YouTube2Text

Tech CEO: Big AI’s monopoly is breaking | Daniela Braga, Defined.ai — Transcript

by The Tech Report · 2,777 words · 515 segments · language en · Watch on YouTube

Full transcript

  1. 0:00everybody is starting to invest in
  2. 0:01smaller
  3. 0:03specialized models. I see a world where
  4. 0:06uh
  5. 0:07this decentralization is for for one
  6. 0:09offload the the the
  7. 0:11concentration
  8. 0:13in certain in a few powers in the in
  9. 0:16terms of AI generation and owning and
  10. 0:20the knowledge
  11. 0:21uh
  12. 0:22and it's and we see that all over the
  13. 0:24world.
  14. 0:26>> [music]
  15. 0:31>> Hi, I'm Isaac and today on the Tech
  16. 0:33Report and at Web Summit kind of I'm
  17. 0:36joined by CEO of Defined AI, Daniela
  18. 0:39Braga. Thanks for coming on.
  19. 0:40>> Hi Isaac. It's a pleasure to be here.
  20. 0:42>> So back in June, Cloudflare announced
  21. 0:45that bot traffic surpassed human traffic
  22. 0:47well ahead of its sort of mid 2027
  23. 0:49expectations.
  24. 0:51And around the same time, the New
  25. 0:52Scientist has reported that AI
  26. 0:54developers and AI data
  27. 0:57data set annotators
  28. 0:59admitted to using chatbots to help do
  29. 1:01their work.
  30. 1:02As a founder of a ethical training data
  31. 1:05platform, if I could put it that way,
  32. 1:07how real is the data wall, the data
  33. 1:09bottleneck?
  34. 1:11And how big of a threat is is model
  35. 1:13collapse?
  36. 1:14>> Hm.
  37. 1:15Right. So Elon Musk a year and a half
  38. 1:19ago, I think he did say that we were
  39. 1:21going to run out of data.
  40. 1:23Uh
  41. 1:24but that he meant out of
  42. 1:26open
  43. 1:28uh not not open. Uh he meant
  44. 1:32data that is out in the web that is
  45. 1:35publicly available, which doesn't mean
  46. 1:38free
  47. 1:39in terms of IP or copyrights.
  48. 1:42But we know that all the Frontier Labs
  49. 1:46have
  50. 1:47crawlers constantly uh
  51. 1:50constantly
  52. 1:51crawling the web and crawling the data.
  53. 1:54So in that regard,
  54. 1:56uh it is true that
  55. 1:59all the internet been
  56. 2:00crawled already
  57. 2:02and and there's a feed of remaining
  58. 2:06digital data that is continuing
  59. 2:08continues to be crawled
  60. 2:11but it's not very different from what is
  61. 2:14already there in the models.
  62. 2:16So
  63. 2:17in that regard, yes, there's a
  64. 2:19bottleneck of data of publicly available
  65. 2:22data
  66. 2:23but it but the real opportunity and
  67. 2:26differentiator is the closed source
  68. 2:29private data.
  69. 2:30That data is much bigger.
  70. 2:33It's just not so accessible. It's
  71. 2:35definitely not accessible. That's what
  72. 2:37we that's what our business model is
  73. 2:39about and our
  74. 2:42product is about is is basically
  75. 2:45making that available to the world but
  76. 2:48for a fee. It has to be paid.
  77. 2:51So that's the big difference. There's a
  78. 2:52third type of data which is synthetic
  79. 2:54data. That is absolutely
  80. 2:58augmenting
  81. 3:00I mean we know we know that our our
  82. 3:02clients are using a fair amount of
  83. 3:05synthetic data in their model.
  84. 3:07Typically synthetic data is the one that
  85. 3:11in the wrong proportions causes model
  86. 3:13collapse
  87. 3:15of course.
  88. 3:16In the right proportions is very
  89. 3:18important because you're going to be
  90. 3:19able to
  91. 3:21supplement
  92. 3:22scenarios that otherwise you can find in
  93. 3:24real data.
  94. 3:26But the world of data is specializing.
  95. 3:29Again, what you can find
  96. 3:31in the open is not going to give the
  97. 3:34edge to the frontier labs. And we're
  98. 3:36starting if in the beginning of this
  99. 3:38race, I mean two or three years ago, I
  100. 3:41remember giving a a talk at CES where I
  101. 3:43was showing the output of all the
  102. 3:44models. They were very similar.
