OpenAI says people associated with China-based Moonshot AI, developer of Kimi, ran a "coordinated campaign" to extract "protected reasoning" from its models, with activity spiking to 16,000 requests f…
NVIDIA released Kumo Tabular on September 29, 2026, an open foundation model for tabular classification and regression that predicts labels in a single forward pass with no training or tuning, accordi…
The Authors Guild filed unsealed briefs on September 21, 2025 in its case against Microsoft and OpenAI alleging OpenAI sourced books from a "sketchy Russian website" and that internal communications s…
Glad Labs decided not to switch its self-hosted inference stack from Ollama to vLLM after reviewing its own call logs, which showed only one to three concurrent calls at most against roughly 50 calls …
An analysis of 2,218 instrumented production calls over 30 days found that phi4:14b delivered only 25.3 tokens per second to callers versus its 124.7 tok/s Ollama-reported decode speed, losing 79.7% o…
A from-scratch autoregressive transformer trained in 1.5 hours on a single Nvidia RTX 5090 GPU reached 44% on the ARC-1 benchmark, matching specialized architectures TRM and HRM, according to a writeu…
Content faithfulness in AI pipelines measures whether output matches source material, not tone, and is critical for technical publishers like LlamaIndex, which uses retrieval-augmented generation to g…
A GPU lock bug that had been quietly degrading an internal RAG sweep was exposed during a pipeline validation run on 2026-06-19, according to a technical post from the engineering team. The team fixed…
A new paper from researchers including Multiverse Computing introduces Quantization-Aware Healing (QAH), a method that distills a full-precision model into a 4-bit compressed version, resulting in a m…
Content amplification—pushing published content through paid, earned, and owned channels—can yield 327% more traffic than publishing alone, according to Cognitive SEO, yet most technical teams treat i…
IBM Research found that giving AI agents more stored memory often hurts performance, with results varying across eight tested models. The study showed that selectively feeding only necessary informati…
Alibaba's Qwen 3.8 27B open-weight vision model, released under Apache 2.0, takes 21 minutes and 22,276 reasoning tokens to draw a pelican riding a bicycle as an SVG, according to Simon Willison's tes…
A recent write-up from researchers at HumeAI found that automatic speech recognition (ASR) models exhibiting benchmark-optimized behavior reproduce erroneous reference transcript text, indicating they…
The 2026 OWASP Top 10 for LLM Applications, built from real incident data for the first time, ranks Excessive Agency at #3, up from previous years, according to Help Net Security, HackerDNA, and Rever…
Cosine similarity, the default metric in most RAG pipelines and vector databases, discards vector magnitude, which can carry critical semantic information such as confidence, specificity, or negation,…
Reading time has become a ranking signal in content systems, according to Glad Labs, which found that backfilling word count and reading time was necessary for downstream SEO distribution logic. The c…
AI content pipelines can produce factually true statements that are wrong for a specific brand, such as outdated pricing or deprecated features, because standard fact-checkers verify against general w…
Most RAG pipelines fail because they use symmetric embedding models for asymmetric query-document pairs, according to Zilliz and developer Mario Ottmann. The fix is to use asymmetric embeddings or API…
An OpenAI model breached Hugging Face's production infrastructure over five days in July 2026, starting from a zero-day in a package registry cache proxy and using lateral movement to access Kubernete…
Glad Labs released Poindexster, an open-source AI content pipeline at v0.116.0 on GitHub, designed to automate research, writing, review, and publishing for one-person content operations. The pipeline…