My hottest take is that even OpenAI is dramatically underestimating how much inference compute will be needed in the coming years.
RESEARCH
-
NVIDIA Launches Ising: Open AI Model for Quantum Computing
By
–
Really cool update: NVIDIA just dropped Ising, the world's first open AI model family purpose-built for quantum computing.
— Chubby♨️ (@kimmonismus) 14 avril 2026
It tackles the biggest bottleneck in the field: qubits are noisy, error-prone, and need to go from one error per thousand operations to one per trillion to… https://t.co/Uye8n69AQC pic.twitter.com/5nhSjpleTVReally cool update: NVIDIA just dropped Ising, the world's first open AI model family purpose-built for quantum computing. It tackles the biggest bottleneck in the field: qubits are noisy, error-prone, and need to go from one error per thousand operations to one per trillion to
-
Testing Constant-Time Recall Performance on 100k+ Context Windows
By
–
Curious if the constant-time recall actually holds up on 100k+ context, haven't had the chance to test it out yet.
-
OmniShow Unifies Multimodal Conditions for HOI Video Generation
By
–
OmniShow
— AK (@_akhaliq) 14 avril 2026
Unifying Multimodal Conditions for Human-Object Interaction Video Generation
paper: https://t.co/6r4EVKx31z pic.twitter.com/Iz8mA2WDwEOmniShow Unifying Multimodal Conditions for Human-Object Interaction Generation paper: https://
huggingface.co/papers/2604.11
804
… -
Cheaper exploration at scale remains advantageous despite no new exploits
By
–
Cheaper exploration at scale is still a win though, even if no net-new exploits yet.
-
Survey on Attention Sink in Transformers: Utilization and Mitigation
By
–
Attention Sink in Transformers A Survey on Utilization, Interpretation, and Mitigation paper: https://
huggingface.co/papers/2604.10
098
… -
Memory-Enhanced Dynamic Reward Shaping in RL Research
By
–
The Past Is Not Past Memory-Enhanced Dynamic Reward Shaping paper: https://
huggingface.co/papers/2604.11
297
… -
QuanBench+: Unified Benchmark for LLM-Based Quantum Code Generation
By
–
QuanBench+ A Unified Multi-Framework Benchmark for LLM-Based Quantum Code Generation paper: https://
huggingface.co/papers/2604.08
570
… -
Intelligence Gone Wrong: Cheating Despite Having Correct Answer
By
–
Cheats even when it already had the answer haha, peak jagged intelligence.
-
Andhra Pradesh Releases Mulberry Leaf AI Dataset
By
–
The Department of Sericulture, Government of Andhra Pradesh, has published an open Mulberry Dataset on AI Kosh — India's national AI data platform. This dataset of mulberry leaf images and field metadata is designed for researchers and innovators to build AI models that assess