  103. 3:47You can start seeing differentiation in
  104. 3:49the models. They are getting specialized
  105. 3:51in domains in verticals
  106. 3:53and this is because of the closed source
  107. 3:55data.
  108. 3:57>> A lot of attention around AI has been
  109. 3:59focused on sort of scaling and growth
  110. 4:01and more and more hardware to sort of
  111. 4:03make things better. But
  112. 4:05it seems to me and may I could be
  113. 4:07accused of being
  114. 4:08not the most informed on all of the
  115. 4:10scopes of AI.
  116. 4:12But it seems to me that the next phase
  117. 4:14of AI development might be fought more
  118. 4:16on sort of creating specialized AI tools
  119. 4:19trained on sort of very high quality
  120. 4:20data sets
  121. 4:22which have more of a defined use case.
  122. 4:24Thompson Reuters AI model comes to mind
  123. 4:26if you're familiar with it.
  124. 4:28What's your take on the the industry's
  125. 4:31focus on generalist models and the
  126. 4:33potential rise of specialist models with
  127. 4:35more sort of bespoke use cases?
  128. 4:38>> Definitely. I I mean even for from a
  129. 4:40cost perspective, specialized models is
  130. 4:43the way to go. Everybody is starting to
  131. 4:45invest in smaller
  132. 4:47specialized models with
  133. 4:50uh and we see that. We see
  134. 4:52uh so if
  135. 4:54three, four years ago our clients would
  136. 4:57come, it would be a handful of
  137. 5:00the typical frontier labs and they would
  138. 5:02come in and
  139. 5:04request large amounts of data
  140. 5:06and not necessarily even
  141. 5:09with well large amounts of data in
  142. 5:11in in in
  143. 5:13basically multi-model aspects of of data
  144. 5:16but not
  145. 5:18uh but things that were in big scale and
  146. 5:21more generic content and topics.
  147. 5:24Uh today we see a surge of other clients
  148. 5:28that are not the frontier labs but they
  149. 5:30are customizing
  150. 5:32open source models or closed source
  151. 5:34models.
  152. 5:35And they want smaller data sets but
  153. 5:38highly curated, highly curated with
  154. 5:41specialists, with humans in the loop
  155. 5:43that are doctors or mathematicians or or
  156. 5:47chemists or so this is this is what
  157. 5:50we're seeing and all all linguists in
  158. 5:53PhDs PhDs including.
  159. 5:56So this is we see
  160. 5:58we see that movement and uh it's
  161. 6:00definitely the way to go
  162. 6:02uh
  163. 6:03from
  164. 6:04from an industry perspective.
  165. 6:05>> And
  166. 6:06>> do open weight models fit into this this
  167. 6:09idea? 'Cause I mean,
  168. 6:11quite simply, I mean, admit you can
  169. 6:13correct me if I'm wrong here, smaller
  170. 6:15bespoke models sort of trained on
  171. 6:17tightly curated and
  172. 6:19even proprietary data is is surely going
  173. 6:21to impact the AI industry more broadly
  174. 6:24and I'm kind of thinking
  175. 6:25the sort of the the open AI's, the the M
  176. 6:28Tropics who are making these very big
  177. 6:30models, but what happens to them if the
  178. 6:32industry is moving already towards
  179. 6:35open weight sort of
  180. 6:37uh yeah, open weight models which they
  181. 6:39can tailor to their use cases.
  182. 6:41>> No, that exactly right. That's that's
  183. 6:43that's what we see in the industry.
  184. 6:45We see
  185. 6:46uh
  186. 6:47consulting companies, startups that are
  187. 6:50not
  188. 6:51for sure, they don't have budgets to
  189. 6:53build
  190. 6:54large scale
  191. 6:56models, world models, large language
  192. 6:59models, they're not going to build that.
  193. 7:00They definitely are are uh
  194. 7:03fine tuning open source models
  195. 7:05for their own purpose with specific
  196. 7:07hardware, with specific industry
  197. 7:09domains, with some private data.
  198. 7:11Uh private data that
  199. 7:13that can be theirs or or or acquired
  200. 7:16through us.
