DDChen (Daniel Chen)
666 posts

DDChen (Daniel Chen)
@chen79639ddc
矩陣號自媒體經營,AI First恐怖份子,商業分析 數據分析 證卷分析狂熱者。 wealth with/by AI
taiwan,taipei Katılım Ağustos 2024
490 Takip Edilen75 Takipçiler

Google 可能是所有模型公司里,最早意识到:
不可能把最昂贵、最高推理密度、最高算力消耗的模型,以免费或低价的方式无限供应给大众。
最强模型重要,但最强模型不是大众产品。
它更像是能力边界展示,是少数重度用户、开发者、研究者和高密度生产者真正需要的东西。
而大众真正需要的,往往不是模型上限,而是:
低成本、低门槛、稳定可用。
模型从 85 分变成 92 分,大多数人感知并不强。
但从“要付费”变成“免费可用”、“打开就能用”,体感极强。
这也是我现在重新理解 Google 的地方。
Google 更像是在做一件更大的事:
把 AI 从前沿能力,压缩成大众基础设施。
比如 Gemini 3.1 Pro、Nano Banana Pro、Nano Banana 2 这些产品线,已经体现出一种分层逻辑:
· 高阶模型负责能力上限
· 轻量或 Flash 版本负责速度、成本和规模化分发。
少数人追求智能上限,
多数人追求使用门槛。
对大众的日常工作而言,某种意义上的“功能性 AGI”已经部分实现了。
当推上AI圈的推友,感觉某些版本降智了,这实际是google 在平衡模型能力、成本、速度、免费额度和用户规模 。
中文
这么早就官宣 9月29 日为 OpenAI 开发者日
难道要放大招?
无责任猜一个 GPT-6,碾压 Claude Mythos🤪
OpenAI@OpenAI
OpenAI DevDay is back. San Francisco September 29
中文

@sama Sam ,Gpt5.5 is truly masterpiece
Immense thanks to you and these staff members. I hope your models become even better. (I hope the multilingual literary capabilities of the model can rise to the level of Claude Opus—you’ve already completely surpassed it in other areas.)
English

lisan say more mean things about us you're being too nice
Lisan al Gaib@scaling01
GPT-5.5 is on par with Claude Mythos - GPT-5.5 average pass rate of 71.4% (±8.0%) - Mythos Preview 68.6% (±8.7%) - GPT-5.5 solved a task that takes a human expert ~12 hours in under 11 minutes at a cost of $1.73
English

@Its_Nova1012 I'm ready to do this...
Using claudecode will give me a heart attack one day.
English

@elonmusk Elon, we all dislike SAM, but ChatGPT 5.5 is extremely powerful, trustworthy, offers crazy usage limits, allowing OpenAI to effectively monopolize a large portion of the user experience.
Please improve Grok's capabilities as soon as possible to change this.
English


@VraserX 5.5 in Codex is amazing, Meanwhile, Opus 4.7 kept causing my blood pressure to rise, so I seriously considered switching.
Opus 4.7 is an extremely talented guy, but always acts without thinking things through, and 80% of the time is spent fixing the problems he causes
English

@marktenenholtz What tasks?
Seriously, I am really interested in this one
English

Anyone who says this doesn’t realize Gemini 3.1 Pro is giga SOTA for certain extremely useful tasks.
3.0 Pro Preview is possibly better at the same tasks than GPT-5.5/Opus 4.7.
Can Vardar@icanvardar
google should just give up on ai at this point
English

@dontbesilent 文案得opus
Gpt5.5是剛學會講話了 哈哈😂
但5.5對計劃 決策 取捨的判斷,我體感比Opus更深(但他的表達不一定更有說服力)
中文

我道歉我给出一个错误建议:这一段时间如此开多窗口实践下来,我觉得人都要被耗散光了。人类还没进化到能适应AI的效率,或者至少我这一代尚且不能。我们还需要更妥当的和AI合作的方式。
Michael Anti@mranti
AI时代新的工作方式:可裂屏的终端开四个窗口,每个窗口claude运行一个项目。这样的效率最高、人的注意力经过适应也能hold住。就是每过一个小时,得休息下、累。
中文

@BoyuanChen0 The refusal to generate is too common!!!
English

Claude 降智的原因终于找到了
是SDK 的 harness 出了问题
所有基于 claude SDK 的 agent 受到了影响
官方已经发了更新解决
这查了得一个月吧…
真不容易,关键时刻还得靠人类
ClaudeDevs@ClaudeDevs
The issues stemmed from Claude Code and the Agent SDK harness, which also impacted Cowork since it runs on the SDK. The models themselves didn't regress, and the Claude API was not affected.
中文















