[Kevin Comics] Zuckerberg AI

🃏Joker

[email protected]

The goal is manifold; one for example is to have AI agents talk to your customers in your Instagram shop when they have questions.

Another example is internal tools at Meta; obviously I can’t go into too much detail, but the AI tools help the development workflow a lot.

AI also pairs well with the smart glasses like the Raybans or Orion. You might not like it, but having your glasses explain your health insurance at the hospital or be a personal fitness coach at the gym are actually very helpful use cases.

Source: I work at Meta, and am very bullish on the future of AI

[email protected]

Creators on Facebook do get paid though, at least if they’re big enough I guess creators.facebook.com/earn-money

Also the AI model Meta maintains (Llama) is the most powerful open-source model that anyone can use and even build their own commercial products on top of for free, so I’m not sure it’s accurate that nobody wants it?

Ephera

Only the inference code of LLaMA (which runs the model) is open-source. The model itself is not, as you’re given neither the training data, nor the model weights.

[email protected]

I don’t know much about AI models, but that’s still more than other vendors are giving away, right? Especially "Open"AI. A lot of people just care if they can use the model for free.

How useful would the training data be? Training of the largest Llama model was done on a cluster of over 100,000 Nvidia H100s so I’m not sure how many people would want to repeat that.

[email protected]

I know there are some very loud and dedicated haters out there, but “that no one wants”? Bro needs to get out of his bubble.

[email protected]

I haven’t heard anyone articulate anything compelling about consumer-marketed AI so please tell me! There’s loads of really good uses of AI (medical imaging seems really promising) but the ones I know about are so specialised that I can’t see why I would need “AI” in my day to day.

[email protected]

Parent council used it to whip up a Halloween story for our event, it fleshed it out which ended up saving time. Needed some shaving, but nothing as intensive as writing out the entire story yourself for something that is essentially a one time thing.

[email protected]

It turns out these clusters are being used very inefficiently, seeing how Qwen 2.5 was trained with a fraction of the GPUs and is clobbering models from much larger clusters.

One could say Facebook, OpenAI, X and such are “hoarding” H100s but are not pressured to utilize them efficiently since they are so GPU unconstrained.

Google is an interesting case, as Gemini is getting better quickly, but they presumably use much more efficient/cheap TPUs to train.

[email protected]

How useful would the training data be

Open datasets are getting much better (Tulu for an instruct database/recipe is a great example), but its clear the giants still have “secret sauce” that gives them at least a small edge over open datasets.

There actually seems to be some vindication of using massively multilingual datasets as well, as the hybrid chinese/english models are turning out very good.