Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
A technical comparison examines the tradeoffs between Ollama and llama.cpp for local LLM inference, framing the choice as one between a managed model service and a toolkit operated directly. The guide argues Ollama suits…
A developer built an open-source autonomous Discord AI companion in Python that uses probabilistic skipping via discord.ext.tasks.loop to let an LLM spontaneously post messages without spamming servers. The bot employs a…
A blogger using two separate Claude instances for drafting and review found that withholding the author's reasoning from the reviewer surfaced two issues the author had missed, including a likely backwards claim about to…
A developer known as zigster64 published atari.zig, a repository that enables an LLVM m68k backend to produce a bare-bones Zig toolchain on macOS, and used DeepSeek tokens to have an LLM generate a Space Invaders clone a…
A developer released ai-sdk-threads, an npm package that persists AI SDK chat threads and messages directly into an existing Postgres or SQLite database via Drizzle, replacing the copy-and-maintain persistence pattern us…
An agent harness is the runtime layer around a model — comprising a tool registry, model, context management, guardrails, an agent loop and a verify step — that makes an agent reliable, according to a technical explainer…
A developer's experiment across eighteen coding-agent sessions found that compaction can cause models to answer from a stale summary rather than re-reading source files, with seven of sixteen compacted sessions reporting…
Rage4J, a Java library for evaluating LLM applications, is now available at version 1.0.4 through Maven, according to its project page. The tool measures accuracy, relevance, and faithfulness of AI responses and offers a…
Developer lencx, creator of the open-source ChatGPT desktop client, has released Minke, a local-first desktop agent workspace powered by DeepSeek Harness under the Apache-2.0 license. The Electron app unifies conversatio…
Legora, the Swedish legal AI startup, is building its own corpus of citable case law and legal data to compete with LexisNexis and Westlaw, hiring former attorneys and striking licensing deals with "tens" of publishers i…
Database administrators should treat AI as a process-following diagnostic assistant rather than a replacement, according to an InfoWorld opinion piece marking PostgreSQL's 30th anniversary. The author argues that LLMs co…
China's state security minister Chen Yixin warned in a government outlet that advanced U.S. models such as Anthropic's Mythos and OpenAI's GPT-5.5-Cyber could pose serious risks to China's critical information infrastruc…
A writer demonstrated that Google's Gemini free tier can generate abstract artwork from scratch through a chat window using iterative prompting, without a paid model. The writer, a Muslim programmer, built the compositio…
Researchers formulated adaptive routing of prompts to large language model experts as a bandit problem with K actions representing experts and d features encoding prompts over a horizon of T rounds, aiming to maximize re…
The Shanghai Artificial Intelligence Laboratory released Atria Dawn Preview, a preview version of a new-generation agentic model built on a 744B-parameter MoE foundation model. The model is designed for research and engi…
A developer has released OpenArch, a repository of hand-written PyTorch implementations of modern open-source LLM architectures, cataloged against Sebastian Raschka's LLM Architecture Gallery. The project currently provi…
Occamy-1.0, an open 35B co-work agent model trained from Qwen3.6-35B-A3B, lands at the low-cost knee of the cost-performance Pareto frontier across four representative co-work benchmarks, according to its arXiv paper. Th…
An unofficial open-source project, codex-chatgpt-web, lets users run ChatGPT Web models directly inside OpenAI's Codex by routing selected tasks through a task-bound ChatGPT Temporary Chat via a local Responses bridge. T…
Daniel Miessler outlined a two-layer personal knowledge management (PKM) architecture that he argues makes a truly personal AI possible, with raw source documents kept untouched in a lower layer and an indexed layer of s…
Y Combinator CEO Garry Tan told TechCrunch he would "do nothing" to curb AI model distillation and instead wants U.S. regulators to let smaller American open-weight labs distill from domestic frontier models, arguing the…