Sida wax-soo-saarrada 'goblin' ugu faafeen moodooyinka AI-ga: jadwalka, sababaha asalka ah, iyo xalalka ka dambeeya dabeecadaha shakhsiyadeed ee yaabka leh ee dhaqanka GPT-5.

Baro sida Model Spec ee OpenAI uu ugu adeegayo qaab-dhismeed dadweyne oo loogu talagalay hab-dhaqanka nooca, isaga oo isu dheellitiraya badbaado, xorriyadda isticmaalaha, iyo isla-xisaabtan marka nidaamyada AI ay horumarayaan.

Sida OpenAI u adeegsato la socodka silsilada fikirka si ay u daraaseyso iswaafaq-la'aanta wakiillada gudaha ee koodh qorista—iyadoo falanqaynaysa adeegsiyada dunida dhabta ah si loo ogaado khataraha loona xoojiyo ilaalooyinka badbaadada AI.

IH-Challenge waxay noocyada u tababartaa inay mudnaan siiyaan tilmaamaha la aamini karo, iyadoo hagaajinaysa kala-sarraynta tilmaamaha, hagidda badbaadada, iyo iska-caabbinta weerarrada weydiinta.

OpenAI waxay soo bandhigtay CoT-Control waxayna ogaatay in noocyada caqliyeynta ay ku dhibtoodaan xakamaynta silsiladaha fikirkooda, taas oo xoojinaysa la-socodka inuu yahay ilaalin amniga AI ah.

Preprint cusub ayaa ku ballaarinaya amplituudhyada hal-minus gravitons, iyadoo GPT-5.2 Pro ka caawisay soo saarista iyo xaqiijinta graviton tree amplitudes aan eber ahayn.