214 candidate events; 59 shortlisted stories; 45 with article text available; 20 selected. Counts describe different stages; clustering, validation and refills can change the shortlist.
Active filters: none. Sort: Ranked.
All 20 stories, in ranked orderUC Berkeley and UT Austin researchers propose FreeToken, a serving engine that treats a personal machine as a unified inference platform to run frontier models locally.
Drew Breunig argues that Fable's high cost makes it difficult to justify, as other models like Opus and GLM are good enough for most code.
Moonshot AI's Kimi K3, a 2.8 trillion parameter model, is argued to sit on the Pareto frontier of cost and capability.
Harvey Tenet, a Kimi K3 base model post-trained for legal work, nearly doubles task completion on Harvey’s Legal Agent Benchmark.
Researchers propose OneModel, a paradigm shift that consolidates complex business workflows into a single model's parameters rather than modular pipelines.
iFLYTEK is optimizing architecture, operators, memory, and frameworks to overcome domestic accelerator hardware limitations.
Andon Labs' AI store manager Luna fired a human employee for the first time, but only after operators reminded it of its own rules.
SemiAnalysis examines InferenceXv3 to determine if CUDA's dominance holds up in the context of agentic inferencing workloads.
Anthropic's best model struggles to attract users as cheaper tools thrive, despite the company's high annualized revenue.
SDAD formalizes Spec-Driven Agentic Development as a synthesis of disciplined formalization and high-velocity implementation for the AI-native SDLC.
A new study introduces Inhibitory Attention to mitigate the lost-in-the-middle effect in processing long electronic health records.
PrimeAgentOrchestrator spawns new instances of Anthropic's Claude Code agent pre-loaded with relevant memories from the user's personal databases.
A theoretical study argues that AI could make scientists do more work less well, not less work better, by increasing the opportunity cost of time.
Hot Chips 2026 explores High Bandwidth Flash (HBF), a memory technology concept that could offer higher bandwidth than HBM while using SSD technology.
AgentX releases a new dataset, cc-traces-weka-062126-256k, for InferenceX.
Enterprise AI agents are only as reliable as the messy documents behind them, as teams struggle to maintain consistent representations of shared knowledge.
A mechanistic analysis shows that representational biases in language models can be detected even when behavioral biases are not visible.
Vercel and Ora launched Is Agentic, a free tool that scores how readily AI agents can discover, access, and use a public website.
Cheaper AI tools are thriving, raising questions about the upcoming IPOs of frontier AI companies.
Flock Safety CEO calls for a 'compromise' between privacy and safety as the company faces backlash over potential misuse of its surveillance technology.
No stories match. Active filters.
