Reading

Ride-hailing company inDrive, Chinese autonomous driving firm Pony.ai, and US chipmaker Nvidia announced on Friday that they have signed a memorandum of understanding to test robotaxis in Astana, the capital of Kazakhstan. According to inDrive CTO Yuri Misnik, testing will begin in the first half of 2027, initially with a safety operator on board, with fully driverless rides as the eventual goal.

Under the proposed plan, Pony.ai will provide its Gen-7 robotaxis, which inDrive will offer through its ride-hailing app. Nvidia will supply the underlying AI technology, including its Hyperion platform and Halos safety system. Ali Kani, Nvidia’s vice president of automotive, stated that the Hyperion platform could lower development costs by providing a ready-made foundation of sensors and computing, a factor he noted is particularly relevant in emerging markets where cost competition is key.

The choice of Astana is partly driven by its harsh winters. The city, home to around 1.7 million people, faces months of snow and ice, offering a challenging environment for autonomous systems. Arsen Tomsky, founder and CEO of inDrive, said that success in Astana could provide valuable insights for expanding autonomous mobility to other locations facing similar conditions. Kazakhstan has previously announced plans to launch autonomous taxi services in 2027 and has introduced rules for testing on public roads.

This move aligns with broader industry trends as ride-hailing giants expand into robotaxis. Earlier this month, Uber announced plans to test Pony.ai’s robotaxis in London. inDrive describes its approach as “capital-light,” partnering with various firms in different markets to combine expertise rather than bearing the full burden of R&D and operations alone.

Bayer Crop Science manages hundreds of machine learning model configurations covering genomics, field trials, imagery, and environmental data. Breeders, analysts, and researchers need to run these workflows repeatedly, but the true bottleneck is not execution—it is understanding. When a breeder asks, 'Why did my phenotyping workflow fail yesterday?', they need to know which system to check for logs, which API to use for status, and which document explains the model behavior. Bayer built the Agentic Decision Engine (ADE), a multi-agent natural language interface, on top of its existing Connected Decision Engine (CDE). ADE uses 22 specialized agents responsible for conceptual Q&A, workflow status queries, fault diagnosis, and visualization, orchestrated by a supervisor agent that handles intent classification and routing.

An AWS blog post recorded five design decisions, the most notable of which were not the architecture itself, but the real failures exposed during the production phase. First: ADE's exception handlers caught internal agent errors and returned format-normal responses, causing monitoring systems to count them as successful. A supervisor architecture turns one user request into a chain of model calls, tool calls, and agent handoffs; the final answer can be completely wrong while every HTTP request returns a 200 OK. The team added custom CloudWatch metrics and alarms for specific conditions within failure paths—tool call errors, throttling, model timeouts, and structured output validation failures—to distinguish transport success from task success. Second: During construction, the team evaluated a newer model version with reasoning mode enabled and found that the combination of tool calling and structured output was incompatible with ADE's agent implementation. The Domain Specialist Engineer (DSE) confirmed this constraint, and the team retained the current model release. The article summarizes this as a principle: for agents using tools, model selection is constrained by the full interface—including tool calling, structured output, streaming behavior, reasoning features, latency, quota availability, and model-specific request semantics—an update does not equal plug-and-play.

The remaining three decisions also point to specific frictions in productionization: orchestration state must be externalized to durable storage (LangGraph checkpoints stored in DynamoDB), otherwise multi-step workflows cannot recover after scaling or restarts; user identity (Microsoft Entra ID), inter-service authorization (AWS IAM), and downstream tool access (AgentCore Gateway outbound authorization) must each define their own token contracts independently, because identity failures cluster at the transitions between user identity, agent identity, and tool identity; and the tool catalog must be scoped per agent, because every tool definition consumes input tokens and expands the model's decision space. CDE manages over 500 model configurations; if each API had a dedicated MCP server, the overhead for deployment, authentication, and version management would be huge. By registering OpenAPI specifications through the Gateway instead, the team reported reducing per-API integration overhead by more than 99%.

Taiwan-based e-commerce technology service provider 91APP published its '18-Day Investigation Report on the 2026 Cyberattack Incident' on October 22, detailing how its Commerce Cloud database was subjected to automated attacks. The report revealed that the primary breach occurred due to an unpatched vulnerability in Appsmith, a third-party tool used internally, which allowed hackers to access the internal network and obtain database access keys from GitLab.

The investigation team discovered traces of the ARTEX automated attack tool within the system. ARTEX is an AI-driven autonomous penetration testing system capable of using AI to automatically find vulnerabilities, plan intrusion paths, launch attacks, and steal data. He Yingqi, Chairman of 91APP, acknowledged in an official announcement that the company failed to alert brand clients when it established an emergency response team on October 8, admitting this as an oversight and apologizing to affected brands and consumers.

