Luca Ceppelli

[Luca Ceppelli @lucaceppelli]

My Codex usage is what happens when “just one more prompt” stops being a plan and starts being a dependency 💉

🤖 I had to upgrade to Pro 20x just to keep the addiction running.

🙏 Thanks, Sam @sama . 😅💸💸💸

[Luca Ceppelli @lucaceppelli]

🧠💣🤤

[natalie @natalie_thenerd]

Clear PCBs feel illegal

[Luca Ceppelli @lucaceppelli]

AMD 9070XT eGPU > 100 TFLOPS 🚀🧠💣

[the tiny corp @__tinygrad__]

107 TFLOPS on Mac M3 with a 9070XT attached. Our GEMM with the AMD_LLVM backend is now beating hipBLASLt on the card. And I love how portable LLVM is, this is just brew install llvm@19.

[Luca Ceppelli @lucaceppelli]

GPT-4o-mini is a ~8B parameter model (ONLY 🧠💣🤯)

[Yuchen Jin @Yuchenj_UW]

A Microsoft paper suggests that GPT-4o-mini is a ~8B parameter model.

It means distillation works pretty well at @OpenAI.

Other model sizes from the paper:
- Claude 3.5 Sonnet: 175B
- GPT-4: 1.76T
- GPT-4o: 200B
- o1-preview: 300B
- o1-mini: 200B

@sama, would you consider open-sourcing GPT-4o-mini? It could run on our local devices.

[Luca Ceppelli @lucaceppelli]

@farzyness @WalterIsaacson it’s look like the this car has only one rear wheel. @farzyness what do you think about it ?

#tesla
#robotaxi

[Luca Ceppelli @lucaceppelli]

@elonmuskADO TwitteX 😎

[Luca Ceppelli @lucaceppelli]

@GoogleDeepMind AlphaFold3 🧠+🧬=❤️

#AI
#AlphaFold
#DrugDiscovery

[Isomorphic Labs @IsomorphicLabs]

We're excited to announce #AlphaFold 3 with @GoogleDeepMind in @Nature: our new AI model for predicting biomolecule structures with unprecedented breadth and accuracy.

Expanding beyond proteins to tackle DNA, RNA, small molecules to fuel advances in biology & drug design 🧵

[Luca Ceppelli @lucaceppelli]

@huggingface
#AI
#LLM

[Luca Ceppelli @lucaceppelli]

@karpathy did it again 😎🚀

#AI
#LLM

[Andrej Karpathy @karpathy]

Have you ever wanted to train LLMs in pure C without 245MB of PyTorch and 107MB of cPython? No? Well now you can! With llm.c:
https://t.co/PoGTZIwASL

To start, implements GPT-2 training on CPU/fp32 in only ~1,000 lines of clean code. It compiles and runs instantly, and exactly matches the PyTorch reference implementation.

I chose GPT-2 to start because it is the grand-daddy of LLMs, the first time the LLM stack was put together in a recognizably modern form, and with model weights available.

[Luca Ceppelli @lucaceppelli]

@herbertong August 8th the best advertisement 🎉🎉🎉 in my opinion would be to the launch 🚀 of the robotaxi service in the tunnels dug by The Boring Companies in Las Vegas. 🤯What do you think @herbertong 🤔?

[Luca Ceppelli @lucaceppelli]

Structured Machine Learning 🧠💣🚀

[Symbolica @symbolica]

Watch @symbolica Principle Scientist Dr. Paul Lessard talk about why category theory is the key to achieving a description of symbolic reasoning in machines and how we are using it to usher in a new era in machine learning.

[Luca Ceppelli @lucaceppelli]

1M context tokens 🤯 🧨🚀

[Google DeepMind @GoogleDeepMind]

Introducing Gemini 1.5: our next-generation model with dramatically enhanced performance. It also achieves a breakthrough in long-context understanding.

The first release is 1.5 Pro, capable of processing up to 1 million tokens of information. 🧵 https://t.co/qT0aXdFL0n

[Luca Ceppelli @lucaceppelli]

🤖+🧠=🧨

[Elon Musk @elonmusk]

Optimus folds a shirt

[Luca Ceppelli @lucaceppelli]

💬+🧠=🎆 @ 🚀

[Stability AI @StabilityAI]

Introducing SDXL Turbo: A real-time text-to-image generation model.

SDXL Turbo achieves state-of-the-art performance with a new distillation technology, enabling single-step image generation with unprecedented quality, reducing the required step count from 50 to just one.

The code, research paper, and weights for non-commercial use are now available on our website.

You can test SDXL Turbo on Stability AI’s image editing platform @Clipdropapp, with a beta demonstration of the real-time text-to-image generation capabilities.

Learn more: https://t.co/L39rZWf9F7

[Luca Ceppelli @lucaceppelli]

🧠+💻=👽

[Andrej Karpathy @karpathy]

LLM OS. Bear with me I'm still cooking.

Specs:
- LLM: OpenAI GPT-4 Turbo 256 core (batch size) processor @ 20Hz (tok/s)
- RAM: 128Ktok
- Filesystem: Ada002

[Luca Ceppelli @lucaceppelli]

@SipeedIO One of my favorite companies for diy electronics geeks like me👨‍💻🧠🚀

#RISCV
#FPGA
#rustlang

[Sipeed @SipeedIO]

10K Followers!🥳🎉🎉

[Luca Ceppelli @lucaceppelli]

@teslaeconomist [2/2] if this is true, why open a new factory in a country where labor costs are lower if it is no longer necessary?

[Luca Ceppelli @lucaceppelli]

@teslaeconomist [1/2] Probably during the first phase of the pilot line of the new open box car manufacturing design the engineers at Tesla are realizing that all the work (or quite all) of building the car could be done by Optimus in 12/18 months further development.

[Luca Ceppelli @lucaceppelli]

🤖🤖🤖

[Tesla Optimus @Tesla_Optimus]

Optimus can now sort objects autonomously 🤖

Its neural network is trained fully end-to-end: video in, controls out.

Come join to help develop Optimus (& improve its yoga routine 🧘)

→ https://t.co/dBhQqg1qya

[Luca Ceppelli @lucaceppelli]

🚀🚀🚀

[the tiny corp @__tinygrad__]

tinygrad is faster than torch at everything on M2 Max

Try it for your next ML project on your MacBook