OpenAI waxay ku bilaabaysaa ChatGPT iyo Codex sawirka qoraalka ee Midowga Yurub
OpenAI waxay sheegtay inay bilaabaysaa sawirka qoraalka (text watermarking) ee nidaamkeeda ChatGPT iyo Codex, laakiin kaliya dadka ku nool Midowga Yurub ayaa hadda heli doona xaqiiqdaas. Tani waa tallaabo loo qaaday sharciga AI-ka ee Yurub, iyadoo OpenAI ay sheegtay inay awooddaas ku jirto inay ka fiican tahay kuwa kale.
Sawirka Qoraalka iyo Isku-dhafka Awoodda
OpenAI waxay sheegtay inay isticmaashay tillaabo loo yaqaan 'textGrain' oo aan la arkin laakiin mashruuc-ku qabsoomay, taas oo ay sheegeen inay 'isku dhacday ama ka fiican' qaababka kale sida SynthID ee Google DeepMind. Tani waxay muujinaysaa inay awooddaas ku jirto inay ka fiican tahay kuwa kale.
Si kastaba ha ahaatee, shirkaddu waxay sheegtay in sawirka qoraalku uusan xaqiijin karin saxnaanta qoraalka, ama inuu xaqiijin karo in qoraalku yahay mid qof insaami ah uu qoray. Waa tallaabo loo qaaday sharciga AI-ka ee Yurub, iyadoo OpenAI ay sheegtay inay awooddaas ku jirto inay ka fiican tahay kuwa kale.
Bilaashka iyo Xaqiijinta
Bilowga todobaadka soo socda, OpenAI waxay soo saari doontaa sawirka qoraalka dadka la xiriira ChatGPT iyo Codex ee ku nool Midowga Yurub, iyadoo aan la beddelin nidaamka caalamiga ah. Dadka isticmaala API-ka ee adduunka oo dhan waxay hadda yeelan karaan doorasho inay isticmaalaan sawirka qoraalka.
Shirkaddu waxay sheegtay inay diyaar u tahay inay bixiso aalad lagu xaqiijin karo sawirka qoraalka, laakiin kaliya xirfadlayaasha iyo ururada aqoontu waxay heli karaan xaqiiqdaas. Aaladdu waxay sheegi doontaa inay aragto sawirka, laakiin ma aysan xaqiijin qofka qoray ama qoraalka.
OpenAI is rolling out an invisible, machine-readable watermark to text outputs from ChatGPT and Codex, but the feature is launching exclusively for users within the European Union. The company states it will not make text watermarking a global default at this time.
Performance and Regulatory Compliance
OpenAI claims its new textGrain watermarking technology matched or exceeded other approaches, such as Google DeepMind's SynthID, which also serves as the basis for watermarking announced by Anthropic in August. Both companies implemented these measures to meet the requirements of the EU's AI Act. Benchmarks included in the announcement showed similar performance between watermarked and unwatermarked text.
Despite the rollout, OpenAI emphasizes that the tool does not guarantee reliable detection. The company clarifies that text watermarks do not verify accuracy, determine ownership, measure human contribution, or prove human authorship.
Limited Availability and API Access
Over the coming weeks, text watermarking will be introduced to eligible ChatGPT and Codex users across all plans in the EU only. While the public rollout is regional, API customers around the world can opt in to watermarked text outputs for select models starting today. Approved researchers and expert organizations may also apply for access beginning today.
The detection tool will report whether it identifies an OpenAI watermark without revealing user identities, prompts, or conversations. However, due to risks of missed watermarks and false positives, the tool is not publicly available at launch.
Ilaha iyo xuquuqda sawirka
Sawir: The Verge Xigasho
