NiceDreamz retweeted
La que se nos viene encima.
Para lo bueno y para lo malo.
Cuánto de cada, yo no tengo ni idea.
In 79 AD Vesuvius buried a library. The scrolls are charcoal now and can't be opened, but X ray scans and AI ink models can find the letters inside. I searched every open ink map from the Vesuvius Challenge, 309 layers across 13 scrolls, and found candidate Greek letters on four unread pieces. Candidates, not translations. Every raw image is public so anyone can check.
#Herculaneum #VesuviusChallenge #Archaeology #MachineLearning #Science
2
3
117
Explore the scans in 3D and check every find yourself
nicedreamzapp.github.io/herc…
9
DaisyRose just took Qwen3.8's spot on my Mac. 18 jobs, both full bf16, graded by code, 94.9 to 88.2. It won on purchasing and fulfillment, the stuff that actually runs a store. Made the whole video locally too.
nicedreamzapp.github.io/agen…
2
96
I asked one quick question and it took a number behind 16 free AI models. Zero finished.
So I built a bouncer: every model gets timed every 6 hours, gets 10 seconds to start talking, or it's the trapdoor. Customer emails skip the line straight to Claude.
The narrator in the corner is me, made on my own Mac. 🛸
2
1
128
NiceDreamz retweeted
This is f*cking insane: OpenAI co-founder Andrej Karpathy says a tiny 1B model can beat most of today's giant LLMs.
"A 1B model can be much more effective than most current LLMs. 90% of current models waste a ton of capacity and money on remembering stuff that doesn't matter"
In 64 seconds he explains why intelligence lives in the system around the model, not in its size.
Memorized Internet → Cognitive Core → Tools → Feedback Loop → Agent That Works
Save this before it gets lost in X's algorithms.
Everything below is free:
Neural Networks: Zero to Hero
Karpathy's own course, from scratch
karpathy.ai/zero-to-hero.htm…
nanochat
Karpathy's "best ChatGPT that $100 can buy"
github.com/karpathy/nanochat
Effective Harnesses for Long-Running Agents
The system around the model, by Anthropic
anthropic.com/engineering/ef…
Effective Context Engineering for AI Agents
What to feed the model and what to cut
anthropic.com/engineering/ef…
All of this costs $0
99% will keep waiting for a bigger model.
The 1% who build better systems will win with a smaller one.
Bookmark it and start with the first link tonight
My November local AI film is up: Hank & Doug's Wild Rescue, a painted storybook trailer. Every still, shot and voice rendered on one M5 Max, with one song running the whole way through and real CC0 sound effects. #LocalAI youtu.be/8_0J0OSYS_k
3
1
42
NiceDreamz retweeted
Google is shaking
Introducing SID-2, a new search model built on compute-scaling swarms.
Our swarms find better results than Opus and Astra, while being 8x faster and 20x lower cost.
By 2030, AI will write an internet's worth of text every 24 hours. Current search pipelines were built on hand-designed rules, adding more won't scale to meet this incoming volume of data.
SID-2 is available on our API today. Try it on your most demanding searches.
NiceDreamz retweeted
I posted this test footage on X, called Charles, and a few weeks later we made Black Thunder.
Neo-Hollywood has no gatekeepers & moves at the speed of a good idea ⚡
teapawt.net/ryanlightbourn/status/…
NiceDreamz retweeted
I got a 1.9B model to watch a restaurant camera and tell staff which tables need cleaning 🍽️
It runs on my laptop. No cloud, and it generates zero tokens.
Built over a weekend with @liquidai d1-2B
NiceDreamz retweeted
This is actually INSANE.
Fable 5.2 just created this in one shot, and I did it live on stream.
I gave Claude one prompt.
It wrote code that builds the entire room in Blender: the desk, the MacBook, the light, even Clawd.
The laptop screen is a web page Claude built to look like the Claude app, played on the laptop inside the 3D scene.
Then it animated everything itself: 13 camera shots, every single key press, Clawd jumping around, and the sun going from golden hour to night to sunrise.
It rented 10 GPUs on RunPod and rendered the whole thing in 4K in 52 minutes. Total cost: $5.75.
It graded it, added the titles, and made the music and sound effects with ElevenLabs.
(More info below)
卧槽,这踏马太毒了
DeepMind 掌门人、诺奖得主 Demis Hassabis 刚刚把延续了上百年的现代学校教育底裤彻底扒光了:
全球几十万名数学老师每天关在办公室里,为同一门中学数学课重复手搓着一模一样的 PPT,这简直是人类智力资源最荒诞的工业级空转。
很多人至今还把名校名师的课堂讲授当成稀缺资源,拼了命去抢学区房和重点班,
但 Hassabis 直接用最冷酷的第一性原理砸碎了这个幻觉,
标准化的知识灌输从来不配占用昂贵的线下课堂,死记硬背和刷题完全应该被踢出教室,交给能按每个孩子毫秒级反应速度实时调频的私人 AI,加上全世界最顶尖的那几个大师录好底层大课,就足以把全球基础教育的讲授环节一次性彻底平替。
过去两百年里,学校本质上是一座把千差万别的碳基大脑强行塞进同一条流水线、按同一个节拍器拧螺丝的工业时代兵营,
老师沦为了人肉复读机,学生被困在枯燥的题海里比拼谁更像一台劣质硬盘,
而一旦 AI 和顶级大师把标准化的知识传递彻底接管,线下教室唯一剩下的合法存在价值,就只剩下了机器永远无法通过权重计算出来的五个维度:
人际博弈、复杂项目实战、创业野心、创造力,以及最致命的判断力。
这里有一个极度冷酷的反直觉真相:
当个性化 AI 把考高分的边际成本砸穿地板,按标准答案培养出来的做题家,在出校门的那一刻就已经直接沦为了算力废料。
未来的阶层分化,从来不再取决于谁的脑子里背下了更多现成的公式,
而是取决于当 AI 把一万种完美方案同时拍在桌上时,谁拥有拍板做决定的判断力,谁能攒起一个活人团队去把事情干成,
所以别再逼着孩子在死记硬背的泥潭里跟硅基芯片比拼记忆容量了,当流水线式的教书匠和做题家同时迎来历史性失业,谁能先把教室变成训练人性直觉与创业野心的实战沙盘,谁才能在算力时代拿到唯一不被平替的入场券
Can you trust a local AI agent? Gemma 4 31B on my M5 Max, nothing in the cloud. It fixes a customer order and stops for my yes before emailing them. Then it reads a law firm's files with the internet cut and pulls every deadline.
2
3
81
NiceDreamz retweeted
Today, we are launching Walkie One. It is the world’s first AI walkie-talkie.
We built it because we wanted a screenless way to interact with technology.
Until recently, that wasn’t really possible in a useful way, but that has changed in the past few months with the latest AI models.
Walkie One connects to the AI assistant of your choice. Once connected, you can use the Walkie One to message your assistant anytime, tapping into the capabilities of your connected AI.
Hit your Claude Code limit? Keep going.
Claude Code Local v1.0. Type exit, then keep going, and the same conversation carries on with a model on your own computer. No API key.
youtu.be/a9vtBSl_j_o
1
1
112
Most of the work went into the tool call parser. Gemma 4 lost its number arguments, two Read calls became one, a screenshot pushed a 32 GB Mac into swap. All fixed. Honest part: a local model is not Opus, and small Macs want one step at a time.
1
11
Free and open source, MIT.
github.com/nicedreamzapp/cla…
1
9











