THE WORLD YESTERDAY.

1. OpenAI is rolling out GPT-Live-1.1 and the lighter Quip model with full-duplex conversation, more natural accents and 13 new voices. 2. Jose runs Did I Do? alone, using home-screen widgets and reminders for daily task confidence while RevenueCat verifies $962 MRR and 323 active subscriptions. 3. Practiceme uses AI voice practice, native accents and real-time feedback for English learners while RevenueCat verifies $3,007 MRR and $4,165 in 30-day revenue. 4. Marc Jacobs recorded 130% year-on-year international online revenue growth and 77% order growth after adding Global-e localization for pricing, payment, duties and returns. 5. WHO and INTERPOL signed an agreement covering falsified medicines, organ trafficking, unusual biological events and chemical or radiological risks through health-law-enforcement cooperation. 6. An IUCN reassessment places 340 of 774 sufficiently documented reef-building coral species in threatened categories, with heat stress the leading driver for 93% of high-risk species. 7. Japan's MMX mission is scheduled to launch on October 20, collect at least 10 grams from Phobos and return the sample to Earth in 2031. 8. University of Bristol researchers routed Braille touch signals through brain organoids, with decoding rising from about 61% for one organoid to 83% for three combined.

• Notebookcheck · ChatGPT Voice: GPT-Live-1.1 brings new voices for six languages
• TrustMRR · Did I Do? verified revenue profile
• TrustMRR · Practiceme verified revenue profile
• Global-e · Marc Jacobs international ecommerce success story
• World Health Organization · INTERPOL and WHO sign agreement to strengthen global health and security protection
• Nature · Nearly half of reef-building corals face elevated relative risk of extinction
• European Space Agency · Mars moon lander set to launch on 19 October
• University of Bristol · Mini brain organoids reading Braille mark step towards biohybrid computing

OpenAI began rolling GPT-Live-1.1 and the lighter GPT-Live-1.1 Quip into ChatGPT Voice on October 15. Both retain a full-duplex design that can interpret pauses, interruptions and changes in speaking pace while deciding whether to respond or continue listening. GPT-Live-1.1 is the default for Plus and higher tiers, while Quip is the default for other users. The release adds 13 voices spanning French, German, Hindi, Japanese, Korean, Mexican Spanish and English and broadens the short phrases used while searching. The system card says production-prompt safety evaluations showed no statistically significant overall regression, but synthetic hard cases showed significant regressions in sexual content, mental health, emotional reliance and gore, so a more natural voice is not the same as risk being solved.

• Notebookcheck · ChatGPT Voice: GPT-Live-1.1 brings new voices for six languages
• OpenAI Deployment Safety Hub · GPT-Live-1.1 system card

Spanish founder Jose operates Did I Do? alone for people managing medication, household duties and recurring routines, particularly users who experience confirmation anxiety around ADHD or OCD. Minimal widgets place task status on the iPhone home screen, one tap marks an item complete and customizable reminders reduce the friction of repeatedly opening an app. TrustMRR verifies through RevenueCat $1,113 in last-30-day revenue, $962 MRR and 323 active subscriptions, updated October 15. Monetization combines a free entry point, Plus subscription and lifetime purchase. Leverage comes from frequent widget visibility, App Store distribution and the founder's existing social audience, while delivery relies mainly on a SwiftUI client and RevenueCat. The limits are small subscription scale, store fees, responsibility in health-adjacent use, notification fatigue and easy imitation.

Practiceme serves English learners from beginners to professionals with AI voice practice, native accents and immediate pronunciation and expression feedback, letting people speak repeatedly without scheduling a human tutor. TrustMRR verifies through RevenueCat $4,165 in last-30-day revenue, $3,007 MRR and 222 active subscriptions, updated October 16. Pro costs $9.99 weekly or $59.99 annually. The listed acquisition channels are App Store optimization and SEO, while delivery runs through a Swift client, Firebase and RevenueCat. The result shows that users will pay for low-friction speaking feedback, but the 5.0 rating comes from only one review, leaving renewal, voice-inference cost, accent quality, learning outcomes and weekly-subscription complaints as major limits.

