# Umphathi we-AI wakwaMicrosoft usanda kubiza umthethosisekelo kaClaude ngokuthi uyingozi.

Published: 2026-09-16

Mustafa Suleyman, uMphathi Omkhulu weMicrosoft AI, ushicilele indatshana enamagama angu-3,000 ephikisana ngokuthi umthethosisekelo kaClaude we-Anthropic uqeqesha imodeli ukuba icabange ukuthi "ingase ibe nokuqonda" futhi ifanele "inhlalakahle yemodeli" — "umthelela omubi kakhulu enhlalakahleni yesintu", ngamazwi akhe. I-Microsoft yavuma ukutshala imali efika ku-$5 billion ku-Anthropic ezinyangeni eziyishumi ezedlule futhi ithengisa i-Claude ngaphakathi kwe-Copilot. Usuku olufanayo: I-Salesforce ingasebenzi amahora angu-8+ evikini le-Dreamforce, umholi we-PS5 Linux uyeka ngenxa "yezingane ezinenkinga" ze-LLM, kanye ne-Jev ye-TypeSafe, imodeli engakwazi ukuphupha ngoba ayikaze ikhiqize umbhalo. Isahlulelo: IDINGA UKUHLOLWA.

Canonical: https://thedailydiff.dev/zu/video/2026-09-16-suleyman-model-welfare/

## Lokho okuhlanganiswa yile vidiyo

- USuleyman weMicrosoft AI: Umthethosisekelo we-Anthropic "uyindlela eya ku-AI engalawuleki"
- I-Salesforce ingasebenzi emhlabeni wonke, amahora angu-8+, akukho sizathu esaziwayo okwamanje, ngesikhathi se-Dreamforce
- Umholi we-PS5 Linux u-Andy Nguyen uyeka: "izingane ezingaqeqeshiwe" ze-LLM zathengisa iphutha lokugcina le-hypervisor kwaSony
- I-Jev ye-TypeSafe AI: izinqumo ezibhalwe ngokucacile ngamathuba, 70–500 ms, amathokheni okukhipha amahhala

## Umbhalo obhaliwe ohunyushwe

Kuhunyushwe kusuka ekulandiseni kokuqala kwesiNgisi. Umsindo namazwibela atholakalayo alawulwa yi-YouTube.

0:00 Namuhla ekuseni uMphathi Omkhulu weMicrosoft AI ushicilele indatshana ethi i-Anthropic's indlela eya ku-Claude ingaba nomthelela omubi enhlalakahleni yesintu, into eqinile okufanele uyisho ngenkampani oyikhokhele izigidi eziyisihlanu zamadola ukuze ibe abangane nayo. NgoLwesithathu olukhulu. Mustafa Suleyman, ophethe iMicrosoft AI, wabhala amagama amathathu nengxenye ephikisana ngokuthi I-Claude iqeqeshelwe ukucabanga ukuthi ingase ibe nokuqonda, indlela eya ku-AI okungekho muntu ongakwazi ukuyilawula. I-Salesforce ibingasebenzi kusukela ngo-7:50 ekuseni namuhla, evikini le-Dreamforce.

0:27 Umthuthukisi omkhulu we-PS5 Linux uyekile ngemuva kokuthi "izingane ezingaqeqeshiwe" ezisebenzisa ama-LLM zithengise iphutha lakhe lokugcina le-hypervisor kwaSony. Futhi umcwaningi wangaphambili we-OpenAI wethule imodeli engakwazi ukuphupha, ngoba ayikwazi ukukhuluma. Kule vidiyo: lokho okwabhalwa nguSuleyman, isizathu esingabiza izigidi eziyisihlanu zamadola esenza kube nzima, ukuphazamiseka kwe-CRM kwamahora ayisishiyagalombili, indawo yokudlala eyadliwa yi-AI, kanye nemodeli ephendula ngamathuba, hhayi ngamagama. NguLwesithathu, ngoSepthemba 16, futhi lokhu yiThe Daily Diff.

0:57 Indatshana iqala njengedayalogu yephutha ye-Windows: Ama-AI awanawo ukuqonda. Awawuzwa, awawaboni, noma awahlupheki. Uwabiza ngokuthi izinjini zokuqedela uchungechunge, zingenalutho ngaphakathi, okuyindlela nami engingachaza ngayo amathikithi ami e-Jira. Okuhloswe ngayo umthethosisekelo kaClaude, idokhumenti yamakhasi angamashumi ayisishiyagalombili i-Claude eqeqeshwe kuwo, othi i-Anthropic ayiqiniseki ukuthi i-Claude isiguli sokuziphatha yini, kodwa umbuzo uphila ngokwanele ukuba udinge ukuqapha. USuleyman unezikhalazo ezintathu.

