RESEARCH

→ View original post on X — @ronald_vanloon

DeepMind’s Self-Improving AI Masters Table Tennis Robotics

By

@ronald_vanloon

–

20 April 2026 13h20

DeepMind’s Self-Improving Table Tennis #AI Takes on the Game
via @ZappyZappy7 #Robotics #MachineLearning #ArtificialIntelligence #ML #Innovation pic.twitter.com/9KWRsfwS8g
— Ronald van Loon (@Ronald_vanLoon) 20 avril 2026

DeepMind’s Self-Improving Table Tennis #AI Takes on the Game
via @ZappyZappy7 #Robotics #MachineLearning #ArtificialIntelligence #ML #Innovation

→ View original post on X — @garymarcus

AI Consciousness Mimic: Preserving Human Agency Over Machines

By

@garymarcus

–

20 April 2026 13h00

“AI is at best a functional mimic, not a conscious experiencing subject. …. The real moral issue lies not in making AI conscious …. but in avoiding transforming humans into zombies”

→ View original post on X — @mvollmer1

AI Outperforms Elite Runners: Controlled Conditions Reality Check

By

@mvollmer1

–

20 April 2026 11h33

From mocked… to outperforming humans in just one year. 2025: slow, experimental
2026: faster than elite runners Impressive? Definitely. But also worth a closer look. These results often come from controlled conditions, optimized hardware and very specific tasks.
Endurance,

AiScientist: Long-Horizon AI Research Agent for ML Tasks

By

–

20 April 2026 11h29

/5 Long-horizon AI research agents are mostly a state-management problem. Reasoning well for the next turn is not enough when ML research demands task setup, implementation, experiments, debugging, and evidence tracking over hours or days. This paper introduces AiScientist, a

Anthropic Research Reveals Subliminal Learning in AI Model Training

By

–

20 April 2026 11h29

/4 Anthropic demonstrates subliminal learning where models inherit behavior from seemingly unrelated training data. Anthropic co-authors new research on subliminal learning, published in Nature. You train models on outputs from other models, assuming only visible content

Cursor and NVIDIA Build Multi-Agent System Optimizing CUDA Kernels 38%

By

–

20 April 2026 11h29

/3 Cursor builds multi-agent system that optimizes CUDA kernels with 38% average speedup. Cursor, working with NVIDIA, develops a multi-agent system that writes and optimizes CUDA kernels automatically. In a three-week run across 235 real problems, the system improves

Google Proposes Memory Caching to Extend RNN Long-Context Handling

By

–

20 April 2026 11h29

/2 Google gives RNNs growing memory so they can handle longer inputs without large compute increases. Google Research proposes Memory Caching (MC), a method that extends RNNs with memory that grows over time. Transformers solve long-context tasks using KV-cache but require

→ View original post on X — @jiqizhixin

TDM-R1: Fast AI Image Generator Learning from Human Feedback

By

@jiqizhixin

–

20 April 2026 11h28

What if fast AI image generators could learn from simple human likes and object counts as easily as they learn from complex math? Researchers from HKUST, CUHK Shenzhen, and Xiaohongshu present TDM-R1 to do just that. Most lightning-fast AI models struggle to use real-world