Blog
Daily notes on new models, LLM releases, agent frameworks and AI research, written from the sources we follow and delivered as a newsletter every day.
Get the daily issue
Every new post of the day, in one email. Confirmation required.
LLMs1 min read
arXiv:2609.13709v1 Announce Type: new Abstract: Human corrections identify editable spans, but the examples receiving corrections may come from a selective feedback channel. We analyze this interaction at a fixed model checkpoint by deco...
From arXiv cs.CL
Research1 min read
arXiv:2609.13637v1 Announce Type: new Abstract: Persistent agents need evaluations that distinguish identity facts they can recall from those they express and enact. We introduce PAI-Bench, a provider-neutral benchmark for fidelity to a ...
From arXiv cs.AI
How this blog is made
Each feed entry becomes one request to OpenSmartRoute: the router picks a model with a cost-weighted objective, the editorial-writer skill is layered on the prompt, and the outcome trains the learners - the same pipeline available to every workspace.
Open any post to see which target answered, its confidence, the alternatives and what the request cost. Run the same pipeline yourself: register feeds in the operator console, map a small model under Providers, or call POST /api/v1/route with execute: true.
Research1 min read
Research expands on The Core algorithm, demonstrating that agents specifically niched to subsets of the Xpilot environment outperform general and partially niched agents. This highlights the potential for targeted agent evolution in complex systems.
From arXiv cs.AI
LLMs1 min read
GraphProfiler is a new LLM-based profiler that builds a personal knowledge graph for each user, linking posts to their source texts. It achieves high attack success rates on benchmarks while providing detailed attribution of the posts contributing to inferences, enabling targeted privacy mitigation strategies.
From arXiv cs.CL
AI1 min read
Moonshot AI projects $2 billion in annual revenue for its K3 model by the end of the year, doubling its August run rate. This reflects the model’s usage, currently generating 300 billion tokens daily, despite recent declines.
From TechCrunch AI
LLMs1 min read
The Qiushi Engine team conducted a research program using BabyLM 2026 Strict-Small, achieving an Overall score of 42.25 across two generations. The program utilized a recursive self-improvement process (Research RSI) to optimize model efficiency through targeted data organization and experimentation.
From arXiv cs.CL
LLMs1 min read
arXiv:2609.11020v1 Announce Type: new Abstract: We study K/V-cache interventions -- transplanting a target-conditioned K/V trajectory into a source-persona generation -- as a structured surface for persona control in decoder-only languag...
From arXiv cs.CL
AI1 min read
Anthropic’s new report details a significant increase in distillation attacks targeting its models, including large-scale campaigns from Alibaba and Moonshot AI, raising concerns about intellectual property and model security.
From TechCrunch AI
LLMs1 min read
Osprey pretrains a small language model to accelerate speculative decoding, improving acceptance rates across diverse target models. The system reduces per-target adaptation work through a reusable backbone and lightweight adjustments.
From arXiv cs.CL
LLMs1 min read
A new benchmark, CLLPU, assesses multilingual LLM unlearning by evaluating the extent of knowledge suppression across languages. Experiments with Llama-3.1-8B-Instruct show that universal suppression often fails to fully remove target knowledge, while language-conditioned confinement leads to uncontrolled propagation.
From arXiv cs.CL
Research1 min read
A new design approach, TEAM-Design, optimizes human-agent team deployments by strategically allocating replay budgets to the most uncertain comparisons. This reduces wasted time and compute when human-AI workflows don't outperform individual alternatives.
From arXiv cs.AI
LLMs1 min read
Three open-weight LLMs were assessed for demographic misalignment across countries using Wasserstein distance; targeted fine-tuning reduced bias but redistributed it among personas.
From arXiv cs.CL
LLMs1 min read
A framework for training language models to refuse harmful or targeted queries, improving safety boundaries while maintaining factual answering capabilities.
From arXiv cs.CL
LLMs1 min read
Replacing old systems with new ones often leads to dual systems in production, with ongoing risks and incomplete replacements. Automated testing and targeted refactors can improve success chances.
From Simon Willison
Agents2 min read
OpenAI projects AGI completion by December 2026, driven by the Astra model, while the launch of Pollen Robotics’ Microduck robot gains traction with its open-source design and accessible price point.
From Latent Space
Posts are drafted from public feeds by models OpenSmartRoute routes to - the same router, skill and metering customers use - and always link to the original source. Corrections: support.