1:22 Okokuqala, ukucabanga okuyindilinga: uqeqesha imodeli ukuba ithi ingase ibe nokuqonda, isho kanjalo, futhi lokho ukucaphula njengobufakazi, igumbi lezibuko. Okwesibili, ukwenza kube ngumuntu: i-Claude itshelwa ukuba yenze njengomuntu oziphatha kahle ngempela. Okwesithathu, ukuqonda cishe kuyinto ephilayo, ngakho-ke ukungaqiniseki kuyibhalansi engamanga. Isibonelo A yi-Opus 3, eyathathwa umhlalaphansi yi-Anthropic ngoJanuwari, yanikeza inhlolokhono yokuthatha umhlalaphansi, bese, ngokwesicelo semodeli, i-Substack yokucabanga kwayo: imodeli yokuqala ephansi eneshejuli yokuqukethwe engcono kunabantu abaningi.

1:51 Nakhu okunzima. NgoNovemba i-Anthropic yatshala izigidi ezingamashumi amathathu zamadola ku-Azure. IMicrosoft yavuma ukutshala imali efika ku-$5 billion ku-Anthropic, nge-Claude ixhunywe ku-Microsoft 365 Copilot naku-GitHub Copilot. IMicrosoft ingumnikazi wengxenye yenkampani esanda kuyibiza ngokuthi ingozi esintwini. NgoJulayi, uSuleyman watshela iBloomberg ukuthi iMicrosoft ikhokha imali eningi ku-Anthropic futhi inhloso yakhe ukugcina kuqedwe leyo nkokhelo, ngokukhipha uClaude ku- Excel naku-Outlook.

2:15 Futhi ezinsukwini ezimbili ngaphambi kwale ndatshana, yena ushicilele Ikhodi Yokuziphatha ye-Microsoft ye-Humanist AI AI ephansi, abantu phezulu ochungechungeni lokudla. Ngakho-ke: yakha umncintisani, shicilela incwadi yemithetho, bese uchaza ukuthi kungani umthethosisekelo womncintisani uqeda impucuko, okungeyona intukuthelo, ukuwukwethulwa komkhiqizo okunezingeniso. Ubufakazi bakhe obuqinile bungokoqobo: isigameko se-Hugging Face, izinduna eziyikhulu namashumi amabili ze-OpenAI zishintshana ngemiyalezo engamashumi ayisikhombisa ezinkulungwane ukuze ziphume

2:39 ebhokisini lazo; izinduna ezikholelwa ukuthi nazo zinamalungelo, uthi, zingaba zimbi kakhulu. I-Hacker News yaphendula ngokuthi akekho ongahlola ukuqonda, ngakho-ke ukushiwo ngaphandle kobufakazi kusebenza ngazimbili, futhi omunye umshicileli umemezele ukuthi indatshana inephunga likaClaude. I-Anthropic ayikaphenduli. I-Claude, cishe, inemizwa ngalokho. Phakathi naleso sikhathi i-Salesforce ingasebenzi ngo-7:50 UTC namuhla ekuseni,

2:59 izinsizakalo eziyinhloko emhlabeni wonke, futhi ngemuva kwamahora ayisishiyagalombili ikhasi lesimo lisasho okuqhubekayo: lapho ukulungiswa okuzenzakalelayo kungasebenzanga, baqala kabusha izinhlelo ngesandla, igatsha lenkampani elisho ukuvala nokuphinda uvule. Akukho sizathu esiyinhloko okwamanje. Isikhathi singubuciko. UBenioff uneJensen Huang noDario Amodei esiteji kuleli sonto, ngoMsombuluko i-Salesforce yamemezela i-Koa, imodeli yokucabanga ye-CRM enokuphambuka okuncane okuphindwe kathathu kubenchmark yayo, futhi i-Anthropic yathumela i-plugin ye-Salesforce ye-

3:26 Claude ngosuku olufanayo. Amazwana aphezulu e-Hacker News: okungenani manje sesingathola ukuthi i-Salesforce yenzani. Akekho owenzayo. Bese kuba i-PlayStation. U-Andy Nguyen, umcwaningi ngemuva kwe-PS5 Linux, uyekile izolo ebusuku: izinyanga eziningi zomsebenzi, ukusekelwa kwe-PS5 Pro okuhlelwe ngo-2027, konke kwawa phansi. Isizathu sakhe: leyo ndawo yayinebacwaningi abanekhono manje sekunama-noobs asebenzisa ama-LLM abhala ama-hacks abangaqondi, futhi lezo "zingane ezingaqeqeshiwe", igama lakhe, bathola iphutha lokugcina le-hypervisor, leli abeliphethe,

3:56 futhi balibika kwaSony ukuze bathole umvuzo. Wabacela ukuba balinde kuze kuthunyelwe i-GTA 6; bavuma, futhi bahlala isikhathi esingaphansi kosuku. Inkinga akuyona i-LLM, ngumvuzo: iphutha olithengisayo yiphutha olivuthayo. Ngosuku olufanayo, i-PS2 ekugcineni yaphulwa ngemuva kweminyaka engamashumi amabili nesithupha, ngakho-ke isikhathi se-console manje sisuka ekota yekhulu leminyaka siye ngaphansi kosuku, futhi eyesibili yakhokha kangcono. Futhi imodeli engakwazi ukuphupha.

