Auto Magic @AutoMagic108

it’s honestly insane how much of the modern tech tree has been unlocked because of a single guy thanks to Elon we have - online payments - mass market electric cars - batteries and grid-scale energy storage - self driving cars and robotaxis - humanoid robots - advanced brain computer interfaces - next generation tunneling - reusable rockets -global satellite internet - basically the modern space industry - Moon and Mars transportation - orbital compute and space data centers - frontier AI - gigantic AI supercomputers - physical world AI and the wildest part is that these branches are now starting to merge there is a very real chance that this one guy ends up being the reason humanity reaches Kardashev II genuinely the man of the millennium

Wrong advice for junior developer: look at the AI generated code, since you need to mature as a programmer. Right advice: save time with AI coding and write by hand "get-better" programs: interpreters, games, ray tracers, ...

Human reproduction is already biological RSI

If you have 1000 hours in factorio I’ll happily interview you

Dillon Mulroy @dillon_mulroy

5 days ago

i’m positive i could make better hiring decisions based on somebodies top 9 games than any leetcode interview

411 84 3K 384K 227

The e/acc conversion funnel NEVER doom

decided this is the week gave astra my email and paypal and told it to analyze the past few years of business and set up a plan to automate the entire thing so i can travel

Reminder: AI capabilities are advancing rapidly AND we might be in the midst of a huge capex financial bubble... AI adoption (= revenues, productivity) can lag capabilities by years. Depreciation on GPUs is of order years (4? 6?) so a lag like this is enough to DESTROY leveraged (debt-fueled) bets 🤔 Remember, HUMANS ARE IN THE LOOP. They can delay adoption or power supplies to DCs or create regulatory constraints. The sensitivity timescale for LEVERAGED BETS is smaller than the delays humans can impose on AI adoption/scaling. Even if humans don't DELIBERATELY delay productive use of AI, human leaders and organizations may STILL TAKE YEARS to figure out how to effectively use AI in the real economy ...

steve hsu @hsu_steve

a month ago

BOTH THINGS CAN BE TRUE AT THE SAME TIME: 1. AI capabilities are real and advancing rapidly 2. It will take longer than singularity or fast-takeoff people think for AI to affect real productivity in the economy, or generate $trillion revenues for AI labs (although capex

25 16 171 25K 39

I told 200 AI villagers the bread was about to run out. It wasn't. Within 18 hours, 156 of them believed it, and they emptied the bakery three days in a row. Every choice is made by an AI (Jev, by TypeSafe). They all panicked lol. On day 3 the truth turned up: a flour cart at the bakery and a villager who'd seen the mill. It won in the end, with 172 of 200 knowing it was false by day 5. But the truth took more than two days to reach as many people as the lie reached in 18 hours, and the bakery kept selling out after it arrived.

I love my computer

there are entire professions and businesses owned by late boomers and genXers who still think AI is an overhyped Google there’s millions to be made by just showing up, demoing them how to integrate AI in their work and charging them as “AI consultant” or something

vittorio @IterIntellectus

6 days ago

i showed my dad how to use codex and claude code for his work. he just called me crying tears of joy because he’d done in 5 minutes what had taken one of his employees an entire month.

186 186 8K 389K 671

May Starship have the Mandate of Heaven

here is an overview of the 4 most advanced efficient architectures: Deepseek V4.1 Flash, MiMo V3, Qwen 3.8 Next Flash and GLM 5.3 Flash Deepseek and MiMo are quite similar, they both use YOCO - only the first part of the network is active during prefill to build the KV cache representation - no linear attention, and token level indexer Qwen and GLM both use a more standard interleaving 3:1 (like Kimi K3 as well) between sparse attention and linear attention (GDN vs KDA) both deepseek and qwen use Engram, they all use a gate or sink except GLM 5.3 Flash, they also all use no or partial RoPE on the full/sparse attention layers they also all have some sophisticated residual network, either mHC (simplified or not) or gated residual and they are all trained using Muon visualization by opus 5.5 and me :)

there are many many people who studied physics and came to the conclusion that working on machine learning would probably create more scientific progress than studying physics including many of the original chatgpt crew like john schulman, liam fedus, amodei & jared kaplan at ant

"People aren't ready, this is gonna change everything" You know what people weren't ready for? Fucking mammoths. Tigers eating them. Monkeys ripping their arms off. The black plague. The ice age. Smallpox. Vikings pulling up on the beach. Drinking water that gives you diarrhea until you die. Getting a scratch and losing your whole leg. The Mongols. The Spanish Inquisition. One volcano can send us back to the dark ages and these safety guys are gonna be shocked. Oh the robots didn't kill us? It was a volcano? Imagine their disappointment as the hot ash rains down on them. A solar flare is gonna explode your phone in your pocket your legs are gonna be covered in battery acid and your bank account will stop working. You're afraid of a computer program doing math? "Things are moving too fast" homie you live on a rock that periodically deletes everything. You are one the most luckiest entities in a 100 light year radius living during a golden era. Cheer up. My ancestors crossed an ocean in a wooden box because they ran out of fucking potatoes. "The social disruption will be enormous!" Nigga World War 2? World War 2? Hop in the people carrier we're landing on the beach. Are these people out of their fucking minds? Do they not know what we've survived already? A couple decades of air conditioning and people suddenly think oh this is the natural condition of mankind. No. We used to have eight kids because seven didn't make it. You're gonna be fine. Man the fuck up. Accelerate.

