LLM Interfaces Workshop and Hackathon
We will be bringing together some of the founders and developers who are building plugins and tools on top of and around LLMs to showcase their projects and inspire you to build!
This will be in our usual format of presentation (25 minute) followed by discussion (25 minute)
If you want to speak about any of these topics or similar ones, reach out to amir@ai.science
APRIL 28th (times are in ET):
9:00 Percy Chen (PhD Student @ McGill University); Using LangChain [includes demo | hands-on]
10:00 Denys Linkov (ML Lead @ Voiceflow); Running LLMs in Your Environment [includes demo]
11:00 Wojciech Gryc (Founder @ Phase AI); Building ResearchGPT
12:00 Amir Feizpour (PhD, CEO @ Aggregate Intellect); Task Automation with LLMs
13:00 Fatemeh Mireshghallah (PhD @ UC San Diego); Learning-free Controllable Text Generation
14:00 Suhas Pai (CTO @ Bedrock AI); Exploring the limits of today's LLMs
15:00 Ehsan Kamalinejad (PhD, Sr. MLE @ Amazon); Comparing the Performance of LLM Models and Techniques
16:00 Sina Shahandeh (PhD, Founder @ Plan with Flow); ChatGPT for Mathematical Financial Models
APRIL 29-30
Our friends at VoiceFlow are running a 48-hour online hackathon where teams will create projects to showcase the best use of GPT4, and other large language models.
SPEAKERS
AMIR FEIZPOUR
Amir is the co-founder of Aggregate Intellect (https://ai.science/), a Smart Knowledge Navigator platform for R&D teams in emerging technologies like AI. Prior to this, Amir was an NLP Product Lead at Royal Bank of Canada, and held a postdoctoral position at University of Oxford conducting research on experimental quantum computing. Amir holds a PhD in Physics from University of Toronto.
Task Automation Using LLMs - TBD
DENYS LINKOV
Denys is the ML lead at Voiceflow focused on building the ML platform and data science offerings. His focus is on realtime NLP systems that help Voiceflow’s 60+ enterprise customers build better conversational assistants. Previously he worked at large bank as a senior cloud architect.
Running LLMs in Your Environment - The availability of Large Language Models as a service (LLMaaS) has sparked a new wave of applications, use cases and companies. But what alternatives exist if you want to host your own LLMs? In this talk we’ll cover the landscape of LLMs and their deployment options. We’ll discuss the tradeoffs of hosting your own LLM vs using a commercial offering, including security, compliance, cost and delivery times. We’ll also cover a number of available open source options and how they can be hosted within your own environment whether a virtual private cloud or on prem
Fatemehsadat Mireshghallah
Fatemeh received her Ph.D. from the CSE department of UC San Diego and will join UW as a post-doctoral fellow. Her research interests are Trustworthy Machine Learning and Natural Language Processing. She is a recipient of the National Center for Women & IT (NCWIT) Collegiate award in 2020 for her work on privacy-preserving inference, a finalist of the Qualcomm Innovation Fellowship in 2021 and a recipient of the 2022 Rising star in Adversarial ML award.
Learning-free Controllable Text Generation for Debiasing - Large language Models (LLMs, e.g., GPT-3, OPT, TNLG,…) are shown to have a remarkably high performance on standard benchmarks, due to their high parameter count, extremely large training datasets, and significant compute. Although the high parameter count in these models leads to more expressiveness, it can also lead to higher memorization, which, coupled with large unvetted, web-scraped datasets can cause different negative societal and ethical impacts such as leakage of private, sensitive information and generation of harmful text. In this talk, we introduce a global score-based method for controllable text generation that combines arbitrary pre-trained black-box models for achieving the desired attributes in generated text from LLMs, without involving any fine-tuning or structural assumptions about the black-box models.
