Russian Families Use AI to "Revive" Soldiers Killed in War — The Ethical Frontier
According to a BBC report, families of Russian soldiers who lost their lives in the war in Ukraine are using AI technology to perform a "digital resurrection" of the deceased. Services that harness generative AI to let people "talk" with their lost loved ones again — through recreated voices and faces, interactive avatars, and more — are reportedly being used.
While this phenomenon highlights a new role that AI plays amid grief, it also raises serious ethical concerns. The technology is being applied in areas where legislation has yet to catch up — from the risk of exploitation for war propaganda to the personality-rights issues posed by posthumous digital recreation without the deceased's consent. In the broader context of deepfake technology being used in warfare, serious debate will be essential going forward.
The Rise of Low-Cost Chinese LLMs — DeepSeek V4 Pro, GLM 5.2, MiniMax M3
On Hacker News, the question "Which cheap Chinese LLM are you using?" drew significant attention. The poster noted that since DeepSeek V4 Pro, high-performance, low-cost Chinese models have been arriving one after another — including Mimo V2.5 Pro, MiniMax M3, and the recently released GLM 5.2. The community has been exchanging opinions on real-world experiences and the pros and cons of each.
Offering performance on par with leading Western models at a fraction of the cost makes these models attractive to developers who rely on API access. At the same time, concerns have been raised about data privacy, geopolitical risk, and output censorship — meaning decisions require criteria that a simple price comparison cannot capture.
$342 Vanished Just by Resuming a Claude Code Session — How to Prevent Token Explosions
On Qiita, a case was reported in which roughly $342 (about ¥50,000) was consumed simply by resuming a Claude Code session. In GitHub Issue #38029, a user reported that 650,000 output tokens were burned through in an instant without any input on their part.
The likely cause is that when a session is resumed, the accumulated context gets reprocessed, unintentionally consuming a massive number of tokens. The article introduces two hook configurations to prevent this "token explosion." For developers using Claude Code, cost management has become a key concern second only to SDK and API usage.
Is Your Content Reaching AI Crawlers? — A CI Check for AI Visibility
Also on Qiita, a method was introduced for verifying whether single-page application (SPA) content is properly reaching AI crawlers. With client-side rendering (CSR), the body of the initial HTML returned by the server tends to be empty; while Googlebot executes JavaScript and can eventually read the content, AI crawlers such as GPTBot, ClaudeBot, and PerplexityBot do not execute JS, leaving the content invisible to them.
The article presents an implementation that uses curl to send actual requests with each AI crawler's User-Agent and automatically checks in CI whether the returned HTML contains the body content. With the spread of AI-powered search, making content "visible to AI" is emerging as a new optimization frontier alongside SEO.
Google Custom Search API to Shut Down in 2027 — Searching for Alternatives
It has been revealed that the Google Custom Search JSON API will be discontinued in 2027. This API is widely used as a web search tool for MCP servers and AI agents, so the impact is far from negligible.
Candidates for replacement include the Brave Search API, SerpAPI, and Tavily, each differing in pricing structure, features, and rate limits. Which option is the right search backbone for your AI agents must be judged based on your use case and budget. Migration should be planned to be completed before 2027.
---