  201. 7:17So that's that's where we see the the
  202. 7:20trend. And
  203. 7:21to give you an idea, if we had been
  204. 7:24mostly serving in the last few years
  205. 7:27around
  206. 7:2850 massive
  207. 7:30Frontier Labs,
  208. 7:32the our list of clients exploded this
  209. 7:34year to
  210. 7:36over 150
  211. 7:38because this is the these are companies
  212. 7:40that are not coming from they are not AI
  213. 7:42native
  214. 7:44um and they are starting to build their
  215. 7:45own models.
  216. 7:47>> Mhm. So I'm I'm just really curious if I
  217. 7:48can just push you on that a little more.
  218. 7:50I'm wondering if you think that the the
  219. 7:53Frontier model developers are
  220. 7:55are already starting to in some ways
  221. 7:56lose lose some of their their piece of
  222. 7:58the pie, lose some ground to to this
  223. 8:01this idea of moving towards models which
  224. 8:03fit specific use cases to open weight
  225. 8:05models.
  226. 8:06>> I not yet. Actually,
  227. 8:08>> I see a different movement. I see the
  228. 8:10movement of the Frontier Labs are
  229. 8:13allowing
  230. 8:15allowing small and medium businesses and
  231. 8:18big businesses to build to move to an
  232. 8:20agent to agents
  233. 8:23which which bypasses the small models.
  234. 8:26I mean
  235. 8:27they they they still have
  236. 8:30their relevance today where they're
  237. 8:32building
  238. 8:33I mean now you're I mean we are
  239. 8:36the AI company in the data space.
  240. 8:39We have now our own legal agents
  241. 8:43which we
  242. 8:45we work we we put
  243. 8:49downstream of our
  244. 8:50our legal counsels
  245. 8:53accelerating and and reducing price and
  246. 8:55accelerating legal reviews. We do we do
  247. 8:58have a lot of legal reviews in our
  248. 9:01model
  249. 9:02so many partners that we need to onboard
  250. 9:04and make sure they are legit and all of
  251. 9:07that. So you can see this is this is a
  252. 9:09good example of
  253. 9:11of how
  254. 9:13the Frontier Labs are enabling
  255. 9:15companies to build agents and and but
  256. 9:18there is and I I think that is the next
  257. 9:21thing not so much the small model
  258. 9:23producers
  259. 9:25those
  260. 9:28those are definitely out there but I
  261. 9:30don't I believe that who's going to win
  262. 9:32the battle
  263. 9:33is going to be actually the agentic all
  264. 9:35the all the folks and the architects
  265. 9:37building agents
  266. 9:39more than the small models.
  267. 9:41>> I I've obviously got to ask you about
  268. 9:44copyright and AI training data given
  269. 9:46what you do for a job. It seems more
  270. 9:48legal battles
  271. 9:50are not going the way of the model
  272. 9:51developer and they're going towards the
  273. 9:53people that are sort of bringing these
  274. 9:55claims against thinking most recently
  275. 9:57Anthropic that I can think of.
  276. 9:59I assume maybe we might both agree on
  277. 10:01this point but how can model developers
  278. 10:03like OpenAI and Anthropic
  279. 10:05balance the cost of licensing data when
  280. 10:09the business from these sorts of
  281. 10:11these communities of people that are
  282. 10:13looking to sort of reclaim the ownership
  283. 10:15of of their
  284. 10:16of their data.
  285. 10:17When the business models of of those
  286. 10:19companies are already priced
  287. 10:22kind of put to perfection and haven't
  288. 10:25kind of taken into account the fact that
  289. 10:27they may have to be paying quite as much
  290. 10:29as they they could do going forwards to
  291. 10:32to the people that I would say they they
  292. 10:34stole a lot of data from.
  293. 10:36>> I wouldn't feel sorry for these
  294. 10:39Frontier Labs model developers in terms
  295. 10:41of balancing costs.
  296. 10:45They they are extremely well funded.
  297. 10:49They are
  298. 10:51they are already
  299. 10:53they have enough
  300. 10:56data
  301. 10:57compute power to crawl the web.
  302. 11:00So the least the only thing they need to
  303. 11:02do is to account for
  304. 11:05paying copyrights
  305. 11:07for content and data that is that they
  306. 11:10can find
  307. 11:11otherwise or it's so unique that needs
  308. 11:14to be paid for. So I wouldn't feel sorry
  309. 11:16for their business model.