AI is getting cheaper more quickly than any other transformative tech in history. At a given level of performance, cost has fallen ~47%/quarter since 2023. That’s 4× faster than DNA sequencing, 6× faster than compute, 18× faster than lithium batteries, and (up to 1973) 54× faster than electricity.

The frontier of mathematics will soon reach far beyond the subset that human minds discover. Working memory of ~10 is a very strong constraint on ape brains! 🧠🐒 youtu.be/q4Uao6leMyw?si… Compression is all you need: Modeling mathematics Abstract: The mathematics humans discover and value (“human math”) is a vanishingly small subset of all valid deductions (“formal math”). I’ll argue that human math is distinguished by its compressibility through hierarchically nested definitions and theorems, like a polynomial-growth space rather than the exponential-growth space one might expect when proofs are viewed as strings of symbols. The argument combines toy monoid models with an empirical analysis of MathLib, a large Lean library of formalized mathematics we treat as a proxy for human math. I’ll close with how compression itself can serve as a measure of mathematical interest, giving agents a sense of direction toward where human math lives.

> We might have penalized response length too much (or something) in RL, as it still gives up on hard tasks (that it can do!) too early For the record, here is Elon's solution to this problem. Grok 4.7 xhigh, Grok Build. Godspeed, you glorious bastard… Per aspera ad Astra!

Elon Musk @elonmusk

3 weeks ago

@farzyness Grok 4.7 needs a few more days to cook. We might have penalized response length too much (or something) in RL, as it still gives up on hard tasks (that it can do!) too early and isn’t yet sufficiently rigorous in checking its work.

2K 998 20K 2.1M 799

最近看到 MiniMax Code CLI 开源了，我其实好奇这些 Harness 都有啥区别，深入研究了一下发现， Mcode 居然对 subagent 做了一次工程优化？！ 很多人用多 Agent 容易翻车，本质是因为市面上大多数方案的工作方式太像一个“微信工作群”了： 主 AI 派活，子 AI 在群里发一段 500 字的小作文汇报 更致命的是，几个子 AI 直接跑到你电脑的同一个文件里抢着改代码。 A 刚删掉第 10 行，B 又插了 5 行，两边一存盘直接互相覆盖，尤其是 Codex 经常会碰到这种问题。不仅文件被改崩，返工的时间还会特别浪费Token Mcode 这次最有意思的地方，就是把这种“群聊模式”改成了“数据库事务”： 1. 透明草稿纸（私有 Overlay）：子 AI 读代码随便读，但只要想改代码，系统会垫一张虚拟草稿纸。所有笔迹都在纸上，真实文件纹丝不动。就算 5 个 AI 同时改同一个文件，物理上也碰不到对方 2. 条形码清单（ChangeSet）：干完活严禁写长篇小作文汇报，必须交出一张“条形码清单”——修改前的原文件指纹是什么、具体动了哪几行。主控拿到的是客观数据补丁，而不是大模型的自我感觉 3. 无冲突合流：改动不重叠就自动拓印回磁盘；一旦改了同一行，立刻亮红灯停下，绝不让大模型临场瞎编一套覆盖掉 横向看几家主流 Harness 的思路，差异特别有意思： - Claude Code：放权型精英团队。给子 Agent 充分自主权和独立上下文，直接在磁盘里查跑改，灵活但多 Agent 并发容易撞车，且极其烧钱 - 小米 MiMo Code：单兵死磕型。不搞子 Agent 群聊，每走一步并行采样 5 个方案挑最优，交卷前由独立的 Verifier 严格挑刺，压榨单线长任务的深度 - 智谱 Z Code：集成调度型。路线更偏向平台整合，倾向于调度集成现成的外部 CLI 当 Worker 来协同，核心在生态路由，底层还没切入这种深度自研的文件级隔离； - Mcode：带保险箱的严苛工长。把通信收敛成结构化补丁，把写入收敛进私有沙箱，走确定性的事务合并 写几十行脚本或改个错别字，单模型直接跑最省事；但一旦遇到大 Monorepo 跨模块改动、或者几十个报错要并发排查，这种物理隔离的价值就完全体现出来了。 说到底，现在 Subagent 的核心破局点就在于工程化。mcode 最大的启发，就是用做数据库事务的思维，给多 Agent 协同做了一次扎扎实实的工程化落地 📦 安装： • macOS / Linux: curl -fsSL filecdn.minimax.chat/public/install… | bash • Windows (PowerShell): irm filecdn.minimax.chat/public/install… | iex • 或通过 npm 全局安装: npm install -g @minimax-ai/code

The funniest thing is he was pro-AI until its capabilities surpassed him

Haider. @haider1

2 weeks ago

Mathematician Terence Tao: "we have to slow down AI. the pace is insane, and there's no reason to be this fast — no reason at all" It's amazing how willing we are to change everything without any idea what happens afterward These are extremely nonlinear dynamics

885 1K 7K 7.5M 3K

DO FUCKING HARD THINGS