US luxury brand Marc Jacobs previously supported only a small set of currencies, international cards and PayPal, leaving cross-border shoppers without local pricing, certain landed cost or familiar payment methods. With Global-e, the brand launched in three weeks with localized welcome messages, market-specific MSRP and currency rounding, checkout in more than 30 languages, prepaid duties and taxes, and local methods including Klarna in Germany and WeChat Pay and Alipay in China. Prepaid returns and local consolidation centers reduced after-sale friction. The provider's case study reports that within three months international orders rose 77% year on year and ecommerce revenue rose 130%; German conversion increased 155% and French revenue rose 484% in five months. The earning mechanism is brand demand converted through localization. These are vendor-customer case results, and profit still subtracts returns, duties, logistics, fraud management, FX and service fees.

The World Health Organization and INTERPOL announced a cooperation agreement in Geneva and Lyon on October 16, linking public-health expertise with a law-enforcement network spanning 196 member countries. The framework covers unusual biological events, suspected misuse of chemical or radioactive materials and health risks at major international gatherings that require coordinated health and security action. It also targets falsified medicines, vaccines and medical devices, counterfeit protective equipment, fraudulently labelled food and drink and illegally produced alcohol. The two bodies will cooperate against trafficking for organ removal and illegal trade in organs, tissues and cells and support implementation of the tobacco-control convention. The agreement creates a channel for cooperation rather than automatic enforcement authority; results still depend on national reporting, evidence exchange, regulatory capacity and cross-border case execution.

The IUCN coral specialist group and more than 60 experts reassessed 892 reef-building coral species, with enough data to classify 774. Of those, 340, or 44%, fall into Vulnerable, Endangered or Critically Endangered categories: 33 Critically Endangered, 251 Endangered and 56 Vulnerable. Climate-related heat stress determines the final category for 93% of high-risk species. Half of assessed Atlantic species are threatened, compared with 43% in the Indo-Pacific. The coral Red List Index fell from 0.96 before 1998 to 0.89 in 2008 and 0.74 in 2024. The authors warn that data ending in 2020 do not fully capture the 2023-2025 global bleaching event or the 2026-2027 super El Niño, so present risk may be higher; estimated declines also use coral-cover trends as a proxy and need more direct observation.

JAXA's Martian Moons eXploration mission is scheduled to launch aboard H3 Flight 10 from Tanegashima at 4:41:03 a.m. JST on October 20. The spacecraft should reach Mars in August 2027, enter a quasi-satellite orbit similar to Phobos and observe the shape, interior, composition and gravity of Phobos and Deimos. No earlier than December 2028, it plans two soft landings to collect at least 10 grams of Phobos surface material, with an Earth return in 2031. The mission will test whether Phobos is a captured asteroid or debris from a giant impact and examine how water and organics moved through the early Solar System. NASA contributes a gamma-ray and neutron spectrometer, a pneumatic sampler and deep-space communications support. Launch, landing, sampling and return remain planned milestones, not completed outcomes.

University of Bristol researchers placed brain organoids containing about 10,000 neurons and measuring under one millimeter on multielectrode arrays. They converted tactile-sensor readings from all 26 Braille letters into electrical stimulation and used a support vector machine to identify letters from the organoids' neural activity. Signals from one organoid were decoded correctly about 61% of the time, rising to 83% when three organoids were combined. The combined system was also more robust to artificial noise than directly decoding the obscured input. Published in Frontiers in Artificial Intelligence, the experiment shows that real tactile information can pass through biological neural tissue while retaining classifiable structure. It does not show learning, memory, Braille understanding or consciousness; organoid variability, standardization, long-term viability, ethics and closed-loop feedback remain central challenges.