NewsroomLive

Anatomical slider LoRA extends Illustrious XL and Pony model controlReleases · 55m ago Gemma 4 26B uncensored weights quantized to GGUF for local inferenceReleases · 2h ago Microsoft's MAI-Thinking-1: 35B-active MoE model trained on 8,000 GB200sIndustry · 3h ago Google DeepMind open-sources Science Skills for AI research workflowsResearch · 4h ago Anthropic adds Services Track and Partner Hub to Claude deployment networkPlatform · 5h ago Ideogram 4 open-weights: 9.3B DiT with Qwen vision encoder, native 2KReleases · 6h ago

Image models Video models LoRA models LLM models

Prompts Gallery Journal News Polls Free Tools Agents Can I run it?

Loading…

contact@uncensoredhub.ai

Image Video LoRA LLM Prompts Gallery Journal News Polls Tools Agents

About Terms Privacy Cookies DMCA

18+ · Adults only · Not affiliated with hosted platforms

← Back to hub

News · 60 articles

What's happening

Releases, platform updates, funding, industry shifts, community work, and research from across the AI scene.

All Releases Platform Funding Industry Community Research NSFW

ResearchNSFW

Google DeepMind open-sources Science Skills for AI research workflows

DeepMind released Science Skills, a Python library of modular functions that let language models call scientific tools, databases, and simulations as part of reproducible research workflows.

Jun 6, 2026

Research

q0 framework trades compute for data via parallel ensembles and cyclic schedules

New pretraining framework runs parallel model ensembles with anti-phase learning-rate and weight-decay cycles, then distills and weights predictions to squeeze 12.9× more value from fixed datasets when compute is abundant but training tokens are scarce.

Jun 6, 2026

Research

LeJEPA latents recover world state up to rotation, LeCun team proves

New proof from LeCun and Balestriero establishes that LeJEPA's Gaussian-regularized embeddings linearly reconstruct ground-truth latent variables, enabling direct planning in learned space.

Jun 6, 2026

Research

Stereological theory reveals LLM benchmark blind spots 100× larger than score gaps

A new arXiv preprint proves that structural uncertainty in LLM benchmarks dwarfs statistical noise, with simulated ranking swaps affecting 92% of trials across three major leaderboards.

Jun 6, 2026

Research

MAOAM unifies object and material selection via text or click

A new vision-language model framework enables precise image segmentation by object or material using either text descriptions or click interactions, trained on synthetic datasets with VLM-generated annotations.

Jun 6, 2026

ReleasesResearch

Music recommender system lifts recall 95% by fusing lyrics, audio, and LLM metadata

A new framework enriches LastFM-1K with audio embeddings, lyric vectors, and LLM-generated semantic tags, lifting session-based music recommendation recall by up to 95 percent over ID-only baselines.

Jun 6, 2026

Research

Lattice Deduction Transformer solves extreme logic puzzles with 800k parameters

A specialized neural architecture combining lattice structures with recurrent transformers achieves perfect accuracy on complex reasoning tasks that stump frontier LLMs, using a fraction of the training cost.

Jun 3, 2026

Research

Neural network curvature exponent decomposed: why convolutions differ from transformers

New preprint proves curvature exponent α—governing how Hessian eigenvalues scale with gradient singular values—decomposes into a geometric alignment term, explaining why convolutions show α≈2 while transformer attention and MLP layers differ.

Jun 3, 2026

ReleasesResearch

VaSE cuts reasoning model KV cache memory 4× while boosting accuracy

New training-free method protects high-magnitude value states and adds stochasticity to KV cache eviction, outperforming prior techniques by 4% on reasoning tasks while enabling static memory footprints.

Jun 3, 2026

ResearchNSFWPlatform

VL-DAC trains vision-language models in simulators, boosts Qwen2-VL-7B by 50%

T-Bank AI Lab's VL-DAC method trains vision-language models in simulated environments before deploying them on real tasks, boosting Qwen2-VL-7B performance by over 50% on interactive benchmarks.

Jun 2, 2026

ReleasesResearch

OpenWebRL-4B reaches 67% success on live-web tasks with minimal supervised data

New open framework trains visual web agents via online reinforcement learning on real websites, achieving state-of-the-art open results with just 400 initialization examples.

Jun 2, 2026

Research

Diffusion models gain internal reasoning loops to fix compositional errors mid-generation

New arXiv preprint wires recurrent MoE modules into multimodal diffusion attention layers, letting models iteratively refine visual tokens across internal latent steps for better compositional accuracy.

Jun 1, 2026

Research

QASM-Eval: 4,100-task benchmark for hardware-level quantum code generation

New dataset trains LLMs on OpenQASM-3's hardware-facing features — mid-circuit measurement, pulse control, timing — revealing current models struggle without fine-tuning.

Jun 1, 2026

ReleasesResearchNSFW

Alibaba's RoTS-32B tops OSWorld with 800k error-recovery trajectories

Alibaba researchers released GUI-RobustEval, a 1,216-case benchmark for GUI agent error recovery, plus RoTS-7B and RoTS-32B models trained on 800k synthesized recovery steps that now lead OSWorld.

Jun 1, 2026

ResearchNSFWPlatform

Confident agents hijack multi-LLM debates, arXiv study shows

Researchers model multi-agent LLM systems as mixtures of experts governed by social-influence dynamics, where confident agents dominate group decisions regardless of correctness.

May 31, 2026

Research

Language Game framework lets LLMs negotiate with gene networks in virtual worlds

Researchers built Language Game, a system that uses adapters and LLMs to translate human text into physical states that prompt gene regulatory networks to respond, then translates their behavior back into natural language.

May 30, 2026

ReleasesResearch

Reflective Prompt Tuning boosts LLM reasoning 12.9 points without retraining

Researchers propose RPT, a framework that uses LLM function calls to diagnose failure patterns and revise prompts iteratively, improving reasoning task performance by up to 12.9 points without parameter updates.

May 30, 2026

ResearchNSFWPlatform

SkillOpt boosts GPT-4.5 reasoning by 23.5 points through text-space instruction optimization

Microsoft researchers released SkillOpt, a text-space optimizer that treats agent skills as trainable external state and compiles domain adaptations into human-readable markdown instruction files.

May 29, 2026

1 2 3 4 Next →