Wink Pings

Six Future Technologies Receive Progress Ratings: AI Is Still Pushing Low-Resource Languages Into a Vicious Doom Spiral

MIT Technology Review assigned progress scores to six technologies including nuclear-powered spacecraft and solid-state batteries; the cognitive immortality roundtable and a batch of tech updates are rolled out as usual. The most noteworthy detail is that machine translation errors left on Greenlandic Wikipedia are being learned by the next generation of AI systems.

Future technological promises call for clear progress trackers. MIT Technology Review's Impossible special issue did exactly that: it pulled six technologies frequently billed as "world-changing", consulted experts, researchers and engineers to identify their bottlenecks, and rated their progress on a 0-to-10 scale. The six technologies are nuclear-powered spacecraft, in-pill exercise technology, cross-species communication, solid-state batteries, PFAS remediation, and zero traffic fatalities. The full ratings and detailed reports for each piece are available in this series. This approach is far more honest than a typical product launch: it does not ask "do we have a vision", but "how far are we from delivering on the promise".

On the same day, another discussion struck a more speculative tone. The Cognitive Immortality Roundtable brought topics including "gradual brain replacement" and "uploading memories to the cloud" to the table. The livestream is scheduled for November 17, hosted by science editor Mary Beth Griggs and senior biomedicine editor Antonio Regalado. The value of this kind of discussion lies in forcing people to define what "you" actually are. The answer itself is far less critical: memories can be copied, but there is no engineering interface to preserve identity continuity. If you upload a copy of your consciousness, the original "file" of you will not automatically disappear.

Here are the most noteworthy items from today's tech briefs:

  • Nvidia CEO Jensen Huang is reportedly in talks with close confidants about the possibility of returning to the company, per FT reports; Elon Musk and 165,000 signatories have called for him to be recalled; the US Treasury Secretary hopes the company will appoint an external leader. The Trump administration plans to designate Nvidia as a national AI "champion".
  • SpaceX successfully caught a rocket returning from orbit for the first time, bringing a fully reusable rocket one step closer to reality. At the same time, Ukraine accused SpaceX of lying about speed restrictions applied to Starshield services.
  • The US and its allies have agreed to protect AI supply chains, with 10 countries participating in the collaboration. Chinese AI stocks dropped amid export control concerns.
  • The recent US nationwide flight groundstop was attributed by officials to a router installation fault, laying bare the fragility of air traffic control technology.
  • The first in-body CRISPR therapy could reach patients next year, with FDA approval expected as early as March.
  • Congress is proposing to make data centers pay billions more in electricity costs, shifting grid operation expenses to tech companies.
  • Apple plans to launch its first smart glasses next year, betting heavily on Siri as the core control system.
  • Bill Gates has earned billions of dollars from climate-polluting landfills, with related investments valued at $29 billion.
  • A new AI device turns everyday household items into virtual pets that users need to care for, designed to help users build healthier daily habits.
  • Scientists had monkeys watch Minions to study brain activity, and the results challenged long-held traditional visual perception models.

Today's Quote of the day comes from Helen Toner, former member of the OpenAI board. She told The Verge: "If our hopes for the future are pinned on any one individual being trustworthy, then we’re pretty screwed." Put plainly: AI requires far stronger governance. Betting the entire future on the personal integrity of a single senior executive carries far too much risk.

The last item deserves further explanation. When Kenneth Wehr started managing the Greenlandic Wikipedia, he found that nearly every single article was written by people who did not speak the language. More and more content was copied and pasted from machine translation outputs, riddled with basic errors. AI systems, from Google Translate to ChatGPT, then learn new languages by scraping text from Wikipedia. These low-quality garbage pages enter training datasets, the next generation of models learns the embedded errors, and low-resource languages are pushed ever closer to the edge of collapse. The absurd reality here is: a person who does not speak Greenlandic pastes a machine-translated sentence into Wikipedia; the AI model treats those sentences as native language corpus; it then generates even more machine-translated sentences. Once this cycle is formed, the cost of correcting all errors will be higher than building a brand new language corpus from scratch. The full article is published on MIT Technology Review.

AI and Wikipedia push low-resource languages into a doom spiral

We can still have nice things: Adelaide Zoo welcomed its first cassowary egg in 34 years; the photos from Epson's Panoramic Photography Awards fill the frame with stunning views; Indonesia called off the world's largest deforestation project ahead of schedule, protecting up to 4 million acres of forest; Luna, a stray dog that went missing in Yellowstone for over a month, was reunited with her owner.

发布时间: 2026-11-16 21:10