📰 Paper reading club: NLP potpourri

We meet biweekly to discuss advances in computer science, ground them down to applications, and see where to the wind is blowing. Read the papers at your own time and come discuss them with us. The talks will be given in English, discussion in En/Ru.

1. 12:00 NLP Potpourri (@add512)

- Stealing Part of a Production Language Model: For under $20, our attack extracts the entire projection matrix of OpenAI's Ada and Babbage language models (the attack extends to gpt-3.5-turbo too).
- Humanoid Locomotion as Next Token Prediction: Our model enables a full-sized humanoid to walk in San Francisco zero-shot and suggest robotics scaling laws.
- The Unreasonable Effectiveness of Eccentric Automatic Prompts: Our findings reveal that the highest-scoring, automatically-optimized prompt exhibits a degree of peculiarity far beyond expectations.
- ChainForge: A Visual Toolkit for Prompt Engineering and LLM Hypothesis Testing: We present ChainForge, an open-source visual toolkit for prompt engineering and on-demand hypothesis testing of text generation LLMs. ChainForge provides a graphical interface for comparison of responses across models and prompt variations. Our system was designed to support three tasks: model selection, prompt template design, and auditing.

⏱ 12:00-14:00 Saturday, 23 Mar
📍 F0RTHSP4CE, Khorava St, 18
🗓 no registration/just come
💰 free/donation/boost
👅 en+ru
👮 @add512
💬 event chat and q&a
🗣 propose talk
🎥 recordings