How DeepSeek is so cheap

Q: How did DeepSeek get around export restrictions? A: They didn’t. They just tinkered around with their chips to make sure they handled memory as efficiently as possibly. They lucked out, and their perfectly optimized low-level code wasn’t actually held back by chip capacity.

Q: How did DeepSeek train so much more efficiently? A: They used the formulas below to “predict” which tokens the model would activate. Then, they only trained these tokens. They need 95% fewer GPUs than Meta because for each token, they only trained 5% of their parameters.

Q: How is DeepSeek’s inference so much cheaper? A: They compressed the KV cache. (This was a breakthrough they made a while ago.)

Q: How did they replicate o1? A: Reinforcement learning. Take complicated questions that can be easily verified (either math or code). Update the model if correct.

Reference: https://x.com/wordgrammer/status/1883712727073607859

Jurnal Sistem Cerdas

Alamat: https://apic.id/jurnal/index.php/jsc/index

Jurnal Sistem Cerdas with eISSN: 2622-8254 is a peer-reviewed journal serving as a publication medium for research findings that support the research and development of cities, villages, sectors, and other systems. The Intelligent Systems Journal is published by the Smart Indonesia Initiative Association (APIC) and is released every four months (April, August, and December). This journal is expected to serve as a platform for publishing research findings from practitioners, academics, authorities, and related communities.

The purpose of the Intelligent Systems Journal is to contribute to the intellectual life of the nation by the mandate contained in the preamble of the 1945 Constitution. This journal also serves as a platform for the publication of innovations, technologies, and policies of the APIC community, related to education and the intelligence of large-scale system components.

The scope of the systems discussed is attached but not limited to;

  1. System engineering
  2. Artificial Intelligence Technology (AI) and Machine Learning
  3. Internet of Things
  4. Big Data
  5. Systems and components for urban, rural or other Smart areas
  6. Smart mobility and transportation systems and components
  7. Systems and Smart energy components
  8. Smart tourism systems and components
  9. Systems and components of smart city security and comfort
  10. Smart infrastructure systems and components
  11. Smart health systems and components
  12. Smart Education systems and components
  13. Robots and Smart Systems.

Prestasi Akademisi Berhubungan Dengan Kekayaan Orangtuanya

Sumber: “Climbing the Ivory Tower: How Socio-Economic Background Shapes Academia” https://www.nber.org/papers/w33289

Studi ini mengeksplorasi bagaimana latar belakang keluarga dan status sosial ekonomi seseorang membentuk karier akademis mereka, mulai dari menjadi dosen di universitas hingga jenis penelitian yang mereka lakukan dan pengakuan yang mereka terima.

  1. Orang-orang dari latar belakang sosio-ekonomi yang lebih tinggi sangat banyak terwakili di antara para dosen di universitas, dimana sekitar 50% berasal dari 20% teratas dalam distribusi pendapatan.
  2. Pilihan bidang akademik seorang dosen sangat erat kaitannya dengan pekerjaan ayahnya, dan anak-anak sering kali menekuni bidang yang serupa dengan pekerjaan ayahnya.
  3. Meskipun rata-rata profesor yang berlatar belakang sosio-ekonomi rendah memiliki produktivitas yang sama, mereka cenderung tidak memiliki publikasi atau termasuk dalam 1% teratas yang paling produktif. Mereka juga memperkenalkan lebih banyak konsep ilmiah baru dalam karya mereka.
  4. Para profesor yang berlatar belakang sosio-ekonomi rendah menerima lebih sedikit sitasi dan kecil kemungkinannya untuk dinominasikan atau memenangkan Hadiah Nobel, bahkan ketika memperhitungkan catatan publikasi dan sitasi mereka.