept
  122. Shot 121, 43:49 to 44:12, 1 of 1 keyframes kept
  123. Shot 122, 44:12 to 44:28, 0 of 1 keyframes kept

123 shot(s).

keyframes kept every frame deduplicated

What was stored

cues
569
whisperx 569
chunks
80
from 569 cues
keyframes
105
kept of 123 captured
frames with text
104
157 lines read
chapters
23
from the source metadata
keyframe bytes
13.2 MB
word timings on 569 cues

Provenance

Each pipeline stage, its state and the model that produced it
stage state model started took
fetch done — 2026-08-10 17:17 3m 13s
stt done — 2026-08-10 17:21 53s
chunk done — 2026-08-10 17:21 0s
text_embed done — 2026-08-10 19:53 1s
keyframe done — 2026-08-10 17:21 5m 11s
ocr done — 2026-08-10 17:27 14s
frame_embed done — 2026-08-10 19:53 18s

Frames, and what the machine read

  • 0:02 #0 done2 line(s)

    shot 0·sharpness 449.9

    1. AlEngineer0.95
    2. World's Fair0.93

  • 0:03 #1 done2 line(s)

    shot 1·sharpness 657.8

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

  • 0:10 #2 done24 line(s)

    shot 2·sharpness 2733.9

    1. LAB & PLATINUM SPONSORS0.99
    2. Amazon AGI Lab0.98
    3. ANTHROP\C1.00
    4. Google DeepMind1.00
    5. MINIMAX0.95
    6. OpenAI0.92
    7. Akamai1.00
    8. arize0.94
    9. aws1.00
    10. Braintrust bright data0.98
    11. B1.00
    12. Browserbase1.00
    13. docker1.00
    14. :neo4j0.92
    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

  • 0:31 #3 done2 line(s)

    shot 3·sharpness 370.9

    1. roboflo0.92
    2. exo1.00

  • 0:49 #4 done1 line(s)

    shot 4·sharpness 205.0

    1. roboflow1.00

  • 1:32 #5 done1 line(s)

    shot 5·sharpness 130.0

    1. roboflow1.00

  • 2:03 #6 done1 line(s)

    shot 6·sharpness 131.8

    1. roboflow1.00

  • 2:18 #7 done1 line(s)

    shot 7·sharpness 133.3

    1. roboflow1.00

  • 2:47 #8 done2 line(s)

    shot 8·sharpness 379.3

    1. robafton0.69
    2. exo1.00

  • 3:21 #9 done2 line(s)

    shot 9·sharpness 77.8

    1. exo1.00
    2. 110.93

  • 3:39 #10 done1 line(s)

    shot 10·sharpness 71.2

    1. roboflow1.00

  • 4:27 #11 done1 line(s)

    shot 11·sharpness 64.4

    1. roboflow1.00

  • 4:39 #12 done1 line(s)

    shot 12·sharpness 372.3

    1. exo1.00

  • 4:47 #13 done1 line(s)

    shot 13·sharpness 101.3

    1. roboflow1.00

  • 4:52 #14 done1 line(s)

    shot 14·sharpness 101.5

    1. roboflow1.00

  • 4:55 #15 skipped

    shot 15·duplicate of #8

  • 5:10 #16 done1 line(s)

    shot 16·sharpness 80.6

    1. exo1.00

  • 5:37 #17 done1 line(s)

    shot 17·sharpness 73.5

    1. roboflow1.00

  • 6:19 #18 skipped

    shot 18·duplicate of #8

  • 6:28 #19 done1 line(s)

    shot 19·sharpness 98.9

    1. roboflow1.00

  • 6:47 #20 done2 line(s)

    shot 20·sharpness 373.3

    1. roboflow0.83
    2. exo1.00

  • 7:29 #21 done1 line(s)

    shot 21·sharpness 75.9

    1. exo1.00

  • 8:00 #22 done1 line(s)

    shot 22·sharpness 83.8

    1. exo1.00

  • 8:26 #23 done1 line(s)

