Live data from Hacker News

Meta Llama 3

llama.meta.com

511–520 of 965 posts

Re: Meta Llama 3

#511
post #504
post #240

Earlier quoted context omitted.

You can see from Zuck's interviews that he is still an engineer at heart. Every other big tech company has lost that kind of leadership.

But also: Facebook/Meta got burned when they missed the train on owning a mobile platform, instead having to live in their competitors' houses and being vulnerable to de-platforming on mobile. So they've invested massively in trying to make VR the next big thing to get out from that precarious position, or maybe even to get to own the next big platform after mobile (so far with little to actually show for it at a str…

Similarly to Google keeping Android open source, so that Apple wouldn’t completely control the phone market.

Re: Meta Llama 3

#512
post #497

I just want to express how grateful I am that Zuck and Yann and the rest of the Meta team have adopted an open approach and are sharing the model weights, the tokenizer, information about the training data, etc. They, more than anyone else, are responsible for the explosion of open research and improvement that has happened with things like llama.cpp that now allow you to run quite decent models locally on consumer h…

Let's be honest that he's probably not doing it due to goodness of his heart. He's most likely trying to commoditize the models so he can sell their complement. It's a strategy Joel Spolsky had talked about in the past (for those of you who remember who that is). I'm not sure what the complement of AI models is that Meta can sell exactly, so maybe it's not a good strategy but I'm certain it's a strategy of some sort

Also keep in mind that it's still a proprietary model. Meta gets all the benefits of open source contributions and testing while retaining exclusive business use.

Re: Meta Llama 3

#513

Earlier quoted context omitted.

Good thing that he's only 39 years old and seems more energetic than ever to run his company. Having a passionate founder is, imo, a big advantage for Meta compared to other big tech companies.

Love how everyone is romanticizing his engineering mindset. But have we already forgotten that he was even more passionate about the metaverse which, as far as I can tell, was a 50B failure?

If you actually listen to how Zuck defines the metaverse, it's not Horizons or even a VR headset. That's what pundits say, most of whom love pointing out big failures more than they like thinking deeply.

He sees the metaverse as the entire shared online space that evolves into a more multi-user collaborative model with more human-centric input/output devices than a computer and phone. It includes co-presence, mixed reality, social sites like Instagram and Facebook as well as online gaming, real-world augments, multiuser communities like Roblox, and "world apps" like VRChat or Horizons.

Access methods may be via a VR headset, or smart glasses, or just sensors that alert you to nearby augmented sites that you can then access on your phone - think Pokemon Go with gyms located at historical real-world sites.

That's what $50B has been spent on, and it's definitely a work in progress. But it sure doesn't seem dead based on the fact that more Quest headsets have been sold than this gen's Xboxes; Apple released Vision Pro; Rayban Smart Glasses are selling pretty well; new devices are planned from Google, Valve, and others; and remote work is an unkillable force.

The online and "real" worlds are only getting more connected, and it seems like a smart bet to try to drive what the next generation looks like. I wouldn't say the $50B was spent efficiently, but I understand that forging a new path means making lots of missteps. You still get somewhere new though, and if it's a worthwhile destination then many people will be following right behind you.

Re: Meta Llama 3

#514

Earlier quoted context omitted.

Having an engineering mindset is not the same as never making mistakes (or never being too early to the market). The only way you won’t make those mistakes and keep a perfect record is if you never do anything major or step out of the comfort zone. If Apple didn’t try and fail with Newton[0] (which was too early to the market for many reasons, both tech-related and not), we might’ve not had iPhone today. The engineer…

His engineering mindset made him blind to the fact the metaverse was a product that nobody wanted or needed. In one of the Fridman interviews, he goes on and on about all the cool technical challenges involved in making the metaverse work. But when Fridman asked him what he likes to do in his spare time, it was all things that you could precisely not do in the metaverse. It was baffling to me that he failed to connec…

and is responsible for building evil products to fund this stuff.

Apple photos and FaceTime are good products for sharing information without ruining your attention span or bring evil. Facebook could’ve been like that.

