How we can use AI for as a "partner in thought", losing faith in long context windows for improved reasoning, and why we should stop anthropomorphizing LLMs
https://www.latent.space/p/brightwave
- BrightWave is a startup founded by Mike Conover, focusing on AI-driven financial analysis
- Raised $6 million seed round led by Decibel, with participation from Point72 and Moonfire Ventures
- Aims to expand individuals' ability to reason about the structure of the economy and markets using AI
- Acts as a "partner in thought" for finance professionals
- Uses multiple AI subsystems for specific tasks rather than a single large model
- Employs RAG (Retrieval Augmented Generation) with context-aware prompting
- Focuses on extracting structured information into knowledge graphs
- Prioritizes grounded reasoning and factuality in outputs
- Can analyze complex financial scenarios and provide insights
- Handles temporality of data, crucial for financial analysis
- Balances private and public data sources in analysis
- Provides highly facetable, pivotable product interface
- Co-founded with Brandon Katara, who has experience in finance and deep learning
- Hiring across AI, engineering, machine learning, and design roles
- Believes AI hedge funds may already exist in some form
- Sees potential for AI in idea generation and thematic investing
- Emphasizes human role in final decision-making and strategy alignment
- Notes convergence in model behavior and diminishing returns on pretraining
- Predicts future innovation in instruction tuning and fine-tuning data creation
- Emphasizes importance of private evaluation datasets
- Observes decreasing incentives for companies to train their own foundation models
- Predicts focus shifting to differentiating models through specific behavioral fine-tuning
- 00:42 BrightWave is a startup founded by Mike Conover, focusing on AI-driven financial analysis
- 04:51 Raised $6 million seed round led by Decibel, with participation from Point72 and Moonfire Ventures
- 05:38 Aims to expand individuals' ability to reason about the structure of the economy and markets using AI
- 09:36 Acts as a "partner in thought" for finance professionals
- 31:56 Uses multiple AI subsystems for specific tasks rather than a single large model
- 35:49 Employs RAG (Retrieval Augmented Generation) with context-aware prompting
- 38:13 Focuses on extracting structured information into knowledge graphs
- 20:04 Prioritizes grounded reasoning and factuality in outputs
- 09:36 Can analyze complex financial scenarios and provide insights
- 31:27 Handles temporality of data, crucial for financial analysis
- 34:00 Balances private and public data sources in analysis
- 40:00 Provides highly facetable, pivotable product interface
- 06:07 Co-founded with Brandon Katara, who has experience in finance and deep learning
- 1:03:55 Hiring across AI, engineering, machine learning, and design roles
- 50:42 Believes AI hedge funds may already exist in some form
- 55:51 Sees potential for AI in idea generation and thematic investing
- 56:02 Emphasizes human role in final decision-making and strategy alignment