According to The Information, citing internal Tencent memos and sources, Tencent employees used Anthropic’s Claude Code during the post-training phase of Hy3, the company’s latest large language model, despite Anthropic’s explicit ban on commercial services to Chinese firms citing national security concerns.
Hy3 features a 295B parameter mixture-of-experts architecture. During the RLHF (reinforcement learning from human feedback) stage, Tencent staff served as human evaluators with usage capped at thousands of tokens per person. Internal memos provided Claude Code installation guides. Tencent employees used Claude Code to generate high-quality reference examples for scoring anonymous model outputs, rather than what they characterized as distillation. An Anthropic spokesperson stated the company actively monitors for distillation attacks but did not directly address Tencent’s use of Claude Code.
Disclaimer: The information on this page may come from third parties and does not represent the views or opinions of Gate. The content displayed on this page is for reference only and does not constitute any financial, investment, or legal advice. Gate does not guarantee the accuracy or completeness of the information and shall not be liable for any losses arising from the use of this information. Virtual asset investments carry high risks and are subject to significant price volatility. You may lose all of your invested principal. Please fully understand the relevant risks and make prudent decisions based on your own financial situation and risk tolerance. For details, please refer to
Disclaimer.
Related Articles
Sam Altman and Dario Amodei are both unbearable! AI doomsday talk and relative deprivation have made Americans increasingly resent AI
A well-known Silicon Valley podcast, 《All-In Podcast》, discussed the AI industry and offered a rather sharp observation: American society’s attitude toward AI is shifting negative, and the most concrete outlet for this sentiment is the data centers that AI companies are racing to build across the U.S. The source of this backlash may include AI doomsday fears, fear of unemployment, or a deeper sense of dissatisfaction: it seems that yet another wave of technological innovation will only make a small number of people get rich, while most people’s lives show no clear improvement.
Local U.S. governments have already overturned data center construction cases
On the show, Chamath Palihapitiya said that the current problems facing the AI industry are not just competition among models, capital expenditures, or a shortage of compute power, but that “the American people, overall, are becoming increasingly repulsed by the whole issue of AI.” He noted that the source of this repulsion may include AI companies
ChainNewsAbmedia3m ago
Community Questions Ideological Bias in Mainstream AI Models on May 4
According to BlockBeats, on May 4, AI community user X Freeze questioned whether mainstream AI models including ChatGPT, Claude, and Gemini show systematic bias by being less aligned with conservative positions on issues such as gender, immigration, and crime. The user suggested that as AI
GateNews1h ago
Cerebras Plans $4B Nasdaq IPO, Starts Roadshow on May 4 at $115-$125 Per Share
According to Reuters, AI chipmaker Cerebras Systems started its IPO roadshow on May 4 for a Nasdaq listing under the ticker CBRS. The company plans to price shares at US$115 to US$125 and could raise up to US$4 billion at a valuation of approximately US$40 billion. This marks Cerebras' second
GateNews2h ago
AI drives the U.S. Q1 GDP growth up 75%, and the top five companies’ capital expenditures in 2027 may exceed $1.1 trillion
Morgan Stanley once again raised its forecast for capital expenditure by the top five U.S. cloud giants: about $805B in 2026 and about $1.116 trillion in 2027, and in 2026 spending on non-tech sectors will be roughly comparable to 2025. David Sacks said AI capex contributes about 2.5% to GDP and could exceed 3% next year, viewed as a new engine for the U.S. economy.
ChainNewsAbmedia2h ago
To cash in on the IPO boom for SpaceX, OpenAI, and Anthropic, Nasdaq and S&P ease standards
The Wall Street Journal said that, to prepare for IPOs from SpaceX, OpenAI, Anthropic, and others, the S&P and Nasdaq have loosened the criteria for component stock selection. The S&P has cut the new-share observation period to six months, or waived the profit threshold; the Nasdaq 100 has introduced fast-track inclusion, allowing ultra-large IPOs to be added to the index on the 15th day after listing. Selection criteria have been changed to rely on total market capitalization, remove the minimum free-float requirement, and adopt dynamic weighting. Experts questioned whether it could amplify the risk of popular stocks and distort price discovery when it may be insufficient.
ChainNewsAbmedia2h ago
ASX Warns Companies Not to Overstate AI Impact on May 4
According to Bloomberg, on May 4, Australia's stock exchange operator ASX warned companies not to overstate the impact of artificial intelligence on their businesses. ASX Chief Compliance Officer Lucinda McCann said the exchange was watching for "ramping" or claims intended to lift share prices
GateNews2h ago
