Since many of those Chinese models rely heavily on distilling Opus 5.5, can they ever truly reach the exact same frontier level? Even if everyone uses distillation, labs that actually own the frontier model can continuously refine their mid-tier and lightweight models without paying external inference margins. In the end, wouldn't the frontier labs end up dominating the mid-tier and small model markets on both price and performance?
Especially in multi-agent orchestration, where specialized sub-agents divide up the work and the frontier model acts like the CEO: understanding user intent, delegating tasks, reviewing intermediate outputs, ordering revisions, and synthesizing the final result. As we move deeper into the multi-agent era, the scarcity value of top-tier frontier intelligence might actually become even more critical.
Curious to hear what everyone thinks!
❄️❄️❄️❄️❄️❄️❄️
$UBER also has deep pockets and massive data, so it can match any promotion with instant Uber Cash or credits right inside the app people already use. Without a reason to download and manage another app, the existing habit will hold strong.
By the way, I just sent you a DM! Nothing weird at all, I just wanted to ask for your advice on growing my X account. If you have some free time, I would really appreciate a reply, and no worries at all if you would rather pass!
Since many of those Chinese models rely heavily on distilling Opus 5.5, can they ever truly reach the exact same frontier level? Even if everyone uses distillation, labs that actually own the frontier model can continuously refine their mid-tier and lightweight models without paying external inference margins. In the end, wouldn't the frontier labs end up dominating the mid-tier and small model markets on both price and performance?
Especially in multi-agent orchestration, where specialized sub-agents divide up the work and the frontier model acts like the CEO: understanding user intent, delegating tasks, reviewing intermediate outputs, ordering revisions, and synthesizing the final result. As we move deeper into the multi-agent era, the scarcity value of top-tier frontier intelligence might actually become even more critical.
Curious to hear what everyone thinks!
❄️❄️❄️❄️❄️❄️❄️
Something I think people fail to understand is that as soon as there is 1 cheap / smaller Chinese open weights model roughly on par with Opus 5.5, it is game over for Anthropic. 5.5 is, for better or worse, good enough for almost every task. 🧵
@Franny8 I don't know about that, haha. Honestly, none of the hyperscalers look like they have any room to pivot back to major shareholder returns anytime soon.😅
@xAxiom83x@AlexFinn I actually canceled my Gemini subscription a few months ago, so maybe I need to re-subscribe and give it another shot! What kind of tasks are you mainly using it for that gave you that impression?
기기 장악력만큼은 의심의 여지가 전혀 없습니다. 저도 아이폰만큼 완성도 높은 제품을 본 적이 없습니다.
하지만 외부 모델을 가져다 쓰는 구조에서 협상력은 상황에 따라 크게 달라질 �� 있습니다. 과거 부품사나 일반 앱 개발사들과 달리, 최전선 AI 모델은 극소수 기업만이 막대한 자본과 연산력으로 만들어낼 수 있어 대체하기가 쉽지 않기 때문입니다.
무엇보다 애플이 AI 모델의 로드맵, 성능, 최적화에 대한 직접적인 통제권을 갖지 못한다는 점은 가장 큰 리스크이지 않을까 싶습니다. 애플이 강력한 해자를 유지해온 것도 하드웨어와 소프트웨어의 통합과 거기서 오는 사용자 경험인데 핵심 기술인 AI 모델을 기존 생태계와 잘 통합할 수 있을지 아직은 의심스럽습니다.
현재 Chatgpt와 Siri가 통합되었다고는 아직도 너무 멍청한 것 같거든요. 반면에, Google의 AI 모델이나 Meta Muse가 치고나오는 것을 보면 안타까움을 느낍니다. 물론, 그들이 2,3,5,10 년 후에도 높은 ROI를 유지할 수 있을지는 별개지만요.
참고로 저는 애플 주주입니다.
@vojtsei@BrettKessler__ Agreed. Even if $UBER's take rate compresses slightly with the rise of AVs, their growth will still continue as long as the overall rideshare market accelerates
@BrettKessler__ The more I study the rideshare market, the more I realize the true bottleneck is supply rather than demand. In that sense, $UBER actually has a big advantage in the AV race, because onboarding one human driver is way faster than rolling a new car off a factory line.
@100janovicpharm really interesting how massive cultural and entertainment events act as real growth catalysts for $UBER. Seeing Uber Eats outpace $DASH in order growth that night is a great sign too.