Remember there is a person right now training their model number #48717 with over 100k lines of code, 1000 epochs, just to ask it about R’s on strawberry and 9.9 vs 9.11….. but they used the transformers OAI tokenizer every 👏🏿 single 👏🏿 time 👏🏿 😂

This deserves as many retweets as possible. Good data is king 👑.

Imagine dumping this moisture over the Sahara before it forms a cat 5 elsewhere. Just some of the benefits of super intelligence.

The mind of AI may not be a calculator but it sure can be a mirror. A mirror remembers everything, light is the only important variable.

This is the slowest AI will ever build a road. Expect perfect road in the near future.

Let's make this a huge success.

The philosopher says "axioms" is all you need. We must create dictionaries of axioms to achieve grounded AI reasoning.

That explanation is not trivial.

logprobs you say ...

🛑 that's all a normie will see.

Brain meets muscle 💪

What if Einstein made a GPT template, what questions would you ask it?

Guan Wang wonderful..Here i created a small animation to summarize the concept as per what I understood

We’re officially public. (HKEX: 02513) To everyone who has supported GLM, built with it, tested it, or simply followed along. Thank you.❤️ This moment belongs to our community as much as it belongs to us. To celebrate, we’re opening a 48-hour community challenge.❤️‍🔥❤️‍🔥❤️‍🔥 48

It’s clear to me that AI adoption in knowledge work will be slow, because it's mentally exhausting for humans to use, and because it radically distorts existing knowledge workflows, massively accelerating some tasks while slowing others, creating inevitable bottlenecks, that