91APP added that between September and October 2026, multiple banks in South Korea, approximately 20 companies in Japan, and over 100 corporate websites in the United States were targeted or locked by similar AI automated attack tools. This finding indicates that AI-based automated attack methods have emerged in cybercrime activities across multiple countries, extending beyond specific regions or industries.

Currently, one order database within Commerce Cloud has been confirmed as illegally accessed, involving order information but excluding payment instrument details such as credit cards. Marketing Cloud, 91APP Payments, and iCHEF were also unaffected. 91APP has removed the compromised Appsmith tool, replaced internal system tokens, database accounts, and passwords, rebuilt GitLab permissions, and restricted connections between CI hosts and database network segments. The company promised full compensation for consumers who suffered financial losses due to fraud resulting from this incident and will negotiate individual compensation measures with each brand for affected members. As of the report's release, the actual number of affected brands, members, and the specific scope of stolen data remain under confirmation; the company stated it is cooperating with Taiwan's Ministry of Digital Development and expects to publish further details on the impact scope and system improvements next week.

Precipitation nowcasting aims to predict rainfall intensity in localized areas over short timeframes (e.g., 0–6 hours), which is critical for generating emergency alerts and airport weather guidance. Traditional methods primarily rely on numerical weather prediction or radar echo extrapolation based on optical flow, but the latter is limited by separated steps and difficult-to-determine parameters. This paper proposes treating precipitation nowcasting as a spatiotemporal sequence prediction problem and designs a Convolutional Long Short-Term Memory network (ConvLSTM). The model introduces convolutional structures in both input-to-state and state-to-state transitions, thereby effectively capturing spatial correlations. Experiments show that ConvLSTM outperforms fully connected LSTM (FC-LSTM) and the ROVER algorithm developed by the Hong Kong Observatory on both synthetic datasets and real radar data, enabling more accurate prediction of future frames. For researchers engaged in meteorological AI, video prediction, or spatiotemporal data mining, it is recommended to start with the introduction to understand the problem background, focusing on Section 3 for model architecture design and Section 4 for experimental comparison results. Readers are encouraged to consult "Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting" (2015) to deeply understand the construction logic and empirical effects of this classic spatiotemporal prediction model.

In the 1986 shareholder letter, Warren Buffett, Chairman of the US-based Berkshire Hathaway Inc., delves into two key traps in measuring a company's true value: the limitations of book value and the misuse of 'cash flow' by Wall Street. This letter is not only an exposition of investment methodology but also a profound analysis of the economic substance behind accounting figures.

First, Buffett points out that for most companies, there is a significant difference between book value and intrinsic business value, with the latter being the number owners truly care about. Although Berkshire's own book value once served as a reasonable proxy for business value, the two are not identical. He specifically emphasizes that 'goodwill' created by excellent managers through improved profitability often does not appear in the net worth on the balance sheet. For example, the managers mentioned in the text, such as the Blumkin family and Mike Goldberg, significantly increased business profitability with almost no additional capital. This unrecorded gain constitutes real economic value, which may be masked by simple book growth. Therefore, investors should not just stare at the growth rate of book value per share, but should focus on the management capabilities that can generate sustained excess returns.

Secondly, Buffett severely criticizes the popular concept of 'cash flow' at that time. He believes that many investment banks and analysts add depreciation back to net profit to calculate cash flow, while ignoring the capital expenditures required to maintain competitiveness. He wrote in the letter: "If all American companies were sold simultaneously through our major investment banks—and believing the sales brochures describing them—the government's projections for national plant and equipment expenditures would be cut by 90%." This satire reveals the absurdity of ignoring maintenance capital expenditures. For enterprises in manufacturing, retail, etc., that require continuous reinvestment, so-called 'free cash flow' is meaningless if it does not deduct necessary update expenditures. Buffett advocates using the concept of 'Owner Earnings', which is net profit plus non-cash expenses such as depreciation, minus the average capital expenditure required to maintain current competitive position.

To help readers understand this mechanism, we can imagine a hypothetical scenario: a century-old printing press reports $1 million in net profit and $500,000 in depreciation annually. The traditional algorithm would claim its cash flow is $1.5 million. However, if the plant must spend $2.5 million every five years to replace core printing equipment to maintain market share, the annual average maintenance cost is $500,000. At this point, the real distributable 'owner earnings' to shareholders are only $1 million, not $1.5 million. Investors who ignore this maintenance cost may overestimate the company's dividend capacity and valuation.

The 'Berkshire Hathaway 1986 Shareholder Letter' is suitable for long-term investors who wish to go beyond surface financial data and deeply understand corporate economic moats and management effectiveness. It is recommended to start with the sections on 'Sources of Reported Earnings' and the appendix, focusing on reading its analysis of the gap between accounting figures under GAAP standards and actual business value.