Large Language Model News

A developer measured 120 AI-generated code samples from Qwen2.5-Coder 1.5B and found that every case flagged as a swallowed error by a naive detector was either a false positive or a documented fallback, concluding that …

DeepSeek engineer Liu Shengyu, who worked on the company's V4.1 models, publicly said he does not trust Anthropic or OpenAI to make cutting-edge AI "open and affordable," writing in a Monday social media post that he esp…

A developer released rag-debugger, an open-source MIT-licensed Python SDK that intercepts and inspects retrieval-augmented generation (RAG) pipelines to surface missing knowledge-base coverage. The tool, installable via …

A security practitioner has published a six-item taxonomy of AI threats, splitting them into four vectors — Shadow AI, Rogue Agent, Prompt Injection, and Model & Data Supply Chain — and two impacts, Data Exfiltration and…

AI industry leaders Dario Amodei, Sam Altman, Elon Musk, and Demis Hassabis have converged on the position that the latest generation of large language models are not safe, according to MIT Technology Review, which notes…

Salesforce unveiled Koa, its first reasoning model, built on Nvidia's Nemotron-3-Super-120B foundation model and trained with Group Relative Policy Optimization (GRPO) for multi-turn reasoning and tool-use tasks inside S…

The performance gap between frontier AI models from US tech companies and the best open-weights models from Chinese companies has narrowed to just 4.4 months, according to a State of Open Source AI report from Mozilla pu…

Salesforce Inc. announced at its Dreamforce user conference that it partnered with Nvidia Corp. to train and release Koa, a specialized AI model built on Nvidia's Nemotron architecture to reason about CRM data within Sal…

Microsoft offers five free GitHub curricula for learning data science and artificial intelligence, ranging from a 10-week, 20-lesson Data Science for Beginners course to a 21-lesson Generative AI for Beginners course. Th…

Salesforce unveiled Koa, its first reasoning model, at its Dreamforce conference, built on Nvidia's open-weight Nemotron model and post-trained by the two companies for sales, marketing, and customer-support tasks. Koa w…

A tutorial demonstrates treating prompt templates as tunable hyperparameters by wrapping a language model in a scikit-learn-compatible classifier and running GridSearchCV over candidate prompts for zero-shot text classif…

TypeSafe announced Jev, a model that takes unstructured state as input and returns typed, probabilistic decisions — Choice, Score, and Noul — rather than free text, claiming it is 20-200x faster and 40-400x cheaper than …

A 2026 guide from Apify lists 12 LangChain alternatives for building AI agents, including Haystack by Deepset, Hugging Face's roughly 1,000-line smolagents library, and the open-source TypeScript framework Mastra. The gu…

A developer's cost analysis of AI agent runs shows that a 20-turn agent reading roughly 59,000 tokens of material is billed for 656,000 input tokens because stateless model APIs resend the full conversation history on ev…

MediaTek launched the Dimensity 9600 Pro globally on Tuesday, a 2nm flagship mobile platform with a 2+3+3 CPU architecture peaking at 4.55GHz, LPDDR6 RAM and UFS 5 storage support, and a dual NPU design. MediaTek claims …

Anthropic CEO Dario Amodei and Microsoft published separate essays calling for a slowdown — not a pause — in frontier AI development, with Amodei urging external evaluators, government-supervised coordination among front…

Orthrus, a hybrid autoregressive-diffusion architecture for accelerating autoregressive language-model inference, claims its intra-model consensus mechanism enables lossless speculative decoding, according to the source.…

GreyNoise reported that a threat actor used hundreds of AI agents built on OpenAI's Codex and DeepSeek models to develop and launch exploits against two known PaperCut NG/MF vulnerabilities, compromising at least 440 ins…

A developer has released Franca, an open-source Java 21 library published to Maven Central that implements the server side of agent wire protocols, terminating OpenAI Responses and Anthropic Messages dialects inside a de…

A developer explains embeddings, the numerical vector representations that let AI systems compare meaning rather than exact text matches. The piece describes how an embedding model converts text into high-dimensional vec…