  310. 11:18They definitely are
  311. 11:20so much so that they both OpenAI and
  312. 11:22Anthropic are filing IPOs. They they are
  313. 11:25in a good shape
  314. 11:27to
  315. 11:29to in they have turned around their
  316. 11:31revenue model. If two years ago or three
  317. 11:33years ago there was not it wasn't clear
  318. 11:36how these these companies were going to
  319. 11:38make money. Now it is very clear.
  320. 11:41So
  321. 11:42I I say that
  322. 11:44litigations continue to happen
  323. 11:47in the current
  324. 11:50political environment
  325. 11:52less so because
  326. 11:54the moat is deregulation in the US.
  327. 11:58But this goes in cycles and I believe
  328. 12:01that
  329. 12:02more should come and legislation will
  330. 12:05come eventually. We
  331. 12:06we cannot see we cannot continue as you
  332. 12:09can see
  333. 12:11whenever a bigger outlet or a bigger
  334. 12:14bigger union comes comes together just
  335. 12:16like the authors guild for
  336. 12:20against entropy
  337. 12:21they are able to
  338. 12:24to win their battle.
  339. 12:26That being said it's still a small price
  340. 12:29to settle 1.5 billion which was the
  341. 12:32largest so far in a settlement it's
  342. 12:35still a small price to pay considering
  343. 12:38the value of the data that
  344. 12:40was taken.
  345. 12:42>> I will say I I I might go off topic here
  346. 12:44with one of my questions. You you said
  347. 12:47that the the path to making money for
  348. 12:50these these companies is quite clear. I
  349. 12:53should be very upfront that the the
  350. 12:54audience of the tech report would I
  351. 12:56think largely disagree
  352. 12:58with the premise maybe they can make
  353. 12:59money but not profit.
  354. 13:02I'm curious I just would love to hear
  355. 13:03your your thoughts on how you think
  356. 13:06that that would come around because you
  357. 13:07seem very confident with it.
  358. 13:10>> Well
  359. 13:11I
  360. 13:13I work in this field for a long time.
  361. 13:16I know
  362. 13:18I know what it goes I mean you could see
  363. 13:20actually in terms of cost reductions
  364. 13:23how
  365. 13:24this year the world in the US
  366. 13:26experienced
  367. 13:28thousands
  368. 13:29tens of thousands of layoffs
  369. 13:32in the biggest frontier lab
  370. 13:35builders.
  371. 13:37If if that's not a way to balance profit
  372. 13:41profitability in a world where more of
  373. 13:44the white collar work is being replaced
  374. 13:48but it but that it's not so much to
  375. 13:50offload the price of data as a matter of
  376. 13:53fact we we haven't seen
  377. 13:56a huge growth in data
  378. 13:59purchases
  379. 14:00it's to offload the the price of data
  380. 14:02centers cost
  381. 14:04and energy and energy cost so
  382. 14:07I continue to disagree I think I think
  383. 14:10we are going we're seeing
  384. 14:13I think their business models are going
  385. 14:15to work out as we can see that. They
  386. 14:16they will it they have to balance
  387. 14:19people,
  388. 14:20infrastructure, and data.
  389. 14:22And certainly right now data is not
  390. 14:26at the center.
  391. 14:28And clearly people is getting cut.
  392. 14:31[laughter]
  393. 14:32It may change next, but that's what was
  394. 14:34this happening this year.
  395. 14:36>> That does actually we we gave um the the
  396. 14:39audience of the Tech Report a chance to
  397. 14:41sort of come up with some of the
  398. 14:42questions that we might ask,
  399. 14:44particularly executives, CFOs, CEOs,
  400. 14:46founders, those sorts of people.
  401. 14:48And one of them was what you think needs
  402. 14:51to be done to sort of address the
  403. 14:53anxiety that's kind of been created
  404. 14:55around
  405. 14:56uh losing losing jobs to AI because on
  406. 14:59one side it's seen as a very
  407. 15:00revolutionary technology, but for a lot
  408. 15:02of people it's seen as
  409. 15:04something that is a a threat to the
  410. 15:05livelihood their livelihoods to put it
  411. 15:07in quite a abrupt way. I was wondering
  412. 15:09what you think can be
  413. 15:11what can be done about that and and it
  414. 15:12would be great if you have any examples
  415. 15:14of of of that yourself.
