Amazon researchers unveiled SSFT and GFPO, techniques that train LLMs to generate diverse reasoning paths and select the optimal strategy.
A self-healing AI agent platform for legacy Windows desktop automation
Clawpatrol, a security firewall that controls AI agent actions based on rules, has been released.
Harbor v0.4.19 has been released, supporting agent execution based on local inference backends and automatic injection of web search tools.
BusPatrol is pushing to let police use license plate data collected by school bus AI cameras.
Anthropic is introducing an 'AI Fluency scorecard' to Claude that assesses users' AI collaboration skills and suggests areas for improvement.
Bank of America (BofA) forecast that AI will trigger a productivity revolution surpassing that of electricity or the internet, bringing an effect more than 10 times greater to the economy.
Chunk sidecars have been launched to quickly validate AI agent-generated code in a CI environment before committing.
WAVE, a portable GPU ISA and toolchain that runs across various GPU architectures, has been unveiled.
The open-source tool Ratel uses the BM25 algorithm to dramatically reduce MCP tool context usage in Claude Code.
Keye-VL-2.0-30B-A3B, the latest 30B-class multimodal model in the Keye series, has been released.
TogetherAI has open-sourced OSCAR, a new KV cache quantization technique that improves inference efficiency.
As frontier model costs rise, combining engineers from low-wage countries with local AI is emerging as a more economical alternative.
Eagle 3.1 was released through a collaboration between the EAGLE, vLLM, and TorchSpec teams.
BoquilaHUB 0.5 has been released, supporting the latest SOTA AI models for bioacoustics analysis.
Hugging Face has released a practical guide that clarifies the complex terminology surrounding AI agents.
A phone calling infrastructure service has launched to help AI agents directly make calls and hold conversations.
The Chinese government has expanded overseas travel restrictions targeting key AI talent at private companies.
Uber is struggling to justify its AI investment because there's a missing link between rising AI token consumption and actual feature improvements.
SkillOpt research has been published, boosting agent performance by optimizing markdown skill files as if they were learnable parameters.
A security vulnerability has been discovered in which insufficient Host header handling in Starlette can allow authentication to be bypassed in major AI infrastructure such as vLLM and LiteLLM.
Anthropic has launched the 'Claude Certified Architect - Foundations' certification exam, which verifies proficiency in using Claude.
An analysis has raised the issue that LLMs may distort instructions by processing hidden system/developer prompts, which take priority over users' custom instructions, first.
The DoJ arrested two men who used AI to create and distribute non-consensual sexual exploitation material.