Search “Reasoning”

AI education and training agents are typically built in five layers: the perception access layer, the brain reasoning layer, the teaching action execution layer, the knowledge and storage support layer, and the security and operations assurance layer. The brain layer uses a central scheduling control unit for intent recognition and routing distribution, and schedules multiple agents such as the master control, grammar correction, and question generation and assessment agents to collaborate. The

Matched fieldsSummary

Matched fieldsSummary

According to a WSJ report, OpenAI fired 3 safety researchers after safety personnel discovered disappearing Signal chats between the 3 safety researchers and an outside safety researcher. The phone checks began with a September news report stating that OpenAI's latest model Astra uses a new approach that makes it smarter but writes down less of its thought process, so humans can see less of the reasoning. OpenAI treated the report as a leak of its secret model design and began tracing the source

Matched fieldsSummary

DeepSeek has integrated the Harness with the web, App, and open platform teams into a product department, led by Cui Tianyi, and the official website has posted recruitment positions in two directions: AI product manager and Agent Harness. The article uses OpenAI Codex lead Tibo Sottiaux and the Anthropic Claude Code team as examples to explain how the execution capabilities of coding tools can be extended into a general work task entry point. DeepSeek, while open-sourcing Harness and the infere

Matched fieldsSummary、Body

The Shanghai Municipal Commission of Economy and Informatization released the "Shanghai Action Plan for Accelerating the High-Quality Development of AI-Empowered Industrial Services (2027-2028)", proposing that by 2028 the penetration rate of AI applications in industrial services reach over 85%, striving to cultivate 10 digitally native unicorn enterprises. The plan provides funding support ranging from up to 5 million to 20 million yuan for stages including R&D and design, supply chain service

Matched fieldsSummary、Body

Cloud hosting platform Railway discovered abnormal activity on October 21 in the system storing customer environment variables, stating that an unauthorized third party may have accessed encrypted environment variables and decrypted some of them, and has advised all users to change credentials such as API keys. Railway believes the attack may be part of a recent series of attacks targeting platform companies with the aim of breaching AI inference service providers, and has notified upstream supp

Matched fieldsSummary、Body

MiniMax announced on October 23 the open-sourcing of the M3.1 Flash model, which processed over 100 trillion tokens within 72 hours of launch. Compared with the previous generation M3, M3.1 Flash has comprehensively improved capabilities, with usage cost reduced by 70% and online inference speed increased by nearly 3 times. MiniMax stated it will work with chip manufacturers such as Huawei Ascend, NVIDIA, and AMD, as well as inference frameworks and communities such as vLLM, SGLang, and Hugging

Matched fieldsSummary、Body

Matched fieldsSummary、Body

Matched fieldsSummary、Body

Matched fieldsSummary、Body

MiniMax M3.1 Flash weights are now public, with 428B total parameters and 23B activated per token. The company claims substantial improvements in agentic tasks, coding, and cybersecurity: BankerToolBench 76.2, DeepSWE v1.1 73.7%, OSWorld V2 61.7, Agents' Last Exam 29.6, CyberGym 82.2. The model uses MiniMax Sparse Attention and DSpark speculative decoding to improve inference efficiency; the weights are now available on Hugging Face and ModelScope, with API and MCode desktop entry points also pr

Matched fieldsSummary、Body

OpenAI announced the full opening of the Decisions API built on GPT-6 Luna, aimed at questions with a limited set of answers, with response speeds up to 10 times faster than calling GPT-6 Luna through the Responses API. The API charges $0.10 per million input tokens, $0.01 per million cached input tokens, bills only for input tokens, and output tokens are free. Since the public beta, it has added native GPT-Live integration, prompt caching, and inference data residency support in the United Stat

Matched fieldsSummary、Body

OpenAI announced the full opening of the Decisions API, which is built on GPT-6 Luna and is aimed at questions with bounded answers, helping applications and agents make near-real-time decisions based on text, images, and structured inputs, with response speeds up to 10 times faster than calling GPT-6 Luna through the Responses API. On pricing, input is charged at $0.1 per million tokens, cached input hits are charged at $0.01 per million tokens, and billing is based only on input tokens, with o

Matched fieldsSummary、Body

OpenAI opened the Decisions API to all customers on Thursday; the product was announced at the September 29 Developer Day and is based on a reconfigured ChatGPT Luna to speed up inference and reduce costs. Nikunj Handa, OpenAI's head of API product, said latency has been cut from 1 to 3 seconds to 100, 120, and 180 milliseconds; the public version adds voice and vision modalities and deploys inference resources by region to reduce latency. In early trials at Box, Datadog, Ramp, and Speak, Speak

Matched fieldsSummary、Body

Matched fieldsSummary、Body

Cerebras deploys the roughly 3-trillion-parameter Kimi K3 on 59 WSE-3T wafers, achieving inference speeds of 1000-2000+ tokens/second. Each wafer has 44 GB of on-chip memory, and a single layer is about 16 GB, so each of the 24 dense layers occupies one wafer, while the 69 linear attention layers are placed two per wafer; only activations are passed between layers, at about 2 microseconds per hop. The remaining space on the dense-layer wafers is used for KV cache, which determines the upper limi

Matched fieldsSummary、Body

Microsoft has announced the research results of its AI diagnostic tool Diagnostic Orchestrator, which uses multiple AI agents to collaboratively reason about symptoms, decide on tests, and provide diagnoses, and is based on OpenAI's GPT-5. In tests drawn from hundreds of case reports from the New England Journal of Medicine, the system's diagnostic accuracy was about 83%, compared with about 36% for general practitioners who used reference materials and about 22% for those who did not; on emerge

Matched fieldsSummary、Body

Aditya Agarwal shared Sergey Levine's conversation at SPC: In his early years, Levine had a roomful of robots in a converted Google conference room spend all day grasping objects, accumulating a great deal of experience yet lacking prior knowledge; this bottleneck ultimately gave rise to Physical Intelligence. The conversation covered the Pi Zero robot foundation model, the evolution of reasoning and generalization capabilities from Pi 0.5 to Pi 0.7, as well as the robot data challenge and world

Matched fieldsSummary、Body

Modal releases the open-source training server Spindle, which can run in your own Modal workspace, is compatible with the Tinker API, and bills by GPU time rather than tokens. Deployment only requires running spindle deploy, and it supports full-parameter training and multiple LoRAs, covering a variety of open-source models. In the official comparison, a single-round DAPO-math task with 8-card H200 training, 8-card H200 inference, and 8 clients sharing costs $1.99 per million output tokens, 80%

Matched fieldsSummary、Body

At Fireworks AI, Bardia Shahali grew the sales team from about $3 million to about $400 million; he is now VP of Sales at Granola. He points out that at the AI infrastructure layer, individual deals can reach tens of millions of dollars and close in as little as two weeks, but all inference providers are competing for the same set of customers, making competition fiercer. Because model capabilities reach new heights every 45 days, compensation plans have shifted from annual to quarterly. At the

Matched fieldsSummary、Body