  416. 15:15>> No,
  417. 15:16definitely. I mean the anxiety is real
  418. 15:19and this is the first year where we I
  419. 15:21mean this we knew the whole world knew
  420. 15:23this was coming.
  421. 15:24Every
  422. 15:25technology revolution
  423. 15:27uh
  424. 15:28it happens with loss of jobs and
  425. 15:30creation of new jobs. And this was the
  426. 15:33pivotal year. Uh we saw these layoffs
  427. 15:36are the
  428. 15:38testimony to that.
  429. 15:39Uh and they happened
  430. 15:42most like they happened we saw that they
  431. 15:44happened in areas
  432. 15:46that can be more automated obviously.
  433. 15:48The
  434. 15:49the the entry-level
  435. 15:52white-collar jobs, analysts, and things
  436. 15:55like that. But but the world
  437. 15:59humanity and technology adjusts. We are
  438. 16:02also seeing huge demand
  439. 16:05uh for people who know how to use AI
  440. 16:08tools, people who know how to build
  441. 16:11agents,
  442. 16:13people who know how to red team their
  443. 16:15agents
  444. 16:16and curate
  445. 16:18data and and their models and keep
  446. 16:21testing them against hallucinations and
  447. 16:24creating the governance and guardrails
  448. 16:26on around the data. So
  449. 16:28all these things are happening
  450. 16:31at
  451. 16:32and and all the physically AI world that
  452. 16:34is exploding in every sector of society
  453. 16:38asking for skills of doctors and
  454. 16:41mechanics and electricians and everybody
  455. 16:44who has hands on
  456. 16:46real the real thing
  457. 16:49in the 3D world. This is this is
  458. 16:51happening. So
  459. 16:55I would the advice I would say I would
  460. 16:57give is
  461. 16:59try and use as many AI tools as you can
  462. 17:02because the differentiator to get a job
  463. 17:05is the person to get or sustain a job is
  464. 17:08the person who actually can
  465. 17:10leverage this knowledge, these tools
  466. 17:14for to improve their productivity and
  467. 17:15their organization's productivity.
  468. 17:18>> And just one final question. I know
  469. 17:19you've got to you've got to run off
  470. 17:20soon, but I am asking everybody that we
  471. 17:22interview.
  472. 17:24Very curious to hear what you're going
  473. 17:25to say with this one as well. Is AI in a
  474. 17:27bubble and what do you think is going to
  475. 17:29happen between now and and 2030?
  476. 17:32>> Well, we hear that a lot.
  477. 17:34I
  478. 17:35I don't
  479. 17:37I don't think
  480. 17:38AI is in a bubble.
  481. 17:41I can I actually I actually think that
  482. 17:44I AI is getting
  483. 17:46decentralized
  484. 17:48which is good. It's a great thing.
  485. 17:51It's getting more in hands. You don't
  486. 17:54It's going to be very much like the
  487. 17:55internet. You don't have to have to know
  488. 17:58HTML
  489. 18:00to create a website back in the day.
  490. 18:03But in the beginning you did.
  491. 18:05So I I see a world where
  492. 18:08this decentralization is for for one
  493. 18:11offload the the
  494. 18:13concentration
  495. 18:14in certain in a few powers in the in
  496. 18:17terms of AI generation and owning and
  497. 18:21the knowledge.
  498. 18:23And it's and we see that all over the
  499. 18:25world already.
  500. 18:27And the open source world plays a huge
  501. 18:31play here.
  502. 18:32More standards are coming together. The
  503. 18:34Linux Foundation has a whole now new
  504. 18:37standard on agentic
  505. 18:39how to build agents in the agentic
  506. 18:41platforms.
  507. 18:42So I do I do think that
  508. 18:45the bubble
  509. 18:47if anything the bubble is always created
  510. 18:50around
  511. 18:52political instability rather than
  512. 18:54the actual technology advances.
  513. 18:58>> Well, Daniela Braga, thanks for taking
  514. 19:00the time.
  515. 19:00>> Thank you so much, Isaac.

About this transcript

This page contains the full transcript of Tech CEO: Big AI’s monopoly is breaking | Daniela Braga, Defined.ai by The Tech Report, generated from the public captions YouTube serves with the video. The transcript has 2,777 words across 515 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.