Key Moments
How AI is Eating Finance - with Mike Conover of Brightwave
Latent Space PodcastKey Moments
AI partner for finance pros, enhancing reasoning about markets and economies.
Key Insights
Brightwave offers an AI 'partner-in-thought' for finance professionals, moving beyond simple information retrieval to deeper reasoning about markets.
The company emphasizes a 'systems of systems' approach, combining specialized AI subsystems rather than relying solely on large context windows.
Domain expertise is crucial in finance; Brightwave integrates deep financial knowledge with AI capabilities for non-consensus insights.
Customer feedback and revealed preferences are key for personalization, as explicitly articulating beliefs is impractical.
Factuality is addressed through multiple evaluation steps, including LLM supervision and product affordances for users to double-check AI outputs.
Brightwave focuses on extracting structured information and building knowledge graphs for deeper, pivotable reasoning, complementing generative AI.
Fine-tuning is used for differentiating LLMs into specific behavioral regimes rather than imbuing them with new factual information.
BRIGHTWAVE'S VISION: AN AI PARTNER FOR FINANCE
Brightwave, founded by Mike Conover and Brandon Katara, aims to provide finance professionals with an AI "partner-in-thought." The core idea is to augment human capabilities in understanding complex market structures and economic systems, moving beyond simple data recall. The company believes that AI can significantly expand individual reasoning abilities, especially in fields like finance where human attention spans are limited. They have secured a $6 million seed round, including participation from major players like Decibel and a large hedge fund, indicating strong market confidence in their approach.
INTEGRATING DOMAIN EXPERTISE WITH AI CAPABILITIES
Building a successful AI startup in a vertical domain like finance requires a blend of deep AI expertise and industry-specific knowledge. Brightwave's team includes individuals with extensive experience in finance, such as co-founder Brandon Katara, a former CTO of a regulated derivatives exchange. This dual expertise ensures the AI systems not only possess methodological sophistication but also understand the nuances of financial markets, regulations, and the nature of valuable insights. This domain knowledge is crucial for generating actionable, non-consensus ideas that are the hallmark of active asset management.
SYSTEMS OF SYSTEMS: BEYOND LARGE CONTEXT WINDOWS
Contrary to the trend of simply increasing context window sizes, Brightwave adopts a 'systems of systems' approach. They find that large context windows, while good for information retrieval, struggle with deep synthesis and complex reasoning. Instead, they build specialized AI subsystems, each designed to perform specific tasks effectively. This modular approach allows for more robust and predictable system behavior, treating modern AI, particularly Retrieval-Augmented Generation (RAG) and agent-based reasoning, as a framework for orchestrating interconnected ML subsystems.
ADDRESSING FINANCE-SPECIFIC CHALLENGES: FACTUALITY AND TEMPORALITY
Finance presents unique challenges, including the critical need for factual accuracy and handling temporal data. Brightwave employs rigorous methods to ensure factuality, using multiple evaluation steps, LLM supervision, and user-driven verification. They acknowledge that hallucinations can occur and build product affordances to manage this. For temporality, the system is designed to be aware of the time-sensitive nature of financial data, ensuring that retrieval systems can distinguish between breaking news and historical thematic analysis, thus avoiding a 'bag of facts' without proper context.
USER-CENTRIC DESIGN AND PERSONALIZATION
Brightwave focuses on personalization through observed 'revealed preferences' rather than explicit user input. Instead of asking users to define their beliefs, the system identifies natural next questions a user might ask, inferring their interests and strategic direction. This adaptive learning allows the AI to evolve with the user, providing tailored insights and essentially learning the user's implicit worldview and investment strategy over time. This user-centric approach also fosters trust by allowing users to engage in a dialogue with the AI, directing its focus and assessing its findings.
KNOWLEDGE GRAPHS AND DATA EXTRACTION FOR DEEPER INSIGHTS
The company is investing heavily in knowledge graph extraction, seeing it as a powerful complement to generative AI. By extracting structured information from vast datasets, Brightwave can build highly granular knowledge graphs that reveal complex relationships (e.g., 'person X testified critically of organization Y'). This structured data artifact allows for more powerful, pivotable reasoning than simple semantic search. It enhances the system's ability to identify second and third-order derivative bets and provides a robust foundation for quantitative and qualitative analysis, moving beyond basic document summarization.
