Figure AI Robots 🤖 // OpenAI Leaks GPT-5 🤫 // Scaling Language Models 📈

Figure, a leading AI robotics company, is making significant advancements in creating robots that can perceive their environment, make decisions, and take action, all in a way that aligns with human expectations. OpenAI may have accidentally leaked details about a new AI model called GPT-4.5 Turbo, which could level the playing field with Google's AI model Gemini. Two papers explore the development and evaluation of large language models (LLMs) for code-related tasks, and propose simple and scalable strategies to continually pre-train LLMs to save on compute. Another paper investigates scaling in the over-trained regime and relates language model perplexity to downstream task performance via a power law, providing useful insights into how language models can be scaled and evaluated more effectively. Contact:  [email protected] Timestamps: 00:34 Introduction 01:28 Figure AI 03:02 Did OpenAI just accidentally leak the next big ChatGPT upgrade? 04:48 Gradio's Grog 05:47 Fake sponsor 07:37 Simple and Scalable Strategies to Continually Pre-train Large Language Models 09:31 LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code 11:11 Language models scale reliably with over-training and on downstream tasks 13:07 Outro