Fable 1.5 Lama Isticmaali Karo
Sharaxaad
Videogan waxaan ku falanqaynayaa models-ka cusub ee ay Anthropic soo saartay, Fable iyo Mythos oo halmar la soo bandhigay, Mythos-ka oo ah kan ugu weyn. Waxa ugu muhiimsan ee ka duwan kii hore waa guardrails-ka: check points ayaa hor yaal labada dhinacba, su'aasha aad gelinayso iyo jawaabta uu soo saarayo. Kadib waxaan dul marayaa benchmark-yada, gaar ahaan qaybta scientific-ga oo ku dhow labo jibaar marka la barbardhigo model-kii hore, agentic coding-ka oo Mythos-ku ugu sarreeyo, knowledge work-ga, computer use-ka iyo reasoning-ka.
Waxaan idinla wadaagayaa tijaabadaydii. Shaqo research ah ayaan model-ka u diray anigoo dhowr sub-agents kiciyay, 47 daqiiqo kadib limit-kii shanta saac ee 20x-ka ayaan 100% ku gaaray, waxaana la ii sheegay inaan afar saac sugo. Halkaas ayuu magaca videogu ka yimid, waa qaali, si fududna looma isticmaali karo. Waxaan sidoo kale sharxayaa qiimaha, million token oo input ah waa $10, million token oo output ahna waa $50, iyo sida tokens-ka loo xisaabiyo. Tokenizer-ka ayaan idinku tusayaa, space-yada waa la tirinayaa, celcelis ahaanna saddexdii ama afartii xaraf hal token ayay u dhigantaa.
Ugu dambayn waxaan ka hadlayaa safeguards-ka la hagaajiyay, maadaama model-ku horay u wareejin jiray su'aalo aan waxba khalad ku jirin. Waxaan idin tusayaa habka aan hadda ugu shaqaysto: model-ka weyn wuxuu ii yahay orchestrator oo shaqada maamula, sub-agents-ka ku shaqaynaya models-ka jaban ayaa qabta shaqada intiisa badan. Sidaas ayaad lacagta ku dhaqaali kartaa, gaar ahaan haddaad qorshaha $20-ka isticmaasho. Qaybta comments-ka iiga sheega waxa aad ka qabtaan model-kan iyo limit-ka.
English summary
A 13-minute Somali-language (Af-Soomaali) breakdown from Dugsiiye of the new Anthropic model, released alongside Mythos. It starts with the guardrails that now sit on both ends of a request, checking the question going in and the answer coming out, then walks the benchmarks for science, agentic coding, knowledge work, computer use and reasoning. The heart of the video is a real trial run: a research job with several sub-agents burned the full five-hour limit in 47 minutes, which is where the title comes from. From there it explains the pricing, $10 per million input tokens and $50 per million output, and shows the tokenizer so you can see how text becomes tokens, spaces included. It closes with the improved safeguards and the orchestrator pattern that keeps the bill down. Free on YouTube.
Waxaad ku baranaysaa
- Waxa cusub ee models-ka Anthropic iyo Mythos-ka la socda
- Guardrails-ka iyo check points-ka su'aasha iyo jawaabta
- Benchmark-yada: scientific, agentic coding, knowledge work iyo computer use
- Limit-ka shanta saac iyo kan isbuucleh ee 20x-ka
- Qiimaha: million token oo input ah iyo million token oo output ah
- Sida tokenizer-ku u shaqeeyo iyo xisaabinta tokens-ka
- Isticmaalka model-ka weyn orchestrator ahaan si kharashka loo yareeyo
Cutubyada
- 0:00Hordhaca model-ka cusub ee Anthropic Af Soomaali
- 1:34Benchmark-yada iyo awoodda Fable Af Soomaali
- 3:31Tijaabada Fable iyo dhibaatada limit-ka Af Soomaali
- 6:16Fahamka Tokenizer iyo xisaabinta tokens Af Soomaali
- 9:25Safeguards-ka iyo xakamaynta model-ka Af Soomaali
- 11:15Isticmaalka Fable Orchestrator ahaan Af Soomaali
- 12:35Talooyinka isticmaalka iyo gabagabada Af Soomaali
Su'aalo
- Model-kan hadda ma isticmaali karnaa?
- Xilliga videoga la duubay wuu furan yahay haddaad subscriber tahay. Laakiin qaali waaye, limit-kana si degdeg ah ayaad u gaari kartaa, sida igu dhacday markaan tijaabinayay.
- Sidee tokens-ka loo xisaabiyaa?
- Celcelis ahaan saddexdii ama afartii xaraf hal token ayaa u dhiganta, space-yadana waa la tirinayaa. Million in-ka waa waxa aad adigu qorto, million out-kana waa waxa model-ku soo saaro.
- Sidee lacagta loo dhaqaalayn karaa?
- Model-ka weyn orchestrator ahaan u isticmaal, sub-agents-ka ku shaqaynaya models-ka jaban ayaad shaqada intiisa badan u dhiibtaa. Qaladkii aan sameeyay wuxuu ahaa inaan sub-agents-ka oo dhan model-ka weyn siiyay.
