Ben Wellington: ML for Finance and Storytelling through Data

Topics covered
Popular Clips
Episode Highlights
NLP Evolution
Ben Wellington reflects on the evolution of natural language processing (NLP) from its early days to the present. Initially, NLP relied heavily on human linguistic knowledge, with models requiring detailed syntactic understanding to function effectively. However, a significant shift occurred when large-scale data models, like those used by Google, outperformed traditional methods by sheer data volume alone. This transition marked the beginning of a new era in NLP, where data-driven approaches became dominant.
You're telling me that they came and did the simplest thing, but they had the biggest data, and they beat us all.
---
Ben's insights highlight the ongoing debate between data-driven models and those incorporating human linguistic knowledge, a discussion that continues to shape the field today 1 2.
  Â
Statistical Role
Statistics have played a crucial role in shaping modern NLP practices, intersecting with linguistic theory to drive advancements. Ben Wellington discusses how the use of large datasets has become central to NLP, particularly at Two Sigma, where data is leveraged to predict financial market movements. The surprising effectiveness of large data models, which can mimic human language without understanding it, has been a revelation for many in the field.
The shock is, wow, they really sound like humans.
---
This realization underscores the power of data in transforming NLP and highlights the importance of statistical methods in developing sophisticated language models 3 4.
  Â
Research Culture
Two Sigma's integration of AI and NLP into their research culture reflects a balance between short-term goals and long-term innovation. Ben Wellington describes how the company collaborates with academic institutions and supports open-source projects to stay at the forefront of AI advancements. This approach allows Two Sigma to adapt to the rapidly evolving tech landscape while contributing to the broader research community.
We have to balance the short term with the long term, like a lot of organizations.
---
Ben's journey from academia to industry illustrates the dynamic nature of AI research and the importance of collaboration in driving progress 5 6.
Related Episodes


Hugo Larochelle: Deep Learning as Science
Answers 383 questions

Max Woolf: Data Science at BuzzFeed and AI Content Generation
Answers 383 questions

Martin Wattenberg: ML Visualization and Interpretability
Answers 383 questions

Kate Park: Data Engines for Vision and Language
Answers 383 questions

2024 in AI, with Nathan Benaich
Answers 383 questions

Miles Grimshaw: Benchmark, LangChain, and Investing in AI
Answers 383 questions

Ben Green: "Tech for Social Good" Needs to Do More
Answers 383 questions

Nathan Benaich: The State of AI Report
Answers 383 questions

2023 in AI, with Nathan Benaich
Answers 383 questions

Luis Voloch: AI and Biology
Answers 383 questions

Nicholas Thompson: AI and Journalism
Answers 383 questions

Antoine Blondeau: Alpha Intelligence Capital and Investing in AI
Answers 383 questions

Chelsea Finn on Meta Learning & Model Based Reinforcement Learning
Answers 383 questions

Chip Huyen: Machine Learning Tools and Systems
Answers 383 questions

Lukas Biewald: Crowdsourcing at CrowdFlower and ML Tooling at Weights & Biases
Answers 383 questions
