Author Archive: Stephen Green

DECOUPLING CUTS BOTH WAYS: Apple’s iPhone sales plummet 24% in China, while Huawei’s skyrocket 64% as Chinese consumers abandon American tech. “Since last summer, Huawei has benefitted from the release of its Mate 60 phone series after struggling to source parts for many months due to Western sanctions. The new phone features 47% Chinese-made parts, 18 percentage points higher than the previous model, and is even powered by a Chinese-manufactured chip. The Mate 60 amounts to a major upgrade for many Huawei users, but Apple’s iPhone 15 isn’t quite as enticing among its user base, according to Counterpoint’s Zhang.”

THE LATEST IN MY UNDERAPPRECIATED ALBUMS SERIES [VIP]: ‘From The Cradle.’ “If ‘Journeyman’ was Clapton’s apology for getting lost in the ’80s, ‘From The Cradle’ was the make-up sex.”

“IT’S WORSE THAN YOU KNOW”; “IT USUALLY IS”: Joe Biden: Confuzzled Warmonger “Presidentish Joe Biden has just issued another stupid and none-too-subtle threat that could extend the bloodshed in Gaza. That’s on top of another recklessly stupid decision last week that will certainly lead to more dead and wounded on both sides — and those two stupid things are just the most recent in a string of stupid mistakes leading the world to wider wars.”

WELL, YES: Sergey Brin says Google ‘definitely messed up’ Gemini image launch.

Brin, 50, spoke to entrepreneurs on Saturday at the “AGI House” in Hillsborough, California, just south of San Francisco, where developers and founders were testing Google’s Gemini model. AGI stands for artificial general intelligence and refers to a form of AI that can complete tasks to the same level, or a step above, humans.

In taking questions from the crowd, Brin discussed AI’s impact on search and how Google can maintain its leadership position in its core market as AI continues to grow. He also commented on the flawed launch last month of Google’s image generator, which the company pulled after users discovered historical inaccuracies and questionable responses.

“We definitely messed up on the image generation,” Brin said Saturday. “I think it was mostly due to just not thorough testing. It definitely, for good reasons, upset a lot of people.”

Except that isn’t the case at all:

Roughly, the “safety” architecture designed around image generation (slightly different than text) looks like this: a user makes a request for an image in the chat interface, which Gemini — once it realizes it’s being asked for a picture — sends on to a smaller LLM that exists specifically for rewriting prompts in keeping with the company’s thorough “diversity” mandates. This smaller LLM is trained with LoRA on synthetic data generated by another (third) LLM that uses Google’s full, pages-long diversity “preamble.” The second LLM then rephrases the question (say, “show me an auto mechanic” becomes “show me an Asian auto mechanic in overalls laughing, an African American female auto mechanic holding a wrench, a Native American auto mechanic with a hard hat” etc.), and sends it on to the diffusion model. The diffusion model checks to make sure the prompts don’t violate standard safety policy (things like self-harm, anything with children, images of real people), generates the images, checks the images again for violations of safety policy, and returns them to the user.

“Three entire models all kind of designed for adding diversity,” I asked one person close to the safety architecture. “It seems like that — diversity — is a huge, maybe even central part of the product. Like, in a way it is the product?”

“Yes,” he said, “we spend probably half of our engineering hours on this.”

Either Brin doesn’t know what’s going on in his own company or he can’t admit it in public. Neither should be comforting to jittery shareholders.

KRUISER’S MORNING BRIEFING: It’s Time for a Nikki Haley Wellness Check. “Nikki Haley has been hanging around thanks to the largesse of leftists who are consumed by hatred for Trump. Many of them still claim to be Republicans, but they’ve thrown in with the Democrats and are hoping that Trump will end up in jail before election day.”

XI’S GOTTA HAVE IT: Hong Kong is seeing growing internet censorship, and some American companies are using burner phones when visiting, says a top US diplomat.

In his first interview since taking up the post in 2022, US Consul General Gregory May cautioned that connectivity issues and data security concerns had prompted some American companies to use burner phones and laptops when visiting the once free-wheeling enclave.

“Hong Kong is starting to go down the slope of trying to take certain content off the internet and blocking certain websites,” he told Bloomberg News on Thursday. “It is kind of a slippery slope once you start that internet censorship. Where does that end?”

The government’s campaign to wipe Glory to Hong Kong, a song popularized during 2019 anti-government protests, from the city’s internet has highlighted operational risks for Western tech giants such as Alphabet Inc.’s Google and Meta Platforms Inc. The city’s internet service providers have also blocked access to a small number of websites, including from the UK-based rights group Hong Kong Watch. Officials have defended such actions on national security grounds.

While Hong Kong’s internet remains largely free, initial curbs have sparked concern the tight controls of mainland China, where many Western platforms are blocked and censors routinely scrub content critical of the government, could spill over into Hong Kong.

The American diplomat’s criticism comes as Hong Kong tries to bolster its position as an international business center and attract tourists to boost its flagging economy. The city’s global reputation took a hit after years of pandemic isolation drove an exodus of talent and a sweeping crackdown on political freedoms turned it into a flashpoint in fraught relations between the world’s two superpowers.

Hong Kong isn’t really a Great Powers struggle. Or at least not nearly as much as it is another example of Xi’s eagerness to stamp out political threats — real or merely perceived — no matter the economic cost.

ONE CHINA, EVENTUALLY: Beijing’s Post-Election Plan for Taiwan. “China’s public response to the election, while swift, hardly denoted a nation sulking over the defeat it was supposedly handed by Taiwanese voters. If anything, Beijing’s bombast bordered on triumphant. Seizing on the fragmented outcome, China dismissed Lai and the DPP as out of touch with “mainstream public opinion in Taiwan.” Beijing also insisted that the result did not alter the fundamental nature or trajectory of cross-Strait relations, suggesting Beijing views the opposition’s gains as validating its view that Taiwan’s population remains receptive to a reunification dialogue. Tellingly, China has not ordered new military drills encircling Taiwan since the election, likely because its months-long marathon of maneuvers already achieved their intended purpose of undermining popular support for the DPP.”