Re: Meta Llama 3

#515

I just want to express how grateful I am that Zuck and Yann and the rest of the Meta team have adopted an open approach and are sharing the model weights, the tokenizer, information about the training data, etc. They, more than anyone else, are responsible for the explosion of open research and improvement that has happened with things like llama.cpp that now allow you to run quite decent models locally on consumer h…

Yes - for sure this AI is trained on their vast information base from their social networks and beyond but at least it feels like they're giving back something. I know it's not pure altruism and Zuck has been open about exactly why they do it (tldr - more advantages in advancing AI through the community that ultimately benefits Meta), but they could have opted for completely different paths here.

Re: Meta Llama 3

#516
post #112

Earlier quoted context omitted.

The bottom of https://ai.meta.com/blog/meta-llama-3/ has in-progress results for the 400B model as well. Looks like it's not quite there yet. Llama 3 400B Base / Instruct MMLU 84.8 86.1 GPQA - 48.0 MATH - 57.8 HumanEval - 84.1 DROP 83.5 -

For the still training 400B: Llama 3 GPT 4(Published) BBH 85.3 83.1 MMLU 86.1 86.4 DROP 83.5 80.9 GSM8K 94.1 92.0 MATH 57.8 52.9 HumEv 84.1 74.4 Although it should be noted that the API numbers were generally better than published numbers for GPT4. [1]: https://deepmind.google/technologies/gemini/

Those numbers are for the original GPT-4 (Mar 2023). Current GPT-4-Turbo (Apr 2024) is better:

          Llama 3 GPT-4   GPT-4-Turbo* (Apr 2024)
    MMLU  86.1    86.4    86.7
    DROP  83.5    80.9    86.0
    MATH  57.8    52.9    73.4
    HumEv 84.1    74.4    88.2
*using API prompt: https://github.com/openai/simple-evals

Re: Meta Llama 3

#518
post #443

Earlier quoted context omitted.

>We’re rolling out Meta AI in English in more than a dozen countries outside of the US. Now, people will have access to Meta AI in Australia, Canada, Ghana, Jamaica, Malawi, New Zealand, Nigeria, Pakistan, Singapore, South Africa, Uganda, Zambia and Zimbabwe — and we’re just getting started. https://about.fb.com/news/2024/04/meta-ai-assistant-built-wi...

That's a strange list of nations, isn't it? I wonder what their logic is.

No EU initially - I think this is the same with Gemini 1.5 Pro too. I believe it’s to do with the various legal restrictions around AI which iirc take a few weeks.

Re: Meta Llama 3

#519
post #497

I just want to express how grateful I am that Zuck and Yann and the rest of the Meta team have adopted an open approach and are sharing the model weights, the tokenizer, information about the training data, etc. They, more than anyone else, are responsible for the explosion of open research and improvement that has happened with things like llama.cpp that now allow you to run quite decent models locally on consumer h…

Let's be honest that he's probably not doing it due to goodness of his heart. He's most likely trying to commoditize the models so he can sell their complement. It's a strategy Joel Spolsky had talked about in the past (for those of you who remember who that is). I'm not sure what the complement of AI models is that Meta can sell exactly, so maybe it's not a good strategy but I'm certain it's a strategy of some sort

You lead with a command to be honest and then immediately speculate on private unknowable motivations and then attribute, without evidence, his decision to a strategy you can't describe.

What is this? Someone said something nice, and you need to "restore balance"

Re: Meta Llama 3

#520
post #112

I was curious how the numbers compare to GPT-4 in the paid ChatGPT Plus, since they don't compare directly themselves. Llama 3 8B Llama 3 70B GPT-4 MMLU 68.4 82.0 86.5 GPQA 34.2 39.5 49.1 MATH 30.0 50.4 72.2 HumanEval 62.2 81.7 87.6 DROP 58.4 79.7 85.4 Note that the free version of ChatGPT that most people use is based on GPT-3.5 which is much worse than GPT-4. I haven't found comprehensive eval numbers for the lates…

The bottom of https://ai.meta.com/blog/meta-llama-3/ has in-progress results for the 400B model as well. Looks like it's not quite there yet. Llama 3 400B Base / Instruct MMLU 84.8 86.1 GPQA - 48.0 MATH - 57.8 HumanEval - 84.1 DROP 83.5 -

This is tantalizingly close in multiple benchmarks though. Pretty sure this one will finally be the open GPT-4 match.
Post reply on HN