DeepSeek's first LLM paper was literally titled "Scaling Open-Source Language Models with Longtermism"
teortaxesTex · x · 2026-10-15
Following up his debunk, teortaxesTex notes DeepSeek's very first LLM paper was titled "DeepSeek LLM: Scaling Open-Source Language Models with Longtermism." The paper's scaling law findings: the higher the data quality, the more compute budget should go to model scaling — high-quality data enables training larger models at the same data scale.
More from Companies & People
- The way to pace the frontier? A confusing product lineup — dreamwieber · 2026-10-15
- Gary Marcus questions Anthropic: decelerating ARR growth vs a $2T valuation — GaryMarcus · 2026-10-15
- Future Silicon World Streams with opencli (30k Stars) Author on Models and AI Coding — vista8 · 2026-10-15
- Kimi researcher Nathan leaves Moonshot after a year of KDA and Attention Residuals work — teortaxesTex · 2026-10-15
- Google Discover clicks on X posts now open X's For You feed, hinting at a Google-X deal — gaganghotra_ · 2026-10-15
- Prime Intellect and Hugging Face kick off Open Source AI Week — huggingface · 2026-10-15