Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
OpenAI's GPT-6 Astra topped the ErdosBench for open math problems, despite chief scientist Jakub Pachocki saying math was deliberately not a priority for the model. OpenAI is instead directing resources toward recursive …
A developer argues that Retrieval-Augmented Generation (RAG) is a retrieval strategy rather than a complete architecture, and that production GenAI systems should choose retrieval approaches based on the problem at hand.…
A prefix-cache simulator replaying 68,266 requests from 393 real Claude Code sessions and 23,608 Mooncake requests failed to beat the production LRU baseline in three separate attempts, according to the study's author. T…
OpenAI launched GPT-6 Astra on September 10, 2026, a model it calls its most capable for business, rolling out across ChatGPT Work, Codex, and the API at $10 per million input tokens and $50 per million output tokens. Op…
BuildBetter, a customer-conversation analysis company founded by Spencer, has spent roughly $7M over six years building a system to read every customer conversation across an enterprise, according to a blueprint the foun…
Meta is preparing on-demand custom voices, a Meta AI-style voice mode, reasoning controls, and new Avocado model updates for its Muse product, according to reporting by Alexey Shabanov dated 10 Sep 2026. Meta is also tes…
A developer released AI-Git-Bot, an MIT-licensed, self-hosted workflow automation bot that reviews pull requests, generates unit and Playwright tests, triages issues, and syncs documentation using a locally run 7B open-w…
A developer launched geoSurge, a free tool that tracks how brands appear in LLM search results by connecting users' existing LLM subscriptions to gather sentiment data and run analysis. The developer said the tool was bu…
UnifoLM-ER-1, a 4B-parameter embodied reasoning model built on Qwen3-VL-4B, leads open-source models on seven of 16 multimodal perception and understanding benchmarks and matches leading proprietary models overall, accor…
AI-generated code fails primarily because large language models predict the most likely next token rather than simulating a compiler, producing hallucinated library methods, outdated API versions, and logic that misses e…
Qualcomm disclosed details of its next-generation premium Snapdragon platform's Oryon CPU, Adreno GPU, and Hexagon NPU ahead of its Snapdragon Summit later this month. The Oryon CPU's two Prime cores are rated at 5 GHz, …
Runway Financial Inc. has pivoted away from accounting software and rebranded as cfo.ai, launching an AI agent called Ari that acts as a chief financial officer for startup founders and independent business owners. The c…
The "Thursdays with Koog" series concluded after 11 weeks with a post examining Koog's near-complete focus on hosted models and its limits for local, on-device inference. The post notes Koog's only on-device path is an u…
A paper by Federico Barbero and co-authors, revised 13 May 2025 as v3 on arXiv, argues that Rotary Positional Encodings (RoPE) in Transformer-based large language models are not primarily useful for decaying token depend…
A practitioner guide recommends securing production LLM integrations with an intermediary validation gateway that strips PII via regex or a local BERT-based classifier, enforces JSON output, and caps requests at a 30-sec…
A developer argues that the reported large-scale harvesting of chain-of-thought reasoning traces from frontier models like Claude, GPT, Gemini, and Grok is not a novel attack category but a repurposed version of long-kno…
Deepseek released V4.1-Flash, a multimodal model with 552 billion parameters that cuts KV cache memory to a quarter of its predecessor's requirement, with only 16 billion parameters active per token. On the DeepSWE codin…
Meta's top-down plan to replace staff with frontier AI models imploded, according to a Reuters investigation, with major security and technical incidents rising 40% and time spent on those incidents up 70% amid a 220% in…
Nvidia and Palantir are partnering to run supply chains with AI, starting with Nvidia's own operation that spans millions of parts and thousands of suppliers, where a single Vera Rubin rack contains 1.3 million component…
Traditional machine learning models are limited to static input-output mappings and cannot handle multi-step, dynamic, or action-requiring problems, according to an article explaining how combining trained models with ag…