Kyle Wiggers / Techcrunch::
Some developers say that the licenses for “open” AI models like Gemma de Google and the Lama de Meta have limits that restrict commercial use without fear of legal reprisals– This week, Google published a family of open AI models, Gemma 3, which quickly praised their impressive efficiency.
Trending
- A deep technical dive into Amazon’s Trainium3 accelerator, including its server SKU specifications, silicon design, power budget, and bill of materials (SemiAnalysis)
- Calls grow for Trump and Pete Hegseth to be indicted for war crimes
- Some Reddit moderators say a wave of rejection of AI on the site is eroding its authenticity and could lead to a feedback loop of AI models training on AI content (Kat Tenbarge/Wired)
- The History of Denim and How Jeans Were Created
- A look at Phreeli, a privacy-focused phone carrier that lets users sign up with just a zip code and uses an encryption system based on “zero-knowledge proofs” (Andy Greenberg/Wired)
- A look at various quirks of AI-generated prose, mostly influenced by "overfitting" in AI models, as humans increasingly imitate AI language in writing and speaking (Sam Kriss/New York Times)
- Short seller Grizzly is shorting London-listed Trustpilot, alleging the review site pressures businesses to pay for subscriptions; Trustpilot refutes claims (Bloomberg)
- Meta delays the release of its MR glasses named “Phoenix” from H2 2026 to H1 2027 for “more wiggle room to get the details right” (Jyoti Mann/Business Insider)

