Reuters reports that, as chip export controls tighten, Chinese teams are using outputs from US models to train homegrown rivals, pointing to a new training route under compute constraints.
The poster noticed that Claude-coded projects start fast but deliver diminishing returns over time, and asks the community who is actually benefiting from AI coding.
The data platform serving top-tier banks moved its agent stack onto Bedrock, using a multi-model setup with strict tenant isolation for financial data.
The post treats multi-agent coordination as a distributed-systems problem and lays out a field-tested set of patterns, anti-patterns and trade-off rules.
An OpenAI physicist pushed forward on a puzzle that had baffled particle physicists for more than 25 years with help from GPT-6 Astra, putting AI's real problem-solving power in frontier theory on display.
Moonshot AI stopped chasing coding benchmark scores and is steering its Kimi models toward biology, chemistry and pharma research, where science tasks demand more capable models.
OpenAI is bringing six Apollo Research evaluators in-house to look for subversive behavior in its models, though limits on access and disclosure cast doubt on how independent the review can be.
Alibaba launched its flagship Qwen4-Max with 2.4 trillion parameters and only 71 billion activated per token, and its hybrid architecture sharply cuts long-context costs while keeping pace with top rivals.
Training an AI model requires thousands of GPUs working together, and Lambda is using NVIDIA Photonics co-packaged optics to tackle the networking speed and reliability challenge.
On Bedrock, prompt caching lets open-weight models reuse the long context that generative AI apps otherwise resend with every request, cutting input costs and latency.
This article argues that as AI agents enter core business workflows, the old pattern of scattered experiments no longer suffices, and enterprises need AI that works across departments.
OpenAI has kicked off its first round of embedded evaluations, with Apollo Research, METR, Transluce, and Goodfire independently assessing how models line up on alignment, safeguards, and monitoring.
Alibaba officially launched its new-generation foundation model Qwen4-Max on November 16, with stronger agent and coding capabilities moving toward real-world, all-scenario applications.
S1-Evo shows that typed System-1 decision models deployed for agent routing, moderation, and authorization can still be bypassed by carefully crafted evidence-competition attacks.
This work quantifies uncertainty for multimodal LLMs with an incoherence-adjusted semantic volume, so plausible but wrong outputs are not mistaken for reliable ones.
This work aligns speculative-decoding draft models with parameter- and data-efficient adaptation, easing the performance drop draft models suffer after target models are fine-tuned.
This study disentangles three factors often confused in multi-agent evaluations, clarifying whether gains come from coordination between agents, ensembling results, or simply more test-time compute.
Attacks that stopped working against single frontier models long ago can still succeed at the system level against multi-agent setups such as Codex and Claude Code, the study finds.
This study evaluates LLMs' risk propensity across contexts and task demands, systematically measuring whether models lean risk-averse or risk-tolerant.
BigCodeBench evaluates LLM code generation with tasks involving diverse function calls and complex instructions, closer to real development than benchmarks that only test simple snippets.
REFORGE benchmarks LLM reverse-engineering capability through decompiled binary function naming, responding to real threat-intelligence reports of LLMs being used for reverse engineering.