Discover the Best AI Resources

Continual Learning Mechanisms Compose for Long-Horizon Memorization

Investigates long-horizon memorization in language models by evaluating how complementary continual-learning mechanisms (data, function, weight anchors and low-rank allocation/LoRA) compose to reduce catastrophic forgetting across 100 sequential tasks; best composition raises final retention from 1.2% to 34.9%.

SAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?

Evaluates whether AI agents can autonomously use Sparse Autoencoders (SAEs) to discover interpretable features and causally steer model behavior. Introduces a 20-task benchmark that has agents design contrastive probes to search a 131K+ feature dictionary in Gemma-2-9B-IT and evaluates activation rank, selectivity, and causal steering against expert references.

Nex-N2.5-mini

A multimodal, agentic LLM optimized for long‑horizon, visually grounded workflows — capable of operating browsers and terminals and autonomously executing and testing code. Open‑source weights are available and the family ships in mini, Pro and Max variants for different compute/quality tradeoffs.

Eliciting Weak-to-Strong Generalization with On-Policy Reverse Distillation

Evaluates a weak teacher's RL-induced policy shift on the student's own rollouts and amplifies verifier-supported updates so stronger models can learn from weaker supervisors and surpass them. It rescales only verifier-supported policy-gradient components to preserve optimization fixed points while accelerating learning, reducing student updates versus standard RL or distillation.

LFM3-Audio-2B

Handles text and speech input, producing text, speech, or both for streaming voice interaction, multilingual audio tasks, and spoken tool calls. Its 2B-parameter size targets local and edge use, not complex multi-step reasoning.

Omni Interaction Agent Technical Report

Unifies continuous streaming perception, full‑duplex real‑time interaction, and agentic tool use across video, audio, and text. Proposes a Cerebellum–Brain split—streaming Thinker‑Talker cerebellum for low‑latency interaction plus a Brain for complex reasoning—and is released with models, code, and data.

NeoHorse-1: Towards Recursive Self-Improvement via Agentic Post-Training with Routing Harness

Explores a practical mechanism for recursive self-improvement by post-training LLMs: uses a routing harness to record agent executions and convert traces into curriculum-guided supervised fine-tuning and on-policy distillation data, closing an evaluation-selection-update loop and improving benchmark performance.

Edge0-35b-a3b Preview

Runs a 35B-class sparse MoE LLM with SSD-streamed experts, 4-bit quantization, prerouter routing prediction and Recover-LoRA adapters to enable ~2.9–3 GiB active memory and interactive decoding (~15 tok/s) for on-device inference.

Swift-Qwen3.8-27B

A reasoning‑efficient fine-tune of Qwen3.8-27B that penalizes overthinking tokens to shorten internal reasoning traces — about 58.3% fewer thinking tokens with <1% accuracy loss and ~1.95× speedup; designed for long-context, multimodal and quantized deployments.

UltraData-Stack-v3-L2

Selected ~245B-token corpus of algorithmically relevant code across 11 languages from The Stack v3, produced by file-level language-adaptive selection (UltraData-Code-L2-Classifier). Concentrates problem-solving implementations for LLM code pretraining while preserving broad language coverage; released under ODC-By.

Nex-N2.5-Pro

Multimodal agentic model for long-horizon computer and browser tasks, with visual self-correction and function-calling. The Pro variant is a 397B Mixture-of-Experts (≈17B active) model with a 262,144-token context window, Docker deployment recipes, and weights currently marked “coming soon.”

NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction

Adds a Next Concept Prediction objective that predicts discrete, product-quantized latent concepts spanning multiple tokens alongside next-token prediction, yielding faster pretraining convergence and measurable downstream gains with an 8.9B-parameter latent-space LLM.