Aaron Ng

@localghost • 33,711 subscribers

@mecha_corp. Founder of the local AI tool @apolloaiapp (acquired by @liquidai). Created Hyper Online, one of the most popular avatar apps. Former CashApp & Meta

Live Cam

Aaron Ng

Shorts

llama 3 70b beamed to my phone from my M1 Max ~7.6 tok/s with mlx. your own little gpt-4 at home

Here’s Deepseek r1 1.5B thinking through a problem — it’s comparable to 4o and Claude 3.5 Sonnet in a number of domains like math. Except… it’s a 1.5B model… and can run on virtually any hardware. Truly a huge efficiency leap.

Llama 3.3 70B streaming at 7.8 tok/s on an M1 Pro to my phone. Not the fastest, but wild that you can serve something this powerful to your whole house without internet.

Run Llama 3.1 1B offline at 60+ tok/s. Entirely on your phone. Siri can’t save you offline, but a local model just might.

Ollama can host too: here’s Llama 3.3 70B streaming at ~6.5 tok/s from my M1 Max to my phone. Your laptop is an offline self-sovereign AI server.

SmallThinker 3B is one of the AI models I’d want backed up on my phone. Here’s it is reasoning at 12 tok/s on my phone. If you’re ever stuck offline, you’ll want a model that can think.

Offline AI, right on your phone. Small versions of Hermes 3, Qwen 2.5, and Llama 3.2 running privately in your hand. Grab them now in Apollo 1.0.17

Wild that you can run a GPT-4 tier AI on a laptop and serve it to your whole network. Llama 3.3 70B is a beast.

It’s never been easier to have your own private AIs running at home, off-grid, or anywhere else. For AI independence, all you need is: - LM Studio on your computer. - Apollo in your pocket.

Videos

Watch Anya Live

1. Embed a book into GPT-3 / ChatGPT. 2. Talk directly to the book to learn. Learn from a book through conversation, questions, and in the order you want.

ChatGPT / GPT-4 can build a bespoke iOS app in minutes. We're entering a world where custom tools can be made for anything you want to do. Reply to get a DM with the prompt and output repo.

Apollo: a ChatGPT-powered app for real-time knowledge. Talk to it all day long through your headphones. If augmented reality is an overlay on the world, this is augmented intelligence — an overlay on your thoughts. Reply for a TestFlight DM soon.

Here’s Ministral 8B running at 10 tok/s on a phone. Siri isn’t going to save you when you’re stuck without internet. An offline backup AI is essential.

GPT-4 + Eye Tracking = an AI that knows what you’re looking at. The line where our tools end and we begin continues to blur.

here’s gptfile, a way to organize files with natural language using gpt-4. new operating system paradigms are on the horizon repo below

Interview Breaker — a ChatGPT tool that tells you what to say in job interviews. AI will upend every process with just a few lines of code. We’ve already seen it with school essays. Interviews will need to adapt, too.

ChatGPT + Computer Vision = A dictionary for what you see. By giving ChatGPT eyes, we allow it to answer questions about the world around us.

Reminder that Ollama can serve AI models to your whole house. Just expose it to the network with OLLAMA_HOST. Access all your AI models over your network — for free.

ChatGPT + Natural Language + Personal Context = a more useful personal home device. By teaching ChatGPT about me, I get answers specific to me and my goals.

Apollo is a ChatGPT-powered app you can talk to all day long to learn from. It can now shift its conversation style based on what you're looking for, from asking questions to just forwarding you articles. What else should we add to the TestFlight?

You’ve seen 1B and 3B models running on phones. Here’s a much smarter 8B model. If your phone supports it, there’s no reason not to have a larger model downloaded.

custom tools on demand mesh gradient editor built and running right in claude. took maybe 30 minutes

here’s deepseek r1 7b running @ 10 tok/s best reasoning model you can run on a phone today. might just be what you need if the internet goes out suddenly.