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 paper proposes Muon-RT, a variant of the Muon optimizer for large language model self-distillation that reuses the exponential moving average (EMA) teacher's trajectory instead of storing a separate momentum buffer, el…
JustFit v2.3.1, released on GitHub Releases and PyPI by developer YuhuaBillChen, serves a 27B language-and-vision model with its full 256K context on Macs with at least 24 GiB of unified memory. At 16K context, JustFit v…
Typemill will add an MCP layer in version 2.29.0 that lets AI services connect to and use a Typemill installation through the platform's existing API, according to the Typemill documentation. The content management syste…
An Apache-2.0 diagnostic package called TrainSight (trainsight-preflight) was released to catch silent fine-tuning failures that occur between raw JSONL training records and the computed loss tensor, including truncation…
A developer who built a menu bar app to tally Claude Code usage found 4.9 billion tokens consumed over 30 days, of which 98.1% were cache reads and just 0.37% were output tokens written by the model, according to the dev…
A 0.6B-parameter veto gate called EditVeto improves LLM-based post-OCR correction across all five rewriters tested, including two unseen in training, according to an arXiv paper (2611.05539v1). On the eight HIPE-OCRepair…
A new arXiv paper (2611.05517v1) introduces BACE (Belief-Aware Critic with Epistemic tracking), a lightweight action-chooser built on a frozen language model backbone that more consistently selects the optimal number of …
A controlled study of 10 LLM models across 1,000 synthetic search worlds found that final answers are more sensitive to the order of initial evidence during active search than when all evidence is supplied in full, accor…
A new arXiv paper (2611.05814v1) introduces Consultation Value Distillation (CVD), a method that estimates offline, via Monte Carlo executor rollouts, how much an LLM agent benefits from consulting a human at a given dec…
Researchers posted arXiv paper 2611.05731v1 introducing Pro-Memory, a memory framework for LLM agents that separates storing future-directed intentions from deciding when to act on them. Pro-Memory links pending objectiv…
A comparative evaluation of six large language models found authorship impersonation remained challenging for most, with Gemini 3 Pro showing the strongest apparent mimicry but only because it extensively copied and reus…
A new arXiv paper (2611.05804v1) finds that mismatching the inference-time expert-deactivation rule to the training-time weight rule inflates perplexity by 14 to 28,740 times the intact baseline, versus 1.4 to 73 times w…
An evidence gate that lets a reviewer model override an upstream agent only after producing a mechanically valid record of a binding-constraint violation and a compliant repair sharply reduced corruption of correct propo…
A new arXiv paper (2611.05827v1) introduces TandemBench, a benchmark of 1,200 questions in multi-agent cooking scenarios that tests how well decision models choose optimal team plans. The best of 40 native decision model…
A new benchmark, CARTBench, finds that 3 of 10 evaluated models exceed 90% on Core criteria for LLM shopping advice, but none exceeds 45% on Excellence, according to the arXiv paper 2611.05862v1. CARTBench pairs timestam…
A study posted to arXiv (2611.05877v1) finds that better confidence estimates in LLM-built knowledge graphs can yield little audit improvement, because an error-rate × fan-out queue leaves 3.98–4.74 percentage points mor…
A new arXiv paper (2611.05392v1) proposes Conformal Reservation for Orchestration (CRO), an online algorithm that converts a sliding window of past execution traces into dispatch-time compute and bandwidth reservations f…
A challenge-aware adaptive recruitment method that uses an external CHALLENGER to inspect an evolving reasoning trace and trigger targeted knowledge retrieval or mid-discussion specialist recruitment improved accuracy on…
A paper posted as arXiv:2611.05394v1 derives a closed-form condition for when no confidence threshold makes tiered LLM inference worthwhile compared with sending every query to the backstop model. Modeling routing as a b…
A benchmark study posted to arXiv (2611.05459v1) tested six AI tools across four literature-review stages against a published review of the Spanish plume weather pattern and its 102-article corpus, finding no single tool…