☕️ China just erased America's AI lead

Hi there, this is your daily ☕️ Techpresso.

In today's Techpresso:

🇨🇳 China just erased America's AI lead

⚖️ Apple warns ex-staff to keep data

🤖 Google delays Gemini 3.5 Pro over coding

🎬 Netflix used AI in nearly 300 titles

🧠 Brain implant lets paralysed man eat

🇨🇳 Xi pitches China to lead global AI

Plus: 🎁 12 other news you might like, 🧰 5 tools, and 📚 4 papers.

🇨🇳 China just erased America's AI lead LINK

Moonshot AI, a Beijing company, has caught up to America's top AI systems with its new Kimi K3 model, matching leading U.S. tools weeks after their release while charging far less to use.

In tests by AI evaluator Arena, Kimi K3 beat Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol on front-end coding, and topped Anthropic's Opus 4.8 in text ranking while costing 40% less.

Moonshot plans to release Kimi as an open-weight model on July 27, letting companies and governments customize it and run it on their own systems, pressuring the pricing and valuations of U.S. labs.

⚖️ Apple warns ex-staff to keep data LINK

Apple has sent legal letters to about 40 former employees now at OpenAI, ordering them to preserve documents and messages tied to its trade secrets lawsuit accusing OpenAI of stealing confidential company information.

The letters repeat Apple's claims of "trade secret misappropriation and breach of contract" and require the ex-staff to meet with Apple's lawyers, while overriding routine deletion policies so destroying the data becomes a legal offense.

The underlying suit names former Apple hardware VP Tang Tan, now OpenAI's Chief Hardware Officer, and comes as OpenAI builds devices with former Apple designer Jony Ive; OpenAI denies knowing of any evidence supporting the claims.

🤖 Google delays Gemini 3.5 Pro over coding LINK

Google has pushed back the launch of its next flagship AI model, Gemini 3.5 Pro, by several months while it tries to fix the model's overall performance, especially its weak coding skills, according to people familiar with the matter.

The delay has reportedly worried Google engineers and researchers who fear falling behind Anthropic and OpenAI, and sources blame the company's layered internal reviews plus the effort to weave AI across Search, Maps, and YouTube.

Google updated Gemini's training data late last month to sharpen its coding, but the results disappointed, and employees say limited computing power for generating code continues to hold back their productivity on the model.

🎬 Netflix used AI in nearly 300 titles LINK

Netflix said it used artificial intelligence in roughly 300 titles this year, mostly for post-production work, and told shareholders in its second-quarter letter that it plans to keep expanding this use.

Shows including "Glory" in India, "Brasil 70: A Saga do Tri" in Brazil, and "The American Experiment" in the United States relied on generative AI to build complicated sequences faster than human visual effects artists.

The company is buying startups and setting up special studios to grow its AI work, but stressed that people must still oversee the results and that it wants to support creators rather than replace whole teams.

🧠 Brain implant lets paralysed man eat LINK

Keith Thomas, a New York man paralysed from the chest down after a 2020 swimming accident, can now feed himself and drink from a cup using a brain implant that bypasses his damaged spinal cord.

The "double neural bypass," built by Prof Chad Bouton's team at the Feinstein Institutes, uses brain electrodes to detect when Thomas wants to move his arms, while hand sensors send touch signals back to recreate the feeling of contact.

After 35 weeks of training his right arm grew 86% stronger and his left 62%, and a method called cortical mirroring restored feeling to a numb area, with these gains lasting more than two years even when switched off.

🇨🇳 Xi pitches China to lead global AI LINK

At the opening of China's biggest yearly AI event, President Xi Jinping made the case for his country to steer worldwide AI development, backing a more open approach and greater help for poorer nations in the Global South.

Xi warned against creating "new historical injustices" as AI spreads, and he pushed countries to keep the technology "secure and controllable" while building stronger awareness of the risks that come with it.

He criticized the way AI is increasingly treated as a security matter, saying nations should oppose stretching national security claims too far or putting one country's safety above others, as export controls and blacklists tighten access to the technology.

Other news you might like

• SoftBank sinks over 9% as Asia chip stocks track Wall Street AI sell-offLINK

• Verizon cuts 3,000 retail jobs as AI replaces customer service and stores move to franchisesLINK

• Twenty-nine countries sign agreement to establish global AI cooperation bodyLINK

• Oracle is one notch above junk after S&P downgrade as AI data-centre spending burns through cashLINK

• Exclusive: Visa launches new platform to provide stablecoin services to more than 200 million merchantsLINK

• Linus Torvalds to critics of AI coding in Linux: "Fork it. Or just walk away."LINK

• SpaceX suddenly aborts second Starship V3 launch after ignitionLINK

• Google continues its renaming streak by turning NotebookLM to Gemini NotebookLINK

• Anthropic and Blackstone bet the next fortune in AI is implementation, not modelsLINK

• Elon Musk secretly acquires $1B energy company to power the AI futureLINK

• Roblox launches an AI-powered game creation feature in its mobile appLINK

🧰 Trending tools

Ventorah: runs browser-based CFD wind tunnel simulations on uploaded 3D models, showing streamlines, pressure fields, and lift/drag coefficients without local installs.LINK

Yapper Leaderboard: tracks and ranks startups and users on Twitter/X by impressions, highlighting daily top performers and weekly trending accounts.LINK

Pebbles Ai: an AI-powered GTM operating system that unifies strategy, lead generation, outreach, and sales execution into one workspace, eliminating tool-switching for B2B revenue teams.LINK

Scribble Party: a browser-based lesson recorder with an infinite drawing canvas and background-removed webcam bubble, saving video locally with no signup or upload required.LINK

Kimi K3: an AI assistant that interprets academic papers, writes code, translates languages, and creates content to boost productivity for students and professionals.LINK

📚 Trending papers & reports

Chatbot conspiracy persuasion works equally well pushing people toward or away from false beliefs among 3,996 Americans tested, but simple accuracy prompts can nearly eliminate an AI's ability to spread conspiracies.LINK

Security AI agents get tested on cost, not just win rate, showing attack-simulation skill improves with more compute spent per task while defensive threat-hunting instead depends on disciplined tool use, not bigger budgets.LINK

Authorship self-awareness in AI writing tools breaks down with heavy use, as frequent users misjudge how much of the text is actually theirs, a gap that matters for learning and honest attribution.LINK

Face recognition testing can now use computer-generated faces instead of real people's photos, with two synthetic datasets, MorphFace and Vec2Face, matching real benchmarks closely enough to fully replace them, cutting privacy and legal risk from biometric data.LINK

See you tomorrow for a new dose of ☕️ Techpresso!