Sabitlenmiş Tweet
A brief summary of recent developments in several Chinese AI companies:
When evaluating DeepSeek at the year's start, it was noted that both ZhiPu and MiniMax, seen as promising, had successfully completed their IPOs. This suggests the initial assessment was correct. Taking another look now:
1/6 ZhiPu: Previously relied on choosing the right direction and persisting to gain a seat at the large models table. But the second phase challenges of large models may exceed capabilities. Interaction design is unfamiliar, and trial-and-error costs will rise 5-10x. Lead probability drops 50-70%. Will survive the second phase and move to the third, like NLP doing rudimentary customer service during CV’s popularity peak.
2/6 MiniMax: Highly likely to experience the fastest growth in the next phase. All three core aspects well-developed. First, customer segment selection and client service experience is smooth, from pure AI company roots. Second, team’s interaction experience is stronger than others. In phase two, it can move quickly among leading companies and US giants. Third, extensive corporate management experience. Likely China's fastest-growing company ahead.
3/6 Really want to talk about Kuaishou, or Kwai AI. All three key aspects present: first, sufficient funding and complete profitability; second, pure technology with strong focus on CV/AI, transitioning from SD to AD. No external burdens, calm and grounded approach. High transition rate; third, scene/ecosystem cultivation highly cohesive. This is critical after second stage of large models, decisive for GAI. Simply: after GAI, users engage with other users in AI activities. Kuaishou completed this closed loop early. Likely the champion in next phase.
4/6 ByteDance/Douyin: Byte has two advantages and one fatal flaw, plus one general flaw. First advantage: broad user base. Second: extensive data accumulation, but modeling data quality may lag Kuaishou's. Fatal flaw: achieves only bottom-tier performance, due to attempting everything operationally and low technology transfer rate. General flaw: over-manipulating users cultivates lab rats rather than real humans, hindering high-quality AI development. Do not rule out Byte as the next Baidu.
5/6 DeepSeek: Perhaps the fastest to falter, opposite of Google's style. From life cycle view, first forked a large model, added engineering skills for short-term leaderboard gains. Second phase not completed. Beyond model training, issues with foundational frameworks, computational power adaptation, and lack of AI-driven mindset transition. AI still viewed as mere tool or product, resulting in significant fatigue. Work incomplete, chance of not passing second stage of big model.
6/6 Alibaba / Qwen: Will remain in domestic second echelon long-term. Extremely challenging to gain international reputation and influence, for three reasons: strong emphasis on engineering techniques over AI-driven mindset; heavy reliance on tools; training data and problem sets overly specialized. Highly likely to stay focused on domestic market for extended period.
#ChineseAI #DeepSeek #ZhiPu #MiniMax #Kuaishou #ByteDance #AlibabaAI
English
