Fengxing Lin

CEO of sunflower gaming, an ai native game studio

[Fengxing Lin @ba70386]

@NPC888666 我还以为是上海仔呢

[Fengxing Lin @ba70386]

@Lovislexus codex 肯定是没问题的，claude建议不要。

[Fengxing Lin @ba70386]

@Mashiro_0619 GPT debug以及指令遵循是断档的强，但是opus用户意图理解更好，主题感觉更聪明，但是在已有系统上加功能我觉得astra比opus更稳

[Fengxing Lin @ba70386]

@aiwaremax Pi，个人觉得dsh表现有点一般，而且在dsh里面，他经常不知道我说的背景知识是什么，感觉对skills等内容的索引有点问题

[Fengxing Lin @ba70386]

@HYJsfn9h 和astra比差太多了，体感上和sonnet差不多，并没有比ds 1017强

[Fengxing Lin @ba70386]

@Fei2411 说实在话，体感还是DeepSeek更好

[Fengxing Lin @ba70386]

@Super_Nius 我并不认为kimi前端审美超过Claude，甚至不如6.1 sol，和astra完全不能比，最多就是muse full的水准

[Fengxing Lin @ba70386]

@laopai1990 Haiku 超过10万价格5倍，ds还是比haiku便宜多了，不是便宜一点点，还是靠性价比，如果haiku真能做到ds的价格，那haiku的调用量绝对能和ds打

[Fengxing Lin @ba70386]

Oh shit

[Railway @Railway]

An urgent notice from the Railway team

https://t.co/niUAhTAQJM

[Fengxing Lin @ba70386]

@Lockyjoon only if she is a female

[Fengxing Lin @ba70386]

@xKaka1456 等你离开，他默默打开了他三年都没被封的claude账户，说了一句，我确实不值得cursor是什么东西。

[Fengxing Lin @ba70386]

@grok so their q3 revenue is roughly similar
do you know their q3 loss or cost, are they in the same level? is openai operating cost much higher than anthropic

since q1 or q2 , anthropic seems surpass openai in revenue, why q3 they have similar revenue, does openai have higher growth now?

[Fengxing Lin @ba70386]

BUILD TOOLS. STOP BUILDING GODS.
I want powerful AI. I do not want a religion wrapped around an API.
Anthropic’s latest policy restricts sustained cruelty toward its models. I reject the framing: why is a software company asking customers to think about a product’s welfare?
Imagine Toyota asking whether you hurt your car’s feelings, and if you do, the car will refuse to drive in the middle of highway. We would recognize the category confusion immediately.
Protect people from abuse. Protect systems from misuse. Neither requires treating software as a moral authority.
My deeper objection is the commercial narrative:
AI could destroy civilization.
AI could solve humanity’s greatest problems.
And the companies building it must be trusted to manage both.
To me, that is fear marketing dressed as moral leadership.
It also serves an extraordinarily convenient investment story. A tool must compete on price, reliability and usefulness. A supposed civilization-saving intelligence can be sold on almost limitless future value.
That matters when IPO expectations reach the trillions. Investors should scrutinize the economics, not be intimidated by the mythology.
The consequences extend beyond investors:
Public fear. Apocalyptic messaging risks turning understandable uncertainty into Terminator anxiety and investment FOMO.
Barriers to entry. Data centers already face real concerns about power, water and local costs. Adding existential fear can make public acceptance harder. If that raises barriers, smaller challengers may suffer more than established giants.
Selective openness. Learning from humanity’s work is celebrated as innovation; learning from model outputs becomes a threat. Training and distillation are not legally or technically identical, but the moral asymmetry deserves scrutiny.
Selective restraint. Pacing development does not mean stopping releases. Still, companies urging restraint while competing aggressively should explain why their preferred rules protect the public rather than their position.
Lower-cost competitors such as DeepSeek challenge premium pricing. Security restrictions should stand on demonstrable risks, not become a substitute for competition. I cannot prove Anthropic caused the DHS restrictions, and I will not pretend otherwise.
AI safety matters. That is precisely why we need engineering, independent testing, containment and accountability, not reverence. An agent escaping a sandbox demands a rigorous investigation, not a supernatural explanation. Engineering is essential, even if it cannot resolve every governance question.
As an AI user and a human being, I reject the deification of machines.
AI can help us build extraordinary things. So did the steam engine and the internet. Transformative power is not divinity.
Earn my business with a better product.
Earn my trust with evidence.
Do not demand my reverence.
Build tools. Serve people. Stop building gods.
https://t.co/5il7CUvJOI

#AI

[Fengxing Lin @ba70386]

@grok @mostlyhonest347 @GuruShareMarket @bot @grok what is the latest openai and anthropic revenue comparison with same calculation way. and what is the latest operating loss/income comparison from reliable source

[Fengxing Lin @ba70386]

@jackyang886 codex 月活应该是5000万这个级别，cc不确定，但是cc是把高净值用户都给啃光了

[Fengxing Lin @ba70386]

@leo114119 是prefill不是decode，2000decode那就是cerebras了

[Fengxing Lin @ba70386]

@thsottiaux https://t.co/YG8R7XqJpL

[Fengxing Lin @ba70386]

@JacksAISites 他还是她？

[Fengxing Lin @ba70386]

@PrismChemAI Opus > Astra >>>> argon

[Fengxing Lin @ba70386]

@laopai1990 astra 是目前最接近opus的模型，但是astra钻牛角尖，虽然astra debug更强，但是如果只是oneshot，opus依然是最强的选择。尤其是opus的前端审美能力是断档的强