Skip to content
Tech News
Topics: Today This Week This Month This Year

Google and Technion find frontier LLMs often 'know' facts they fail to recall

A new study from Google Research and Technion shows that when models like GPT-5 and Gemini-3 hallucinate, the underlying fact is usually already stored in their parameters rather than missing entirely. The researchers found these models encode 95-98% of tested facts, but often fail to surface them during generation due to retrieval, not storage, limits. They propose 'fact-level profiling' to separate whether a fact is encoded versus reliably retrievable across different phrasings.

Google redesigns Clock app's timer interface with flatter numbers and grid view

Google is rolling out a refreshed Timers tab in its Clock app as part of the September Android Drop. The update flattens the digit style, removes circular backgrounds from the keypad, adds quick presets for 1, 5, 10, and 15 minutes, and moves the start button above the number pad. Multiple active timers now display in a grid instead of a scrollable vertical list.

Fermi Explorer Mission uses AI to design $15M Alpha Centauri probe route

The nonprofit Fermi Explorer Mission, led by Philip Johnston, has spent a year designing a low-cost interstellar spacecraft meant to actually launch toward Alpha Centauri, unlike Yuri Milner's stalled Breakthrough Starshot project. Facing a persistent engineering puzzle over powering a lightweight solar craft without adding excessive weight, the team turned to artificial intelligence tools to help chart a viable trajectory. The mission, backed by private donors and budgeted at $15 million, plans to carry a small payload including art, messages, and a copy of the Golden Record.

Sonos Demos Beam Ultra Soundbar and Ace Ultra Headphones in Hands-On Preview

Sonos let journalists test its new Beam Ultra soundbar and Ace Ultra headphones in several setups, including standalone, paired with wireless Play speakers as surrounds, and streaming soundbar audio directly to the headphones. During a movie clip demo, the Beam Ultra delivered clear, well-balanced sound with surprising bass despite no subwoofer, while adding the Play speakers as rear surrounds noticeably improved immersion through Sonos's new Fabric technology, which lets devices sense their placement and surroundings.

Azure OpenAI assistant leaked SharePoint files to low-privilege users, engineer finds

Egiziago Cioffi, CEO of Microsoft partner SynSphere Italia, discovered that an Azure OpenAI email assistant he built was returning SharePoint content to a low-privilege test account that the account could not access directly in SharePoint. The assistant had passed all evaluation scores and unit tests, but comparing outputs from a high-privilege and a low-privilege account against identical queries exposed the mismatch, revealing that retrieval was happening under the indexer's permissions rather than the requester's.

Ted Nelson's decades-old Xanadu hypertext vision resurfaces amid rise of AI agents

A new essay revisits Ted Nelson's 1960s Project Xanadu, which proposed a 'docuverse' where documents kept every version and quotations linked back to their original source rather than being copied. That system never took off because the web instead adopted simple, breakable links that pushed maintenance onto users rather than the system itself. The piece argues that the rise of AI agents is now reviving interest in Nelson's original approach to tracking provenance and versioning.

Google pursues licensing deals with Hollywood studios to train AI on copyrighted films

Google is reportedly courting major studios like Disney, Warner Bros. Discovery, and Universal, offering large payments—potentially tens of millions per character and billions overall—for rights to train its AI models on their copyrighted content. According to unnamed sources cited by the Los Angeles Times, no studio has yet signed on, but Google is pushing hard to secure these licensing agreements as it races to strengthen its AI capabilities.

Bruce Schneier flags mysterious spam wave replying to his newsletter confirmation emails

Security researcher Bruce Schneier reported receiving a flood of AI-generated, one-line 'thank you' replies to the confirmation emails his Crypto-Gram newsletter sends to verify subscriptions. The messages came from dozens of throwaway Gmail addresses that never actually completed the subscription process, and none of the senders responded when he replied back.

Today's top topics: openai apple spotify anthropic meta muse amazon deepseek artificial intelligence chatgpt tiktok
Browse all topics →  ·  Today's trending topics →