AI / Large Models
OpenAI and Hugging Face jointly disclose a model evaluation security incident
Both companies identified and addressed security vulnerabilities during model evaluation; specific details were not disclosed, but the incident has raised community concern about AI supply chain security.
Hot topic: Developers are discussing the trust boundaries of third-party model libraries.
ChatGPT officially launches its advertising platform
OpenAI has launched ads.openai.com, allowing brands to place ads within ChatGPT, further clarifying its monetization strategy.
Anthropic reaches $1.5 billion copyright settlement
The judge approved a $1.5 billion settlement by Anthropic for using pirated books to train Claude, the largest payout in AI copyright litigation.
Hot take: $1.5 billion is like giving a red packet to every book, but the model has already learned.
Kimi K3 matches Fable in performance, becoming the current SoTA.
Fireworks AI tests show that Moonshot's Kimi K3 performs comparably to Fable, making it one of the strongest models currently available, despite having far less computing power than OpenAI.
Hot discussion: The Chinese community is surprised by Kimi’s performance despite resource limitations, debating how “small compute power can produce large models.”
Fireworks AI | Hacker News | Zhihu
Microsoft is testing the Kimi model in Copilot
Microsoft is testing Moonshot's Kimi model within Copilot, and the Reddit community is discussing whether this signals the AI industry entering an "Intel Inside" era—where major companies no longer rely solely on their own models.
Google released Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber
The new model series covers lightweight, standard, and secure scenarios, but the community commonly complains that Google's AI product line is too fragmented, leading to a fragmented user experience.
DeepMind researcher Alex Turner publicly states his reasons for leaving
Turner posted an explanation for leaving Google DeepMind, highlighting a divergence in AI safety research directions, sparking discussion within the AI alignment community.
Crypto / Web3
Movement Labs files for bankruptcy, $141 million in funding lost
Despite raising $141.4 million, the Movement chain has had consistently single-digit daily fees, with only $1 in fees over the past 24 hours, and its token FDV has plunged over 99% from its ATH; the development company is now bankrupt.
Hot take: Raised $140 million, earns $1 per day—treating investors like prepaid cards.
Chip / Hardware
Hynix is said to be in talks to acquire Intel's Ohio campus
SK Hynix is negotiating with Intel to acquire its Ohio semiconductor campus, with plans to begin memory chip production within five years; if completed, the deal would alleviate Intel's foundry business challenges.
Intel has begun shipping High-NA EUV wafers.
Intel is taking a new step in advanced processes, with High-NA EUV technology set to be used in future nodes, though the timeline for mass production remains unclear.
U.S. semiconductor stocks surged 5.2%, with NVIDIA unveiling a new chip design.
AI-driven chip stocks rebounded, with the Nasdaq 100 rising 1.9%, as NVIDIA unveiled a new chip architecture and Intel continued its layoff plan.
Niantic, Flexion, and NVIDIA collaborate to bridge the Sim2Real gap for humanoid robots.
Three companies collaborate to solve the sim-to-real challenge, accelerating the real-world deployment of humanoid robots.
Technology company
The United States is pressuring South Korea to share the "excess profits" of Samsung and SK Hynix.
The U.S. government proposed to South Korea that they share the excess profits of two major chip giants, sparking discussions on geopolitical tech competition.
Apple wins iCloud CSAM scanning lawsuit
The court ruled that Apple is not liable for not scanning content in iCloud, but the judge expressed dissatisfaction, reigniting the focus on the balance between privacy and security.
Eric Goldman Blog | Hacker News
Short interest in SpaceX surges to 32%, Musk warns short sellers
SpaceX has become a prime target for short sellers, with Musk warning shorts on social media that they have "almost no chance of survival."
U.S. stocks
Stock A's Changxin Technology faced 6.6 million shares unsubscribed; the broker underwrote RMB 57.31 million.
Before ChangXin Technology's IPO, a large number of retail investors abandoned their payments, forcing brokers to underwrite the offering, with significant market divergence on the valuation of memory chips.
U.S. tech momentum stocks posted the largest single-day gain in history
AI-driven tech stocks rebounded sharply, but analysts warn that the plunge may not be over, with volatility still elevated.
New Product / New Trend
NASA's new space telescope, equipped with a deformable mirror, may be able to discover Jupiter-like planets
The deformable mirror technology of the Roman Space Telescope will enhance exoplanet detection capabilities and has the potential to discover planetary systems similar to our solar system.
Oriental Space's Gravity-1 completes its first offshore launch.
China's private rocket company Dongfang Space successfully completed a sea-based launch and plans to conduct three more batch launches this year.
Today's dark line
The AI industry is showing two contrasting signals: OpenAI and Anthropic are grappling with security and compliance issues (security incidents and copyright settlements), while China’s Kimi, despite computational disadvantages, has surged to the top in performance and has even been integrated by Microsoft in testing. This reflects a divergence in technical approaches—U.S. giants are struggling with scaling and regulation, while Chinese teams have found breakthroughs in efficiency optimization. Meanwhile, geopolitical tensions in the chip supply chain are intensifying (the U.S. is pressuring South Korea to share profits, and SK Hynix is acquiring Intel’s campus), as the AI compute war expands from the model layer to the hardware layer.