🔥 if You Read One Thing...

🔥 If you read one thing about AI this week, let it be Google’s leaked doc “We have no moat, and neither does OpenAI”

(View Tweet)

Full article here: https://t.co/l7TyOCCIeq Some quotes 🧵 (View Tweet)

Open-source is innovating faster and cheaper than us, or OpenAI, ever can. Examples:

(View Tweet)

Large expensive models are still slightly better, but the gap is quickly closing.

(View Tweet)

Meta’s LLaMA leaked, and the open-source community tan with it.

(View Tweet)

Some innovations are now coming from single people, tinkering on their laptop “in their garage”

(View Tweet)

After StableDiffusion it shouldn’t have come as a surprise. (Anyone remember DALL-E? many people sure don’t..)

(View Tweet)

Seems like having a lot of money to train large models might cause them to pay less attention to far cheaper methods (inversely, that’s what the tinkerer at home works with)

(View Tweet)

And maybe those small, cheap, incremental updates can outrun progress on large, expensive models.

(View Tweet)

To save money, “tinkerers” are training on smaller, but better curated data-sets. Outputs from ChatGPT that have been shared by users because they were cool are used.

(View Tweet)

Highly specialized individuals are making highly specialized products.

(View Tweet)

And Meta might be benefiting here… because the leaked model was theirs, they can now use all the innovation that’s happening on top of it 🥲

(View Tweet)

Is Google about to open-source a lot more stuff? “If you can’t beat ‘m, join ‘m”?

(View Tweet)

And why be secretive? “OpenAI already knows everything we do because they hire our people.”

(View Tweet)