Tech CEO: Big AI’s monopoly is breaking | Daniela Braga, Defined.ai
Build specialized AI with curated private data, and learn to use AI tools and agents to improve your productivity.
- Use licensed private data to differentiate models; synthetic data can help fill gaps but may cause collapse if overused.
- Fine-tune open-weight models with small, expert-curated datasets for specific industry tasks.
- Learn to use AI tools, build and test agents, and add safeguards against hallucinations.
The interview offers perspectives and examples, but no detailed demonstrations or implementation steps. · Best for: AI leaders, data teams, and professionals adapting to AI-driven work.
Defined.ai CEO Daniela Braga discusses data constraints, specialized models, agents, copyright, jobs, and AI decentralization. She argues that private data and domain expertise enable differentiation, while frontier-model companies are also expanding access through agents.
Data availability and model collapse1:47
- Public web data is increasingly exhausted, but private, closed-source data remains a major potential differentiator and must be paid for.
- Synthetic data can supplement real data for otherwise missing scenarios; using it in the wrong proportions can contribute to model collapse.
Specialized models and open-weight adoption4:13
- Clients increasingly seek smaller, carefully curated datasets reviewed by domain experts such as doctors, mathematicians, chemists, and linguists.
- Startups and businesses can fine-tune open-source models with private data for particular industries instead of building large models from scratch.
Agents and frontier-model companies7:55
- Braga sees agents as a major direction, with frontier labs enabling businesses to build them.
- Defined.ai uses legal agents downstream of its legal counsel to speed up reviews and reduce their cost.
Copyright, costs, and profitability10:18
- Braga argues that well-funded labs can pay for unique or licensed data and says litigation and legislation will continue.
- She says companies are balancing people, infrastructure, and data costs, while noting that data purchases have not seen huge growth.
Jobs and skills14:16
- She says automation is affecting some entry-level white-collar work, while demand is growing for people who use AI, build and red-team agents, curate data, and test for hallucinations.
- Her practical advice is to use AI tools to improve personal and organizational productivity; she also points to demand for hands-on skills in physical AI.
Decentralization and the AI outlook16:41
- Braga does not consider AI to be in a bubble and expects it to become more decentralized, with open source and shared standards playing a role.
- She compares AI's growing accessibility to website creation becoming easier over time.
Notes are AI-generated from the video's transcript and may miss nuance.