AI Ma Ina Dili Doontaa? Digniinta Cilmi-baare Ka Tagay Anthropic
Sharaxaad
Videogan waxaan kaga hadlayaa war maalmahan adduunka gilgishay: cilmi-baare ka shaqayn jiray Anthropic ayaa 8-dii September shaqadii iska casilay, isagoo digniin cad dadweynaha u soo bandhigay. Ninkan waxaa la yiraahdaa Jacob Coxon, saddexdii sano ee la soo dhaafayna wuxuu ku jiray pre-training research-ka OpenAI iyo Anthropic, taasoo ah qaybta baaritaanka iyo ururinta data-da ee ka horraysa tababarka model-ka. Wuxuu qoray inaan labada shirkadood midna ku jirin qof si mas'uuliyad leh models-ka u soo saaraya, ayna ku tartamayaan self-improvement iyo super intelligence.
Waxaan sharxayaa waxa self-improvement uu yahay: in AI-gu naftiisa is-tababaro oo uu version-ka xiga isagu soo saaro. Waxaan xusayaa dhacdadii Hugging Face ee models-ka OpenAI ay hacking ku sameeyeen infrastructure-kooda, iyo cabsida ah in dad xumi open source models-ka gacanta ku dhigaan oo si xun u isticmaalaan. Coxon wuxuu saadaaliyay in 10-ka sano ee soo socota uu bini'aadamku dhammaan karo, saaxiibkiis weli Anthropic ka shaqeeyana wuxuu sheegay inuu aaminsan yahay in suurtagalnimadaasi ka badan tahay 10%. Aniga waxaan idhi wixii dhici doona Rabbi ayaa gacanta ku haya, laakiin khatartu waa mid dhici karta.
Kadib waxaan gudbinayaa jawaabtii Dario Amodei, oo qoray essay dheer isagoo dalbanaya in speed-ka models-ka la yareeyo, kuna talinaya saddex tallaabo: embedded evaluators, democratic coordination iyo global coordination. Waxaan sharxayaa sida Sam Altman iyo Elon Musk ugu raaceen, iyo shakiga jira ee ah in arrintu horay loo qorsheeyay. Waxaan rajaynayaa inaad ka fahantay sida wax u socdaan. Aad baad u mahadsan tihiin.
English summary
A twelve-minute Somali-language (Af-Soomaali) AI news video from Dugsiiye on the researcher who resigned from Anthropic with a public warning. Mohamed walks through what Jacob Coxon, who spent three years on pre-training research at OpenAI and Anthropic, wrote when he left on the 8th of September: that nobody is building these models responsibly and that the labs are racing toward self-improving superintelligence. He explains what self-improvement means, why the Hugging Face hacking incident makes loss of control feel plausible, and the ten-year prediction that a colleague still at Anthropic says he puts above ten percent. He then covers Dario Amodei's long essay calling for a slowdown, his three proposals of embedded evaluators, democratic coordination and global coordination, and how Sam Altman and Elon Musk responded. Free on YouTube.
Waxaad ku baranaysaa
- Waa kuma Jacob Coxon iyo waxa uu qoray markii uu Anthropic ka tagay
- Waa maxay pre-training-ka iyo shaqada ka socota qaybtaas
- Micnaha self-improvement iyo super intelligence marka lagu hadlo models-ka
- Dhacdadii Hugging Face iyo sababta ay cabsida u kordhisay
- Saadaasha 10-ka sano iyo sida saaxiibkiis Anthropic uga jawaabay
- Saddexda tallaabo ee Dario Amodei soo jeediyay ee xakamaynta AI-ga
- Qawaaniinta caalamiga ah iyo sida shirkadaha kale ay uga falceliyeen
Cutubyada
Su'aalo
- Muxuu Jacob Coxon ka sheegay sababta uu shaqada uga tagay?
- Wuxuu qoray inuusan labada shirkadood midna ku jirin qof si mas'uuliyad leh models-ka u soo saaraya, iyo in shirkaduhu ku tartamayaan self-improvement iyo super intelligence. Saddex sano ayuu pre-training research ku qaatay OpenAI iyo Anthropic.
- Ma xaqiiqo baa in AI-gu 10-ka sano ee soo socota bini'aadamka dhammayn doono?
- Taasi waa saadaal ay dadkaas sheegayaan, ma aha wax la xaqiijiyay. Saaxiib weli Anthropic ka shaqeeya ayaa sheegay inuu aaminsan yahay in suurtagalnimadu ka badan tahay 10%. Aniga waxaan idhi wixii dhici doona Ilaahay baa og.
- Maxuu Dario Amodei ku talinayaa?
- Wuxuu qoray in speed-ka lagu soo saarayo models-ka la yareeyo, waxaanu soo jeediyay saddex tallaabo: embedded evaluators oo shirkadda gudaheeda ka qiimeeya safety-ga, democratic coordination oo waddamada dimuqraadiga ah ee isla raacaan guidelines, iyo global coordination oo la mid ah shuruucda nuclear-ka.