    shot 23·sharpness 84.8

    1. exo1.00

Transcript

569 cues· 8,542 words· 45,642 chars

  1. 0:12 Can you guys hear us?
  2. 0:14 Sound check?
  3. 0:14 All right.
  4. 0:15 Give it up for Local AI, everyone.
  5. 0:23 I hope you guys are excited as we are.
  6. 0:25 This is the local AI summit.
  7. 0:28 So we're gonna be here all day talking about local AI.
  8. 0:31 And the reason why is we hit an inflection point this year.
  9. 0:35 Not only did the models get really good, but the harnesses got really good.
  10. 0:39 And this happened really fast.
  11. 0:41 It's been, I think, a struggle for anyone here to keep up.
  12. 0:44 I felt that most when I saw one of Andrej Karpathy's tweets.
  13. 0:47 In November, he tweeted that you can't really trust these coding agents alone yet.
  14. 0:50 You have to monitor them with an eye like a hawk.
  15. 0:53 Three months later, he tweets that he's struggling to keep up with the capabilities of how good this all has gone.
  16. 0:59 And the thing is, both times he was right.
  17. 1:01 This space is progressing really quickly, and it's wild how much he can do.
  18. 1:05 And honestly, the only thing to do is just try to use it a little bit more today than you did yesterday.
  19. 1:10 That's how to keep up with this space.
  20. 1:11 And that's exactly what you guys are doing right here.
  21. 1:14 And so I'm really excited about this panel.
  22. 1:16 The way that we use AI has also changed.
  23. 1:18 I'm not just using chat bots.
  24. 1:20 I'm not just asking simple questions.
  25. 1:22 When we got reasoning models, the profile of how the AI model responded changed.
  26. 1:28 Not only is it bursty and responding to me, but before the burst, it kind of plateaus for a bit.
  27. 1:33 It's reasoning.
  28. 1:34 It's churning on tokens that I'm not consuming.
  29. 1:38 And then we got agents.
  30. 1:39 And suddenly, I don't even want these agents to turn off.
  31. 1:42 I want these always on agents.
  32. 1:44 They can always be productive if we set them up.
  33. 1:47 And so we have enterprises that want to put a lot of their IP into this because it becomes more useful.
  34. 1:53 We have consumers with the same thing.
  35. 1:55 I want to give it my health data, my medical records.
  36. 1:58 I want to give it footage from my home camera.
  37. 2:00 And both enterprises and consumers, we don't want that stuff to leak.
  38. 2:04 And as you also have a profile of tokens continuously generating, suddenly costs matter.
  39. 2:11 So local is amazing for both of those things.
  40. 2:14 You get to make sure that you are plateaued on the costs for the tokens that you're generating.
  41. 2:20 And also, everything sits in that room.
  42. 2:22 So we have amazing demos here after these talks where everything that's being run stays on those devices.
  43. 2:28 It stays in this room.
  44. 2:29 And so that's a really nice guarantee.
  45. 2:31 So as we turn over the panels, first, do you guys want to introduce yourselves?
  46. 2:36 Yeah, should I start, please?
  47. 2:37 You got the mic?
  48. 2:38 Yeah, your mic's up.
  49. 2:40 So yeah, I'm Alex.
  50. 2:42 I'm the co-founder and CEO of ExoLabs and also the creator of Local.ai.

Chapters

  1. 0:00 Welcome to the Local AI Summit
  2. 0:40 Karpathy twice right on keeping up
  3. 1:16 Reasoning models and always on agents
  4. 2:32 Panelist introductions
  5. 4:41 When the inflection point hit
  6. 6:36 GPT 4o quality in your pocket
  7. 7:14 Llama 405B to DeepSeek to GLM 5.2
  8. 8:47 The airplane accessibility story
  9. 10:25 Harnesses give models the real world
  10. 11:27 What language learns from vision
  11. 13:19 A multimodel world in practice
  12. 13:57 Coinbase: tokens up, costs flat
  13. 15:03 Control, sovereignty, no rug pulls
  14. 17:50 Small specialized models and data flywheels
  15. 19:45 A second headquarters inside NVIDIA
  16. 21:50 10x on the DGX Spark by swarming
  17. 24:32 Desk and data center share an architecture
  18. 26:11 ODS and point and click onboarding
  19. 27:14 Where local still falls short
  20. 32:33 Why finetuning as a service stalled
  21. 35:34 Distillation down to a submarine
  22. 39:42 The biggest open problems in local
  23. 42:01 Open source advocacy and closing

Open at this second