cofounder & buidler @sanctumso | member @superteamsg
Joined November 2019
- Tweets1.1K
- Following1.8K
- Followers2.3K
- Likes4.5K
悟空到底在菩提祖师那里学了什么？
这么说吧：
老师「我教你线性代数、概率统计、论文精读，再从零把模型训出来，以后做个真正的研究员。」
你「不学，不学！」
老师「我教你系统、CUDA、推理引擎和分布式训练，以后做个 AI 基础设施大佬。」
你「不学，不学！」
老师「我教你数据、评测、对齐和产品落地，以后做个能把模型用到正经事上的人」
你「不学，不学！」
老师砸了三下键盘，转身而去。
半夜，你跑到老师办公室，老师小声说：
「我教你提示词套壳、批量生成爆款、换脸换声、自动群发话术，以及把开源模型包装成『自研 AGI』的融资秘籍，保你一毕业就能靠 AI 骗到大把大把的钱。」
你「对对对，我就想学这个！老师，我以后发大财怎么谢你？」
老师「说什么谢不谢的，你以后出了事，别把老师供出来就行。」
HY | Sanctum retweeted
1 day out 🫣
In Seoul for KBW? Make your Sanctum card in a minute, it shows everyone where to find you.
Post it on X and you're in the Seoul Card Raffle. Card holders only, drawn at SANC Nine on 1 Oct (Tomorrow).
Make yours here: seoul.sanctum.so 👈🏻
HY | Sanctum retweeted
After weeks of user conversations, the same problems keep coming up for Sanctum users.
So we're going all in on fixing them, in our biggest update yet.
More details coming soon 👀
Consider using @high52weeks ‘ github.com/peonist-ai/haloge… for consistent 3x prefill speeds and 1.5x decode speeds even at long contexts
HY | Sanctum retweeted
We voted against our own interest in the last second to swing the outcome to pass.
As the No.1 protocol on Solana, we wanted to lead by example and when it came down to it, we decided this was best for the ecosystem and for our community.
Thank you for your trust in @sanctumso
It’s no surprise that @everstake_pool, @krakenfx, @Figment_io and @Ledger with captive stakers vote for short term interest and abuse their position to vote against reducing $SOL inflation. They didn’t participate and wouldn’t participate in the discussions because it’s not defensible.
SGP-002 won’t pass for the same reason SIMD-228 didn’t pass.
you usually don't even realize you're exploiting this bug in the "believable organic" cases
nobody cares about this, but the new program binary is also ~7x smaller
