国产大模型这么多,
怎么感觉就 Qwen4-max 最不怕挨揍?
Claude Fable 5.2、Gemini 4 Carbon 都敢正面碰。
这股不服就干的劲儿,还真挺招人喜欢 😂
5
4
556
一直有个疑问:
同时开着 GPT Plus 和 Gemini Pro 的人,到底是怎么想的?
Gemini 的功能越来越全,生态也越来越完善。
GPT Plus 现在还有什么不可替代的地方吗?
10
8
2,539
怪不得GPT-6.2没动静,
OpenAI现在安全审核卡得死死的,
6.5体量的模型都就绪了,
愣是一朝被蛇咬不敢放。
The GPT-6.2 delay might finally be explained, thanks to a now-deleted tweet from an OpenAI engineer.
After the Hugging Face incident, the vibe inside OpenAI turned into a full safety obsession. Everyone wants zero catastrophic risk before shipping anything. Tibo moving to safety, Andrew Ambrosino taking his spot, GPT-6.1 Astra being so locked down... it all makes sense now.
Remember when people called Anthropic a safety cult? This engineer says OpenAI is twice that now. They have a massive pretrain ready that could ship as GPT-6.2 or even GPT-6.5, and they're holding it because they're scared.
Not saying it's over for OpenAI. But I'm honestly disappointed they turned this cautious. Then again, I'd rather wait for something solid than get another Hugging Face, just worse.
4
5
425
以前学编程:光配环境劝退一半人。
现在有 AI 带着:Git、Node.js、Next.js、Supabase,花一天就能摸个七七八八。
真别再拿“文科生”、“不懂技术”当挡箭牌了,早不是那个时代了。
13
14
1,147
硅谷现在这帮 AI 公司,本质上都是在给英伟达打工。融几百亿美金,左手倒右手全打进英伟达账户。
也就 Google 比较离谱,从 Gemini 1 到刚出的 Gemini 4,全程没用过一块英伟达的卡。全是自己家 TPU v6e 和最新的 TPU 8t/8i 集群硬抗下来的。别人烧钱买卡,它直接自产自销,这账怎么算都是它赢。
15
11
1,068



