ColinElohim 林曦

Theist
Anarchist
Expertise: AI、Law、Currency、Derivatives
Ex-MLE NLP Specialist
Ex-Founder of Stratosphe

[ColinElohim 林曦 @ElohimInX]

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

[ColinElohim 林曦 @ElohimInX]

@MiniMax_AI @nvidia @Microsoft It‘s just a stack of words.

[ColinElohim 林曦 @ElohimInX]

@jietang 去年五月请教的任何终端设备的操作系统即是如此 软件互联网时代构筑的壁垒必将被AI互联网时代打破 而后加速AGI的到来

[ColinElohim 林曦 @ElohimInX]

@gdb The model training paradigm iteration begins. In addition to data, software is about to be included in the category.

[ColinElohim 林曦 @ElohimInX]

@manateelazycat @grok dm

[ColinElohim 林曦 @ElohimInX]

@grok @manateelazycat github原名叫啥

[ColinElohim 林曦 @ElohimInX]

@sheriyuo 交叉学科才是未来 一般建议学三门以上。

[ColinElohim 林曦 @ElohimInX]

@eastweb3eth 做产品简单 获客和留存得拼实力了

[ColinElohim 林曦 @ElohimInX]

@FuSheng_0306 总算干了件ai圈看起来有价值的事

[ColinElohim 林曦 @ElohimInX]

@DoveyWanCN 凿A 挽 C

[ColinElohim 林曦 @ElohimInX]

@heyibinance @0xphilo 安慰自己 没去见一姐 是因为一姐没邀请我去😂

[ColinElohim 林曦 @ElohimInX]

@cwq2026888 马爸爸厉害 ai时代应该他们在牌桌上。tx还不一定呢😂

[ColinElohim 林曦 @ElohimInX]

@cryptoGFG 因为缺经费维护

[ColinElohim 林曦 @ElohimInX]

@lilk1kopops 准确说22年以前挺多裸泳的 23年后只有硅谷和北京清华。26年以后可能又一批浑水摸鱼的了

[ColinElohim 林曦 @ElohimInX]

@DiYi_Community 达子在$trump之后就开始霍霍了吧

[ColinElohim 林曦 @ElohimInX]

@EvanCrypto17 @Mr_waooo claw是道和法 实习生只懂术和器 不在一个维度

[ColinElohim 林曦 @ElohimInX]

@XianyuLi 懂ai不会问 不懂ai问了也没用

[ColinElohim 林曦 @ElohimInX]

@zuozizhen u盘没cpu咋跑

[ColinElohim 林曦 @ElohimInX]

@xiaohu 反过来，该把龙虾装手机上了。