STRATEGIC USE OF FINE-TUNING VS. RAG
Brightwave views fine-tuning not as a way to infuse new information but as a method for differentiating LLMs into specific behavioral regimes. They liken LLMs to stem cells that differentiate into specialized functions. Fine-tuning is employed to create specific subsystems that occupy distinct behavioral states, essentially turning an LLM into a finite state machine optimized for particular tasks. Retrieval-Augmented Generation (RAG) is preferred for grounded reasoning, ensuring the AI attends to verifiable facts available at inference time, making it a reliable tool for analysis.
IMPACT ON FINANCIAL ANALYSIS AND FUTURE OF WORK
Brightwave deliberately avoids creating simple Excel spreadsheets, focusing instead on a 'partner-in-thought' modality. This approach allows for a more dynamic, dialogic interaction where users can question, refine, and pivot on AI-generated insights. While the company excels in quantitative reasoning, the emphasis is on qualitative analysis and strategic insight generation. This aligns with the evolution of financial roles, where tools abstract away syntax-driven tasks, allowing professionals to focus on higher-level analysis and strategic decision-making, potentially reshaping how financial models are created and utilized in the future.
Mentioned in This Episode
●Products
●Software & Apps
●Companies
●Organizations
●Books
●Concepts
●People Referenced
Common Questions
Brightwave is a company founded by Mike Conover and Brandon Katara, offering an AI-powered 'partner in thought' for finance professionals. It helps users identify mispriced assets, develop investment theses, and gain deep insights into complex financial and market dynamics.
Topics
Mentioned in this video
Financial technology that emerged in the late 1970s, used as an analogy for how powerful tools can shift the core tasks of a profession, like accounting.
Company where Mike Conover previously led the open-source large language models team.
A financial data provider whose current knowledge graph resolution is compared to what Brightwave aims to achieve.
A company focused on providing AI-powered tools for financial professionals, acting as a partner in thought.
Company whose GPU market position in relation to rare earth metal shortages was used as an example of a Brightwave query.
Social media platform whose dataset was used in Mike Conover's PhD research on network structures and polarization.
Social media platform used for global economy insights and where Mike Conover worked on homepage Newsfeed relevance and the Economic Graph Challenge.
Represents a type of financial institution that employs many people who could potentially benefit from advanced AI tools.
Consumer internet company used as an example for personalization strategies based on user behavior rather than explicit input.
A hedge fund mentioned as an example of a firm that has long used models for high-frequency trading, indicating AI's presence in finance.
An open-source large language model developed at Databricks.
A large language model mentioned in the context of its difficulty in handling very long context windows and its role as a potential foundational standard.
An example of a company with generative work product valuable to most humans, contrasted with the specialized domain of finance.
A platform mentioned as an example of a source with many opinions, some good and some mediocre, highlighting the need for systems aware of user preferences for different sources.
A framework mentioned in the context of using multiple weak supervision data sets for evaluation.
An open-source dataset that was released, signifying a significant moment in the open-source LLM space.
A venture capital firm that participated in Brightwave's seed round.
Agency that funded Mike Conover's PhD research on propaganda and misinformation campaigns using Twitter data.
Lead investor in Brightwave's $6 million seed round.
Company where Mike Conover was director of financial machine learning and where his co-founder Brandon Katara worked.
More from Latent Space
View all 178 summaries
86 minNVIDIA's AI Engineers: Brev, Dynamo and Agent Inference at Planetary Scale and "Speed of Light"
72 minCursor's Third Era: Cloud Agents — ft. Sam Whitmore, Jonas Nelle, Cursor
77 minWhy Every Agent Needs a Box — Aaron Levie, Box
42 min⚡️ Polsia: Solo Founder Tiny Team from 0 to 1m ARR in 1 month & the future of Self-Running Companies
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Try Summify free