Techyon

Claude discovers a novel enzyme system with CRISPR-like repeats

Anthropic’s life sciences team prompted Claude to scan a large DNA‑sequence database for uncharacterized reverse transcriptases (RTs). Over 21 hours, roughly 950 Claude agents processed 210 million tokens, extracting &gt…

→ View original source

Mercury 2.5 LLM hits 770 tokens per second

Mercury 2.5, Inception’s proprietary text-only reasoning model, was evaluated on Artificial Analysis’ Intelligence Index, a composite benchmark covering reasoning, knowledge, mathematics, and coding. The evaluation gener…

→ View original source

Best AI model/stack for one RTX 6000 blackwell

A Reddit user asks the LocalLLM community what AI models and software stacks they are running on a single RTX 6000 Blackwell GPU, seeking recommendations for optimal performance. The post invites discussion on model quan…

→ View original source

Once Claude can measure something, it can make it faster

In a two-week performance sprint, Anthropic engineers used an internal research model, Claude Tag (comparable to Opus 5.5), to accelerate the claude.ai web and desktop applications by approximately 3x. By integrating Cla…

→ View original source

GPT-6 Sol and Luna

OpenAI has introduced the GPT-6 Sol and Luna models, expanding its series of generative pre-trained transformers. Further details about the models' capabilities and release are available at the original announcement.

→ View original source

Uranus: Building the Next-Generation Simulation Infrastructure for Embodied AI

Uranus introduces a data‑driven robot simulator that leverages a joint‑trajectory‑conditioned autoregressive diffusion model to generate visual observations on‑the‑fly. At each simulation step the model consumes a future…

→ View original source

All modalities are equal, but video is more equal: Closing the Cross-Attention Gap in Joint Video Generation

The paper investigates an asymmetry in multimodal diffusion transformers where companion modalities such as 3D body motion or audio form strong attentional links to video tokens, but the reverse links that allow those mo…

→ View original source

InternW0: A Foundational Physical World Model for Efficient Real-World Interactions

Researchers from Shanghai AI Laboratory have introduced InternW0, a foundational physical world model designed for efficient, actionable real-world interactions. The architecture utilizes an asymmetric video-action frame…

→ View original source

Six Layers Less: Encoder Pruning for Whisper with Label-Free Recovery

Researchers have introduced a novel method for optimizing OpenAI's Whisper Automatic Speech Recognition (ASR) models by specifically targeting the encoder architecture. While previous optimizations like whisper-large-v3-…

→ View original source

Calibration as a First-Class Criterion in LLM Evaluation

The paper argues that calibration— the agreement between a language model’s expressed confidence and its empirical correctness—should be treated as a core evaluation criterion alongside traditional performance metrics. I…

→ View original source

GeoPair: Geometry-Preserving Cross-Layer Factorization for Training-Free Transformer Compression

GeoPair introduces a training‑free compression framework that targets cross‑layer redundancies in transformer models by jointly optimizing weight pairings and learning a shared dictionary factorization. Rather than forci…

→ View original source

Jev isn't new tech. Its marketing targets people who think AI started with LLMs.

Jev is being marketed as a novel decision model, but it essentially replicates existing zero-shot classifier techniques that output constrained class probabilities without autoregressive generation. Its capabilities—such…

→ View original source