Hello Dwarkesh Listeners
Like you, we have an unwavering belief that AI will fundamentally transform our world. See how we’re achieving this and come join us in building the future.
Research
Humanity's Last Exam
Leaderboards
SEAL Leaderboards: Expert-Driven Private Evaluations
Research
LLMs Easily Jailbroken as Browser Agents
Research
Multi-Turn Human Jailbreaks on LLM Defenses
Research
Examination of LLM Performance on Grade School Arithmetic
Research
Goal-Conditioned Representations for Reward Models
Research
Measuring and Reducing Malicious Use With Unlearning
Product Overview
Scale AI products are integral to boosting AI model performance. We combine frontier AI research with our expert talent network to help leading AI organizations build better, more capable AI systems.
Data Engine
Advance frontier model capabilities like agentic tool use and complex reasoning with SFT and RLHF data.
Evaluation
Identify loss buckets for model performance to pinpoint areas for improvement.
GenAI Platform
Build, test, and optimize Generative AI applications that unlock the value of your data.
Donovan
Leverage leading LLMs, fine-tuned models, and custom AI workflows for mission-critical use cases.
News
Research
Humanity's Last Exam
Blog
Scale's SEAL Research Lab Launches Expert-Evaluated LLM Leaderboards
Research
MASK Benchmark: Disentangling Honesty From Accuracy in AI Systems
Research
MultiChallenge: A Realistic Multi-Turn Conversation Evaluation Benchmark Challenging to Frontier LLMs
Research
ENIGMAEVAL: A Benchmark of Long Multimodal Reasoning Challenges
Blog
Scale Public Sector: Building on Our Progress in 2025
Blog
The New ChatGPT-4o Update Promises Better Writing, How Does It Compare To The New Claude 3.5 Sonnet?
Blog
First Impressions of OpenAI’s o1
Blog
Advancing Safe and Reliable AI: Scale's Research in Post-Training, Reasoning, and Evaluation