Små AI-modeller for skrivebordet kan konkurrere med skysystemer, viser Stanford-studie

27 kilder
  • En Stanford-studie fant at små språkmodeller på stasjonære datamaskiner svarte nøyaktig på 88,7 % av reelle spørringer, noe som konkurrerer med store skybaserte AI-systemer.
  • Investeringsstrateg Joachim Klement argumenterte i en Reuters-spalte onsdag for at funnene truer lønnsomheten til Anthropic, OpenAI og xAI.
  • Studiens «intelligens per watt»-metrikk forbedret seg 5,3 ganger fra 2023 til 2025, og en optimal rutingmetode kan kutte energikostnadene med over 80 %, fant forskerne.
Kilder (27)
  1. 1 Intelligence Per Watt: A Study of Local Intelligence Efficiency scalingintelligence.stanford.edu
  2. 2 Study Finds Small Desktop AI Challenges Data-center ... letsdatascience.com
  3. 3 The future of AI may be small, cheap and unprofitable klementoninvesting.substack.com
  4. 4 The future of AI may be small, cheap and unprofitable: Joachim Klement www.inkl.com
  5. 5 Avanika Narayan's Post - LinkedIn www.linkedin.com
  6. 6 Intelligence Per Watt: A Study of Local Intelligence Efficiency huggingface.co
  7. 7 The future of AI is in your hands research.ibm.com
  8. 8 Small Language Models Deliver 10-30× Efficiency Gains in 2026 byteiota.com
  9. 9 Small Language Models and Edge AI: The 2026 Shift to Local ... zylos.ai
  10. 10 Distilling Step-by-Step! Outperforming Larger Language Models with ... research.google
  11. 11 [PDF] arXiv:2501.05465v1 [cs.CL] 3 Jan 2025 arxiv.org
  12. 12 Distilling step-by-step: Outperforming larger language models with less training research.google
  13. 13 The environmental impact of LLMs vs. SLMs www.techtarget.com
  14. 14 [PDF] Demystifying Small Language Models for Edge Deployment aclanthology.org
  15. 15 A Comprehensive Survey of Small Language Models in the Era of Large Language Models: Techniques, Enhancements, Applications, Collaboration with LLMs, and Trustworthiness arxiv.org
  16. 16 ‘Tiny' AI model beats massive LLMs at logic test www.nature.com
  17. 17 Minions: the rise of small, on-device LMs - Hazy Research hazyresearch.stanford.edu
  18. 18 Energy Efficiency Benchmarks: SLM vs. Cloud LLM in the 2026 ... aaia.app
  19. 19 Intelligence Per Watt: Measuring Local Inference Viability ... semiengineering.com
  20. 20 intelligence per watt x.com
  21. 21 AI cuts tech project times from years to months, says ... economictimes.com
  22. 22 AI to broaden investment opportunities in 2026, smaller firms to benefit: Report economictimes.com
  23. 23 Small Models May Be Right, But… | Current Affairs | Vision IAS visionias.in
  24. 24 How fast is AI going to develop? - by Joachim Klement klementoninvesting.substack.com
  25. 25 paper: https://t.co/56fcwFuPPg blogpost: x.com
  26. 26 H/T Asif Razzaq OpenJarvis: Personal AI, ... www.facebook.com
  27. 27 AI demand to keep memory tight through 2027, chipflation ... economictimes.com

Legg igjen en kommentar