“Hey Gemini”: Is This the Voice That Could Silence Siri??

Good morning, AI enthusiasts! Google’s Gemini update preps an Assistant takeover, and Meta scores a precedent-setting copyright win. Plus: DeepSeek pauses its R2 launch, and Anthropic’s Claude now fills in the “therapist'“ role.

In today’s TLDR AI:

• “Hey Gemini”: Google’s new assistant update starts its Siri-killer countdown

• Meta beats a star-studded author lawsuit over Llama’s training data

• DeepSeek’s much-hyped R2 model stalls amid GPU shortages and CEO jitters

• Anthropic lifts the lid on how often people really treat Claude like a shrink

• Google unveils Gemma 3n—multimodal muscle that runs on just 3 GB of RAM

LATEST DEVELOPMENTS

📱 “Hey Gemini”: Google’s Siri-slayer starts its takeover

• TLDR: Starting ON July 7, Gemini will be incorporated into all Android phones and it will be able to call, text, and control apps using voice commands.

• If you keep “Gemini Apps Activity” turned off, your data isn’t fed back into model training.

• The change is a dress rehearsal for Gemini’s full replacement of Google Assistant later this year, bringing the same AI you query on the web straight to your lock-screen mic.

• Google is making an aggressive bid to own the “universal assistant” slot before Apple and OpenAI plant their flags.

⚖️ Meta dodges a multimillion-dollar copyright bullet

• TLDR: Thirteen writers—including comedian Sarah Silverman (“The Bedwetter”) and novelist Richard Kadrey (“Sandman Slim”)—claimed Meta scraped their books from the Books3 dataset (7 million pirated titles ripped from LibGen/Z-Library) to train Llama.

• Plaintiffs couldn’t show Llama reproduces “large verbatim chunks” of their books or how Llama harmed their book sales, so Judge Vince Chhabria tossed out the case.

• The judge also stressed that book scraping “is hard to imagine as fair use” if real market harm can be shown. He hints that a better-framed lawsuit could still land.

🐋 DeepSeek R2: the launch that wasn’t

• TLDR: China’s DeepSeek stole headlines in January when R1 (up to 70 B params) used RL to match OpenAI’s o1 on math & coding benchmarks at open-source prices.

• Their new update, R2, promised stronger multilingual reasoning and code-gen and could rival OpenAI’s o3 model, but CEO Liang Wenfeng recently called the current results “not good enough.”

• One big reason for this is that DeepSeek, a Chinese company, is unable to get the latest Nvidia GPU chips due to U.S. sanctions.

• Chinese clouds run R1 on Nvidia H20 chips—the only GPUs the U.S. still lets Nvidia ship to China. New export curbs have pinched even that supply, further threatening any big R2 rollout.

• DeepSeek is the most-watched open challenger to GPT-4-class models. A delay here signals that cutting-edge AI still lives and dies by GPU geopolitics.

🧡 Claude confessions: AI therapist or passing fad?

• TLDR: Is Claude your new therapist? Anthropic anonymized 4.5 million chats: only 2.9 % were “affective”—seeking coaching, counseling, or companionship. Romantic role-play was a mere 0.1 %.

• Users’ sentiment gets happier over the course of these sessions, and Claude refuses just 1 in 10 requests—mainly to block unsafe advice.

• The data punctures fears that chatbots are sweeping people into mass parasocial romances

• Anthropic and other AI giants have a grow duty of care as users lean on LLMs like Claude for real emotional support.

COMMUNITY

🎓 JOIN OUR FREE WORKSHOP 

Want to master AI & automations? Ready to turn your AI skills into a new revenue stream? Get started with our FREE workshop.

Learn how you can monetize AI and get certified as an AI specialist during our free web class. Click here to register.