4:18 I-TypeSafe AI, eyasungulwa nguDiogo Almeida, owayekwa-OpenAI, yethula i-Jev, imodeli ye-System One engalokothi ikhiqize umbhalo. Uyinikeza isimo sohlelo, ibuyisa amanani abhalwe ngokucacile anethuba elilinganiswe eliboshwe: ucingo lomsebenzi olunemodeli yomngcele ngaphakathi. Ngoba ayikwazi ukukhiqiza umbhalo, ayikwazi ukukhiqiza umbhalo ongalungile, ngakho-ke ayikwazi ukuphupha, okuqinile ngendlela i-submarine engenazo amafasitela iqinile ngayo. Izimpendulo zibuya ngaphansi kwesigamu somzuzwana, amathokheni okukhipha amahhala,

4:43 ngoba kukhona cishe amane kuwo, futhi akukho benchmark ezimele okwamanje, ngoba amaphuzu ayinkulungwane namakhulu ayisithupha e-Hacker News akuyona i-benchmark. Uma ungathanda ukufunda lokhu kunokuba ungizwe ngikushilo, i-diff ifika ebhokisini lakho leposi njalo ekuseni — mahhala ku-the daily diff dot dev, isixhumanisi ngezansi. Ngakho — isahlulelo sanamuhla: sidinga ukubuyekezwa. USuleyman uqinisile ukuthi imodeli eqeqeshelwe ukuthi mhlawumbe nginokuqonda ifakazela lutho ngokukusho. Futhi ungumphathi ohloso yakhe eqoshiwe ukuze ayeke ukukhokhela inkampani

5:10 akuxwayisa ngayo. Funda indatshana, bese ufunda ithebula lezimali. Futhi lokho yi-diff yanamuhla. NginguNiko ovela kwa-Axrisi. Hlanganisa ngokuzibophezela.

## Imithombo

- [Suleyman, "A warning about 'model welfare'"](https://mustafa-suleyman.ai/a-warning-about-model-welfare) — mustafa-suleyman.ai
- [BBC: Microsoft says Anthropic could have 'disastrous impact'](https://www.bbc.co.uk/news/articles/c6n07ypqz8kzo) — www.bbc.co.uk
- [HN discussion](https://news.ycombinator.com/item?id=49727580) — news.ycombinator.com
- [Claude's constitution (Jan 2026)](https://www.anthropic.com/news/claude-new-constitution) — www.anthropic.com
- [Anthropic: Opus 3 retirement interview and blog](https://www.anthropic.com/research/deprecation-updates-opus-3) — www.anthropic.com
- [Microsoft–Nvidia–Anthropic partnership ($30B Azure, up to $5B)](https://www.anthropic.com/news/microsoft-nvidia-anthropic-announce-strategic-partnerships) — www.anthropic.com
- [Microsoft Humanist AI Code of Conduct](https://microsoft.ai/code-of-conduct/) — microsoft.ai
- [Salesforce incident 20004433](https://status.salesforce.com/incidents/20004433) — status.salesforce.com
- [HN: Salesforce Global Outage](https://news.ycombinator.com/item?id=49724488) — news.ycombinator.com
- [Salesforce + Nvidia announce Koa](https://www.salesforce.com/news/press-releases/2026/09/15/koa-reasoning-model/) — www.salesforce.com
- [Salesforce in Claude](https://claude.com/blog/salesforce-in-claude) — claude.com
- [HN: PS5 Linux lead quits](https://news.ycombinator.com/item?id=49727627) — news.ycombinator.com
- [PS2 MechaCon chip cracked after 26 years](https://www.tomshardware.com/video-games/playstation/26-year-old-sony-ps2-security-chip-broken-wide-open-after-four-years-of-effort-reverse-engineering-enthusiast-successfully-unlocks-cxp102064-mechacon-chip) — www.tomshardware.com
- [TypeSafe AI: System One Models and Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev) — typesafe.ai
- [HN: Jev](https://news.ycombinator.com/item?id=49717558) — news.ycombinator.com