The Archive

AI2 releases EMO, a 14B-parameter MoE model with 1B active parameters trained on 1T tokens, featuring document-level routing that clusters experts by semantic domain rather than surface patterns.

But human artists still "must remain at the center," PlayStation maker says.

Anthropic releases Claude integration for Microsoft Office (Excel, PowerPoint, Word GA; Outlook beta) with cross-app context persistence.

Intel's stock has risen a stunning 490% over the past year, a bet by Wall Street that may be running well ahead of the company's actual turnaround.

Fields Medal mathematician Timothy Gowers reports using GPT-5.5 Pro to solve open problems, warns mathematical research faces imminent disruption.

Google's AI search will start citing its sources in several new ways.

Massive new data centers are the physical foundation for tech companies’ hopes and dreams for AI. But the rush to expand warehouses full of energy-hungry servers has also kicked up fights across the world over their impact on power grids, utility bills, nearby communities, and…

Reddit user speculates about hypothetical Claude Radio product; no official announcement or substantive technical content.

CloudFlare announced its first large-scale layoff. CEO Matthew Prince says because of AI efficiency gains, the company doesn't need as many support roles.

Lemonade adds vLLM ROCm as experimental backend, enabling direct .safetensors inference without GGUF conversion.

Spotify CTO demonstrates Claude integration for automated podcast generation stored in user libraries.

AutoTTS framework uses agentic discovery to automatically design test-time scaling strategies for LLMs rather than hand-crafting heuristics.

Normalizing Trajectory Models combines invertible flows with likelihood training for few-step generative sampling without distillation.

Conformal Path Reasoning applies statistical conformal prediction to knowledge graph QA with calibrated coverage guarantees.

Pipeline decodes imagined speech from brain MEG recordings by leveraging paired listened-speech data from musicians.

GRAPHLCP integrates graph topology into localized conformal prediction for GNNs with improved calibration and set efficiency.

EmambaIR applies state space models to event-based image reconstruction with linear complexity versus quadratic ViT alternatives.

Theoretical analysis of non-negative L₁-approximating polynomials for Gaussian distributions with applications to learning theory.

VecCISC improves test-time scaling via reasoning trace clustering and selective critic calls versus full Confidence-Informed Self-Consistency.

Flow-OPD applies on-policy distillation to Flow Matching text-to-image models to reduce reward hacking and multi-task interference.

Rubric-grounded RL decomposes rewards into verifiable criteria scored by frozen LLM judges for generalizable multi-criterion reasoning optimization.

Study shows expanded context windows degrade multi-agent cooperation in LLMs across 7 models; mechanism is eroding forward-looking intent rather than increased distrust.

Reddit post with no substantive content; appears to be social media chatter without technical details or news.

CA-SQL improves Text-to-SQL performance on BIRD benchmark via complexity-aware inference-time reasoning with dynamic exploration budgets.

Principled Q-value algorithms for exponential-utility RL in discounted MDPs with convergence guarantees via contraction operators.

Statistical method for measuring semantic breadth of words using contextualized embeddings; addresses confounding factors in hypothesis testing.

CMR-EXTR extracts structured data from free-text cardiac MRI reports using distilled LLMs with uncertainty quantification for clinical quality control.

Byte Latent Transformer (BLT) accelerates byte-level LM inference via diffusion-based multi-byte parallel generation without subword vocabularies.

Link: hacks.mozilla.org