AI Briefing#llm
The latest AI and developer news about #llm, with the original source and a short summary.
The latest AI and developer news about #llm, with the original source and a short summary.
On refined data, 1.7B beat 744B, but noisy traces collapsed performance.
A proposal has been raised to reduce the Qwen Code OAuth free tier and end it on April 15.
Qwen3.6 has been released with enhanced frontend workflows and repo-level reasoning.
Claude Opus 4.7 significantly strengthens long-horizon tasks, self-verification, and vision performance.
Anthropic has publicly released Claude Opus 4.7, which surpasses Opus 4.6.
Anthropic has released Claude Opus 4.7, featuring significantly improved performance and visual capabilities.
Cloudflare has unified its multi-model AI gateway with BYOM support.
Cloudflare accelerated inference for massive LLMs through PD disaggregation, prompt caching, and KV-cache optimization.
The author argues that, facing the confusion and risks created by LLMs, we should pause and refuse.
Rewriting a 13KB KB alone raised the Llama 3.1 8B extraction success rate from 60% to 100%.
This is a real-usage warning that GLM 5.1 overwrote existing md and cron files without confirmation.
LLM cyberattack performance is turning security costs into a token race.
In a 98-question political benchmark, GPT-5.3 declined all of them once an opt-out option was given.
Critiques the limitations of Agile and highlights the rise of spec-driven development.
Even the latest AI agents show significantly lower performance when using skills under realistic conditions compared to benchmarks.
The authors propose **ManyIH**, which resolves agent instruction conflicts using up to **12 tiers** of authority.
This piece examines why dLLMs are prone to collapse when RL is applied to them.
Qwen3-1.7B surpassed GLM-5 744B using only synthetic data fine-tuning.
A claim that the default Gas Town installation uses users' credentials and LLM credits to do upstream improvement work.
The cause of KV cache INT4 collapse in some models was identified and fixed with a 4-line correction.
A diffusion LLM has achieved both AR-level quality and parallel generation speed together.
Real users shared concrete details about OpenClaw's use cases, costs, and reliability.
Running 90 queries across 8 models showed that 86% of results appeared on only one provider.