ThursdAI - The top AI news from the past week

📆 ThursdAI - Jan 23, 2025 - 🔥 DeepSeek R1 is HERE, OpenAI Operator Agent, $500B AI manhattan project, ByteDance UI-Tars, new Gemini Thinker & more AI news

71 snips
Jan 24, 2025
This week in AI is nothing short of explosive! DeepSeek's open-source R1 model is making waves, promising advanced reasoning capabilities. Meanwhile, a staggering $500 billion infrastructure initiative is on the horizon, shaking up the landscape. OpenAI's new 'Operator' is aiming to bridge the gap between chat and action but faced some live adventures. Plus, ByteDance's UI-Tars model adds to the innovation buzz. All these advancements signal a revolutionary moment in artificial intelligence!
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

Open-Source R1

  • DeepSeek's R1 reasoning model is now open-source under the MIT license.
  • This allows for unprecedented freedom to use, modify, and distribute the model.
ANECDOTE

R1 1.5B Performance

  • DeepSeek R1's 1.5B parameter model outperforms GPT-4 and Claude on math benchmarks.
  • Nisten ran the model locally and found it surprisingly performant, even generating a Tetris game.
ADVICE

Speed Up R1 Inference

  • Use Llama.cpp and Llama Server with speculative decoding to speed up R1 inference significantly.
  • Alex Volkov saw speeds increase from 5 to 11 tokens per second on an M3 Mac.
Get the Snipd Podcast app to discover more snips from this episode
Get the app