GPT-6 Waa Insane Daawo Adiga Yaabi Doone

Videogan waxaan ku falanqaynayaa GPT-6 ee ay OpenAI ku dhawaaqday kadib hype aad u weyn. Si ka duwan muuqaalladaydii hore ee models-ka cusub, markan si deep ah ayaan u galayaa waxyaalaha model lagu qiimeeyo: pre-training-ka oo ah marka raw data-da la soo ururinayo, reinforcement learning-ka oo lagu qeexo behavior-ka model-ka, iyo alignment-ka. Generation cusub waxaa loogu yeeray sababtoo ah training-ka qaab cusub ayaa laga bilaabay, ee aan ahayn improvement lagu sameeyay model-kii hore.

Kadib waxaan dul marayaa benchmark-yada: ARC-AGI oo bishii Maarso 6% lagu joogay hadda ku dhow yahay boqolkiiba boqol, FrontierMath, iyo CyberGym-ka cybersecurity-ga oo aan hal exploit ka dhicin. Waxaan isbarbardhigayaa qiimaha iyo token efficiency-ga: kharashka API-ga ee shaqo kasta iyo sida models-ka Anthropic ay uga qaalisan yihiin. Waxaan tusayaa computer use-ka iyo xawaaraha Codex, shaqooyinka professional-ka ah sida presentation-ka iyo dukumentiyada, iyo game development-ka iyo assets-ka 3D ee lagu sameeyay Unreal Engine iyo Blender.

Ugu dambayn waxaan idin geeynayaa Artificial Analysis oo ah qiimayn madax-bannaan, halkaas oo models-ka lagu kala saaro caqliga, xawaaraha iyo cost per task-ga. Opus iyo Fable ayaa caqli ahaan weli hor socda, Gemini-na waa kan ugu xawaaraha badan laakiin ujeeddada shaqada si fiican uma raaco. Hadda model-ka waxaa la siiyay shirkado kooban, maalmaha soo socdana wuxuu u soo bixi doonaa Plus, Pro, Business, Enterprise, API-ga, Azure iyo Bedrock. Aniga weli access ma haysto, markuu ii soo gaadho waan tijaabin doonaa. Qaybta comments-ka iiga sheega waxa aad ka qabtaan model-kan.

A 49-minute Somali-language (Af-Soomaali) deep dive from Dugsiiye into GPT-6, the model OpenAI announced after a huge hype campaign. Unlike the usual new-model round-up, this one explains how models are actually evaluated: pre-training, reinforcement learning and alignment, and why a fresh training run makes this a new generation. It then walks the benchmarks, ARC-AGI, FrontierMath and the CyberGym cybersecurity set, and compares API cost and token efficiency against other frontier models. There are long looks at computer use, Codex speed, professional work such as presentations, spreadsheets and document handling, and game development with 3D assets built in Unreal Engine and Blender. It closes with the independent Artificial Analysis rankings for intelligence, speed and cost per task, plus the rollout plan. Free on YouTube.