Ehsan Kamalinejad
Ehsan is a machine learning scientist. He is currently a lead scientist working on NLP developments at Amazon. Previously he co-founded Visual One which was a YCombinator startup in computer vision. Before that he was working at Apple for several years as a tech-lead machine learning scientist working on projects such as Photos Memories. Ehsan is also an associate professor at California State University. He got his PhD from the University of Toronto in applied mathematics.
Modern Innovations in Fine-Tuning Large Language Models - In this presentation, we will explore the latest breakthroughs in fine-tuning large language models. Our conversation will encompass various fine-tuning techniques, including instruction following fine-tuning and reinforcement learning through human feedback (RLHF). Additionally, we will delve into computational aspects like scaling laws, parameter-efficient fine-tuning (PEFT), and the zero redundancy optimizer (ZeRO).
Percy Chen
Percy is a PhD student at McGill University. His research interests include model-driven software engineering, trustworthy AI, verification for ML systems, Graph Neural Networks, and Large Language Models.
Building with LLMs Using LangChain - This workshop focuses on Large Language Models (LLMs) and their capabilities in language understanding and generation. Despite their impressive performance, LLMs still face challenges in tasks like retrieval and math reasoning. Fortunately, several tools are available for these tasks. LangChain is a Python library that enables the integration of LLMs with external tools to accomplish a wide range of tasks. The workshop will provide an overview of LangChain's basics and demonstrate how it can interface with external tools. Additionally, we will create a simple system using LangChain to answer questions about itself LangChain itself.
Sina Shahandeh
Sina holds a PhD in scientific computing and has led data science teams at three startups in Toronto's ecosystem. Before founding Plan with Flow, Sina served as the Vice President of Data Science at ecobee Inc., leading to an exit valued at $770 million. Currently, Sina is based in Madrid.
ChatGPT as an interface for construction of a mathematical financial model in a no-code application - LLMs alone generally struggle with complex mathematical tasks. This limitation is particularly evident when a problem requires intricate simulations rather than basic arithmetic or a few simple calculations. For instance, while GPT-4 can compute the interest paid on a loan, it cannot determine the loan's duration over several years. In this talk, we show how we used ChatGPT to build an interface for a no-code financial planning application and allow users to use the chat interface to inspect and inquire about the financial projection model in the background
SUHAS PAI
Suhas is the CTO & Co-founder of Bedrock AI, an NLP startup operating in the financial domain, where he conducts research on LLMs, domain adaptation, text ranking, and more. He was the co-chair of the Privacy WG at BigScience, the chair at TMLS 2022 and TMLS NLP 2022 conferences, and is currently writing a book on Large Language Models.
Exploring the limits of today's LLMs - GPT-4 and similar LLMs, powered with the ability to interface with external systems, have ushered in a whole variety of use cases into the realm of possibility that were previously considered impossible. However, possibility doesn't imply feasibility, with several limiting factors like effective retrieval hindering implementation. Tools like langchain and llamaindex are possibly the most impactful libraries of the year, but they are not a panacea to all problems. We will explore some of the limiting factors, approaches to tackle them, and showcase cutting-edge use cases that make the most of current capabilities.
ckground
Wojciech Gryc
Wojciech Gryc is the co-founder and CEO of Phase AI, where he helps startups and scaleups launch AI-driven products. He is the developer behind PhaseLLM and ResearchGPT. Wojciech started his career as an AI researcher at IBM Research, and completed his graduate studies at the Oxford Internet Institute. Prior to Phase AI, he was the founder and CEO of Canopy Labs, a customer data platform funded by Y Combinator and acquired by Drop.
Building ResearchGPT: automated statistical research and interpretation - ResearchGPT is a natural language statistics agent. Users provide the agent with a data set and natural language queries. The agent writes code to answer their questions and provide interpretations based on its analysis. Raw data is never shared with the LLM itself, and generated code is run locally. You can see a demo video here. This session will cover the underlying architecture of ResearchGPT and how it's been tested using PhaseLLM, a developer tooling framework for testing and robustifying LLM-powered apps.
PARTNERS
