Condition-anchored generative distillation (CAGD) reduced four-task final held-out loss in a 219M masked diffusion language model from 2.927 to 1.114 in one task order and from 2.168 to 0.891 in exact…
A developer rebuilt a five-agent research pipeline after discovering that accuracy degraded as accumulated context grew, a phenomenon Chroma's 2025 research terms "context rot" — testing 18 frontier m…
A new paper, REAL-Q, proposes a dynamic gradient descent approach to post-training quantization that the authors say fixes structural flaws in GPTQ, including upstream and downstream misalignments and…
A developer published a 10-minute quickstart for running GraphRAG entirely locally using Ollama and the open-source Chaos Cypher stack, requiring no cloud API keys. The walkthrough covers pulling the …
Independent tokenizer fingerprinting of the API model "Jev" found its token counts match an o200k_base tokenizer with a Qwen3 pretokenizer on roughly 77% of probes, with every difference consistent wi…
A developer ran an exploratory study testing whether modern LLMs contain internal self-models of other LLMs, using GPT2-medium and Qwen3-4B on 12 post-cutoff news headlines. By having Qwen continue GP…
Neurono ML released typed-lm, an Apache-2.0 Rust inference engine that turns dense decoder models including Llama, Qwen2, Qwen3, Mistral, Gemma, Gemma2 and Gemma3 into a typed semantic-routing API ret…
A developer built RelayLab, a local multi-agent system that pairs a Builder model (Qwen2.5-Coder) with a Reviewer model (Qwen3) via Ollama and a Python orchestrator to autonomously develop an app over…
Intel released OpenVINO 2026.4, adding EAGLE-3 tree drafting that lifts speculative decoding throughput on Qwen3-8B from 1.65x to over 1.85x at batch size 1, Multi-Token Prediction for Gemma 4, Qwen3.…
Lateos AI released Reflex, an open-source GGUF-native Rust and CUDA inference engine that compiles every CUDA kernel ahead-of-time with nvcc at build time to eliminate the multi-second JIT tax on firs…
A new technical blog post details a three-way logprob comparison method for debugging probability mismatches between the rollout engine and trainer in online reinforcement learning for large language …
A developer released Reactive Agents, an open source MIT-licensed TypeScript framework now at v0.16, designed to make agent harness behavior explicit rather than hidden inside a black box. Each .with(…
A developer documented four context window management patterns for long-running AI agents after observing that a four-hour agent session spent roughly three quarters of its token budget on its own tra…
Researchers introduced EnvCraft, an automated framework that synthesizes executable environments and training data for Agentic Reinforcement Learning, detailed in arXiv paper 2609.05576v1. EnvCraft's …
Researchers proposed cross-model KV cache transfer, a method that reuses a source model's key-value cache to skip prefill when swapping between different-sized models in a family, and found a closed-f…
An engineer who fine-tuned Llama 3.3 70B on Amazon Bedrock found that the pricing page lists training costs only for Llama 2, not for the newer model they actually used. Probing all 33 Bedrock regions…
A new arXiv paper (2609.04565v1) reports that large language models can be effectively trained for reasoning with as few as one or two tokens per trajectory, just 0.05% of all generated tokens, challe…
Seattle-based AI research startup Resect AI announced Thursday it has raised $25 million in early funding to build an accountability layer for enterprise AI that captures and reduces hallucinations at…
Perplexity launched Hybrid Compute for Mac on September 1, an AI system that automatically routes sensitive data to on-device processing via its open-sourced PII-Tracer classifier, achieving 0.965 rec…
Researchers introduced SCX Router, a lightweight GLiClass-based model for zero-shot selection among LLM inference endpoints, achieving an aggregate top-1 score of 0.707 on a 1,000-task subset versus 0…