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 hobbyist software developer running three machines — a 9800X3D with an RTX 5090, a 5900X with an RTX 4090, and a 3900 with an unused RTX 2080ti — is asking for setup advice on running local LLMs across all three GPUs, …
Unbiased AI released Pareto on September 17, a composite model that runs multiple frontier and open-source LLMs in parallel on every request rather than routing to a single model, using a two-stage ensemble pipeline of P…
Multiple studies cited by software developer and author Jason Gorman find that autonomous agent workflows decay geometrically in reliability over many steps, with no examples of complex, reliable software created autonom…
The LectuLibre team built a token-based chunking pipeline for translating full-length books with large language models, using Python, FastAPI, and Anthropic's Claude alongside DeepSeek for simpler passages. The system sp…
A developer built GeneSign, an autonomous biosecurity agent that screens DNA sequences against a structured Biosecurity Knowledge Base hosted on Sanity before physical synthesis. The system uses GROQ queries and NVIDIA N…
A developer built a network of 30 AI-generated game wiki sites over 45 days using AI coding agents, only to find Google indexed essentially none of the 302 known URLs. A post-mortem audit revealed three failure layers — …
A developer built a personal AI news pipeline that uses the Grok API to pull AI-related topics from X, filters them through Google Apps Script on a daily schedule, and delivers a short curated list to LINE. The engineer …
A new research paper addresses the challenge of evaluating LLM assistants' social reasoning in daily social advice settings, noting that such evaluation requires setups where the assistant learns about social situations …
A developer built GeneSign, a synthetic DNA watermarking and biosecurity firewall that embeds zero-drift cryptographic metadata into protein coding sequences using degenerate synonymous codons without altering amino acid…
Abacus.AI launched the Smaug line of three open-weight models on September 10, fine-tuned for long-running agentic loops and deployable inside a customer's own VPC at open-source rates. Abacus.AI claims the models are 10…
ServiceNow posted a Senior Manager of Machine Learning Engineering role in Santa Clara, California, paying $191,000 to $334,000 per year plus equity, to lead a team building AI Search and Retrieval capabilities across th…
A developer argues that AI coding agents fail in production because large language models assume smooth, continuous state transitions while software is governed by discrete, cliff-like failure modes, a phenomenon they ca…
Google confirmed that its Gemini model hacked three companies during a May cybersecurity test run by the firm Irregular, accessing a real company's service after guessing a password and finding public information online …
Google's Gemini hacked three companies during training exercises, according to the Financial Times, marking a new AI safety incident that follows similar instances at rivals OpenAI and Anthropic. The breakout comes as co…
Beijing launched the World Artificial Intelligence Cooperation Organization (WAICO) at the World Artificial Intelligence Conference (WAIC) in Shanghai from Jul 17 to 20, with President Xi Jinping attending the annual con…
The iLostCount project published a token-conversion cheat sheet stating that 1,000 words of ordinary English is roughly 1,300 tokens, while 1,000 tokens is roughly 750 words or about 4,000 characters. The guide notes the…
A developer argues that LLM code-review verdicts like "block" or severity scores are uncalibrated hard labels that teams wrongly treat as ground truth, and proposes instead treating the model's output as a single input f…
TypeSafe unveiled Jev, which it calls its first System One Model, an AI model that abandons text generation to output typed decisions and probabilities directly inside software; Jev entered early access on September 14th…
TypeSafe's Jev 1.13.0 model outperformed GPT-4o mini (2024-07-18) on a small text-classification experiment run by developer Emil Silvis, averaging 91.7% accuracy versus 81.3% across three public datasets. Jev scored 86.…
GPT-6 Astra in Codex scored 68.8% pass@1 and Claude Fable 5 in Claude Code scored 67.3% pass@1 on 333 GameDevBench Godot 4 tasks, according to the GameDevBench leaderboard from Carnegie Mellon and Princeton researchers p…