Other AI research in last week's AI Lab newsletter: The evolutionary tree of AI models. How to stop chatbots learning to be unsafe. And the insane power needed for future AI training runs. …
AI HARDWARE
-
Humanoid Robot Masters Fried Rice Cooking From Scratch
By
–
Look at this robot make fried rice from scratch!
— Amitav Bhattacharjee (@bamitav) 17 août 2025
pic.twitter.com/Zk8N1tmJk8#Cooking #chef #AI #ArtificialIntelligence #technology #tech #robotech #Robots #robotics #humanoidrobot #humanoid @sonu_monika @enilev @Jagersbergknut @TysonLester @chidambara09 @labordeolivier…Look at this robot make fried rice from scratch! #Cooking #chef #AI #ArtificialIntelligence #technology #tech #robotech #Robots #robotics #humanoidrobot #humanoid @sonu_monika @enilev @Jagersbergknut @TysonLester @chidambara09 @labordeolivier
-
SLMs offer efficiency; LLMs require massive resources. Keep improving.
By
–
clean point, slms got that efficiency while llms need a whole server farm to stretch, keep stacking those gains
-
China semiconductor dominance threatens global tech access
By
–
Future footage of all of us tweeting when China controls the world’s advanced semiconductors and decides who gets access to it. pic.twitter.com/nwIOctEvFy
— 🖤 Christine (@christinelu) 17 août 2025Future footage of all of us tweeting when China controls the world’s advanced semiconductors and decides who gets access to it.
-
Chinese Robot Decompresses After Hard Work Day
By
–
🇨🇳 A robot in China decompressing after a hard day at work!
— Amitav Bhattacharjee (@bamitav) 17 août 2025
pic.twitter.com/80RNIKL0p8#AI #ArtificialIntelligence #technology #tech #robotech #Robots #robotics #HumanoidRobots #humanoid @sonu_monika @enilev @Jagersbergknut @TysonLester @chidambara09 @labordeolivier @BetaMoroney…A robot in China decompressing after a hard day at work! #AI #ArtificialIntelligence #technology #tech #robotech #Robots #robotics #HumanoidRobots #humanoid @sonu_monika @enilev @Jagersbergknut @TysonLester @chidambara09 @labordeolivier @BetaMoroney
-
Beijing Robot Games Showcase Advanced Robotics and AI Technology
By
–
Beijing Robot Games!
— Amitav Bhattacharjee (@bamitav) 17 août 2025
pic.twitter.com/LiLgFYF0ed#AI #ArtificialIntelligence #technology #tech #robotech #Robots #robotics #HumanoidRobots #humanoid @anand_narang @antgrasso @chidambara09 @CurieuxExplorer @Eli_Krumova @Fabriziobustama @FrRonconi @ingliguori @ipfconline1…Beijing Robot Games! #AI #ArtificialIntelligence #technology #tech #robotech #Robots #robotics #HumanoidRobots #humanoid @anand_narang @antgrasso @chidambara09 @CurieuxExplorer @Eli_Krumova @Fabriziobustama @FrRonconi @ingliguori @ipfconline1
-
Multi-GPU Training Strategies for Deep Learning Models
By
–
By default, deep learning models only utilize a single GPU for training, even if multiple GPUs are available.
— Akshay 🚀 (@akshay_pachaar) 17 août 2025
An ideal way to train models is to distribute the training workload across multiple GPUs.
The graphic depicts four strategies for multi-GPU training👇 pic.twitter.com/rEKkFw3pF3By default, deep learning models only utilize a single GPU for training, even if multiple GPUs are available. An ideal way to train models is to distribute the training workload across multiple GPUs. The graphic depicts four strategies for multi-GPU training
-
Ring Approach for Scalable Model Weight Synchronization Across GPUs
By
–
And there you go! Model weights across GPUs have been synchronized. While the total elements transferred is still the same as we had in the “single-GPU-master” approach, this ring approach is much more scalable since it does not put the entire load on one GPU. Check this
-
Phase 2: Share-only segment transfer across GPUs
By
–
Phase #2) Share-only
— Akshay 🚀 (@akshay_pachaar) 17 août 2025
Now that each GPU has one entire segment, we can transfer these complete segments to all other GPUs.
The process is carried out similarly to what we discussed above, so we won’t go into full detail.
Iteration 1 is shown below👇 pic.twitter.com/7MssxcJfoEPhase #2) Share-only Now that each GPU has one entire segment, we can transfer these complete segments to all other GPUs. The process is carried out similarly to what we discussed above, so we won’t go into full detail. Iteration 1 is shown below
-
GPU Segment Transfer and Distribution Strategy
By
–
In the final iteration, the following segments are transferred to the next GPU.
— Akshay 🚀 (@akshay_pachaar) 17 août 2025
This leads to a state where every GPU has one entire segment, and we can transfer these complete segments to all other GPUs.
Check this 👇 pic.twitter.com/0EknVEN7IvIn the final iteration, the following segments are transferred to the next GPU. This leads to a state where every GPU has one entire segment, and we can transfer these complete segments to all other GPUs. Check this