Large language model (LLM) news — GPT-4, Claude, Gemini, Llama, Mistral and the latest research on training, fine-tuning, RLHF, and deployment of LLMs.
Steve Yegge shut down Gas Town, admitting that despite spending many thousands of dollars a month on coding agent subscriptions, Gas Town was the only thing he ever built with them. Separately, Databricks reported a 60% …
A developer documented a verification workflow for archiving a GitHub software release to Zenodo, using version 1.1.0 of an LLM-as-a-judge cost calculator as the worked example. Checks on September 17 confirmed the Zenod…
JB Financial Group signed a memorandum of understanding with AI company Upstage on Monday to develop AI agents for banking and financial services using Upstage's proprietary large language model technology, the group sai…
A developer published a technical explainer tracing the evolution of attention mechanisms in transformer models, from the original Multi-Head Attention (MHA) introduced in the 'Attention Is All You Need' paper to newer v…
A developer's side-by-side test found that Google's Gemini 3.8 Flash cloud model produced a faster Rust parsing function than the local Gemma 4 model (gemma-4-26B_q4_0-it.gguf), cutting starts_with calls from 4 per byte …
A developer built a local personal AI assistant running on two 12 GB consumer GPUs, combining an LLM with RAG, graph memory, MCP tools, and voice input/output to handle documents, email, calendar, Telegram, and smart-hom…
A developer ran 32 controlled generations across four open-weight models (DeepSeek v4 Pro, Llama 4 Maverick, GLM 5.3, Mistral 3 14B) using the same prompt but swapping in two different commits of their own football quiz …
Mathematician Terence Tao argues that large language models should be understood not as intelligent agents but as "a new kind of cultural and social technology," citing a paper by Henry Farrell, Alison Gopnik, Cosma Shal…
A developer behind the Codenames AI web game added Simplified Chinese support by using an LLM to generate candidate translations while building deterministic validators and human review checkpoints around them. The devel…
ContextPress, a Python library that compresses LLM context through rule-based logic rather than model calls, added five new compression stages since its initial release, according to a follow-up post by its developer. Be…
Google Research introduced Retrieve-for-Train (R4T), a search framework that trains a fan-out language model with reinforcement learning once, then uses that model to synthesize training data for a 53.9M-parameter diffus…
A tutorial published on Modal Cloud's platform shows how to fine-tune the 135M-parameter HuggingFaceTB/SmolLM2-135M-Instruct model on the b-mc2/sql-create-context dataset using a single YAML configuration file and no Pyt…
Union Alpha is a multimodal model built for research, coding, and agentic workflows that delivers frontier-level performance across diverse general-purpose tasks, according to third-party information. The model's documen…
A developer analysis published on tamiz.pro warns that codebases heavily composed of AI-generated code accumulate a distinct class of structural technical debt that typically surfaces around the third month of maintenanc…
A developer proposes a hard ceiling of roughly 300 lines for always-on agent instruction files such as CLAUDE.md, arguing that selective attention and per-request token spend cause long configs to be quietly ignored. The…
TypeSafe AI launched Jev, a "System One Model" that returns typed, structured outputs with calibrated confidence scores instead of generating text, priced at $0.042 per million tokens and trained with a method TypeSafe c…
Moonshot AI released Kimi K3 on 2026-07-16, a model that handles text, image, and video input with a 1049k-token context window and 131k maximum output tokens. No retirement date has been announced for Kimi K3, which is …
Google released GLM-5.2 on 2026-06-13 under the model identifier zai-org/glm-5.2-maas, offering a 1000k-token context window and 64k maximum output, with no retirement date announced. The model listing on models.dev poin…
SK hynix showcased its next-generation AI memory portfolio under the slogan 'New Spectrum' at the AI Infra Summit 2026, held September 15th-17th in Santa Clara, California, where the company doubled the size of its booth…
A developer surveyed five coding agents — Claude Code, Cursor, Codex, VS Code, and Windsurf — and found that all of them already support resuming a prior conversation, so the common complaint that agents "forget" history…