
Head of hype @lobehub Prev @listenhub @Sapient_Int 没啥好看的,也就发点AI相关的内容🫡 同名小红书4万粉丝,公众号01Founder(长文首发) 观点仅代表个人 更多请访问网站⬇️(欢迎商务、合作
Singapore
Joined September 2023
- Tweets12.9K
- Following3K
- Followers43.8K
- Likes27.8K
Max For AI retweeted
Over the past two months, we have been leading a coalition to mitigate one of the most severe bugs in open source software found in recent memory, php2shell, found as part of our Patch the Planet initiative.
Max For AI retweeted
“super intelligence” will probably never catch on in china
“si” = 死 = death
“ai” = 爱 = love
Introducing Underdog, your Private Personal AI on devices you already own
Today we're announcing our backing from @a16z @khoslaventures @HummingbirdVC Anthology (@AnthropicAI @MenloVentures) @patrickc @naval @rauchg @polynoamial @Thom_Wolf @Mascobot @tszzl @OfficialLoganK and other top AI leaders
Our mission is to provide free, capable, reliable and private AI to billions of people.
Underdog’s Law: today’s frontier intelligence reaches your devices in six months.
We’re starting with fast, capable AI that runs on consumer hardware. By co-designing models and inference engines, we’re pushing the frontier of capability, speed, power and data efficiency.
Our research also spans agentic commerce, confidential inference, and how AI will reshape the internet economy.
Privacy and capability no longer needs to be a tradeoff. If you believe in this future, join us. Time to build.
Max For AI retweeted

Tibo@thsottiauxSep 25
United States
United StatesConnected via WebAccount-level information, not a live location or per-post device.
We are aware that codex is down and are working hard to bring back normal service.
Thrilled to announce @alibaba_cloud as a Founding Corporate Patron for the Omacom Foundation! $3 million in funding, collaboration on Omarchy China, and bringing Omarchy to the newly announced Qwen Book. Agentic computers need a native agentic OS! omarchy.org/news/2026/09/ali…
Max For AI retweeted
Introducing Step 5 Preview: Advancing the Pareto Frontier.
Step 5 Preview is our new flagship model for agentic work, delivering frontier-level performance across software engineering and professional knowledge work, with particular strength in finance.
- 600B total / 27B active MoE, with 1M context + Vision
- Substantially lower task cost at comparable intelligence
- Broad software engineering capabilities with sustained execution over long horizons
Try Step 5 Preview: platform.stepfun.ai
Model page: stepfun.com/step-5-preview
Open weights on Oct 15.
Max For AI retweeted
Replying to @PessimistsArc
Right. Dario was already claiming that GPT2 was too dangerous to open source back in 2019.
I made fun of them then.
Everyone should make fun of them now.
Dario is right
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must…
Max For AI retweeted
DeepSeek 开源了一些新的代码仓库,方便更容易地部署 V4.1 Flash 以及后续的开源模型:
github.com/deepseek-ai/deeps…
github.com/deepseek-ai/DeepS…
github.com/deepseek-ai/DeepJ…
Max For AI retweeted
Replying to @MaxForAI
他说的是前者:在受影响的时间段里,使用过 banked reset(储存/银行式重置券)的人,会再补发一个 reset
Max For AI retweeted
鉴于 DS V4.1 Flash 模型在性能、费用、速度、总用时等各项指标上都全面超越了 V4 Pro。再以更高的价格、更慢的速度和更多的算力消耗给 DS 用户提供原有的性能较差的 V4 Pro 模型会不太合适。V4.1 Flash 正式上线之后,V4.1 Pro 上线之前,V4 Pro 模型的请求都会被路由到 V4.1 Flash,并按 Flash 的价格计费。
Max For AI retweeted
As a kid, I always wished Pokémon battles in the games could play out like they do in the anime.
So I built a real-time Pokémon battle experience with @fal H3 Max.
choose a move, the game calculates the outcome, and AI generates the animation on the fly. Paralysis, fainting, and Pokémon is hurt are reflected on screen.
Project:github.com/xflare-bot/pokemo…
Max For AI retweeted
看到今天AA连夜改index对应的benchmark,想到前两天和 @CyouSakura 聊到的。不过以下都是我的个人观点:
1. build一个bmk很大程度依赖taste,集成一堆bmk也是一样。这次AA被打脸的深层原因就是显然AA和OAI的taste有巨大的gap。具体来说,有的人可能觉得语音模型重要,有的人可能觉得模型需要有cu能力,有的人可能觉得任何的东西都可以被coding agent覆盖。不管是数据公司还是众包团队在造一个benchmark的时候,都是bet这个方向是一个promising的下一个方向,AA也是一样,AA index无非就是AA认为最重要的几个能力,通过一定的权重加和得到的,也就是AA的taste。只能说人家都要AGI了你还搁这儿玛卡巴卡(没错说的就是你GPQA)
2. 以bmk为导向其实并没什么错,错的是overfit到极少数的bmk。本身LLM很大程度上是“崎岖的智能”,如果存在足够多的bmk来观测,我认为几乎不会有人会说刷榜这件事了。不过目前的一个问题似乎是没有足够多高质量的bmk,你能想象critpt scicode ALE这些bmk都有bug么。另一个问题是目前的感觉不像是有足够多的bmk,然后我们一次把它们都刷完了;更像是一个domain刷完了,看看OA两家最近在刷什么榜,我们也去造那个domain的bmk然后刷。总而言之,如果大模型厂商内部如果有足够多高质量的bmk来观测+刷,这应该不太会是问题。
3. bmk在今天仍然是不可或缺的&充满human bias的。前面说到需要有足够多高质量的bmk,很显然让人容易联想到让agent来scale&build bmk。不过目前还不太可行。我认为判断&发现模型的弱点依赖于taste以及verification的能力,比较显然的是现在的LLM不太具有这样的能力,比如可以试试让他给你十个能得test of time 的idea。anthropic也写过一个相关的blog:alignment.anthropic.com/2026…。我觉得一个很本质的问题是没有足够多的这样的数据。 @ZimingLiu11 的播客上提到过这件事情,以及他们想通过做一个meta model来让模型拥有一些sense/taste,eg判断一个训练的curve走势。本质上是同一件事。所以目前看来,RSI中的一个大断点就是taste,因为归根结底,目前的work都是在一个bmk上自提升,这太narrow了。一个朴素的想法是,模型可以提出各种各样的榜单,然后优化自己在榜单上的分数,这可能是一个弱化版的RSI。



























