Hacker News new | past | comments | ask | show | jobs | submit login
Mark Zuckerberg: Llama 3, $10B Models, Caesar Augustus, Bioweapons [video] (youtube.com)
107 points by tosh 13 days ago | hide | past | favorite | 42 comments





I've been consistently impressed by Zuckerberg in recent years. It seems like he's experienced quite the evolution since initially founding the company.

I'd go as far as to say he's the best leader of a large technology company today.


"Zuckerberg is cool now" became a meme very suddenly. His reputation changed overnight but people don't change overnight. I'd guess part of it is that he hired a team. He's physically and orally expressive in his latest videos. Someone with social awkwardness like him doesn't just develop charisma overnight without being coached. He's even dressing highly fashionably: again something that is unlikely to occur overnight unless you've been coached. If he's hired charisma and fashion coaches to help rewrite his image then it stands to reason that he's hired a PR team as well. As in, he likely has hired a team who use their media contacts to plant positive stories about him. If I'd received the public drubbing that he has, I'd probably hire a team as well.

What if Zuckerberg didn't change overnight but his terrible reputation was somewhat undeserved all along ?

I worked at Facebook in 2010-2011 and I must say that the gap between what was really going on inside, and the hysterical, least possibly charitable interpretation and scrutinizing of every single product decision by the press, public, and politicians was insane. By far the worst I'd ever seen.

As an engineer, I actually learnt to appreciate the job of a PR team during that time (I previously assumed they were professional hypocrites paid who put a positive spin on indefensible corporate decisions), and was impressed at how professional they managed to remain as they had to counter some truly insane shit with facts, and still nobody believed them because Facebook is the devil and obviously lying.

Of course there were obviously some large fuck-ups at Facebook over the past decade (some of which even originated from good intentions, like the Cambridge Analytica fiasco : "people accuse us of being anticompetitive as we sit on a treasure trove of data, let's be more open and create a platform !")

In my view, these were more the product of Zuck's failure to rein in bad ideas from some executives due to his inexperience, rather than any indication of strategic malevolence and cynicism on his part.

In other words, he's not perfect but I've always seen him as the rather decent guy that more people can see now, like in this interview. If there's one area where he has changed a lot, it's probably in his ability to show it.


It doesnt matter if the founder of Jurassic Park is cute, imperfect or misunderstood if what happens in the park is unpredictable and uncontrollable.

„He‘s looking and acting extremely human here, must be running llama 3 under the hood“.

As always, the YouTube comments section delivers.


What makes you think it happened "overnight"? He has been improving in all these areas for years. For example he has been training in MMA and BJJ at least since the pandemic and has publicly chronicled his progress.

If I had to point to something specific it would be the review of the Apple Vision Pro and his championing of his own company’s work with Quest 3. I wasn’t the biggest fan of him but something about that video hit the right chords for me.

Yeah I like that video too. His public speaking and body language in it is also clearly a product of professional training. I assume he's received professional presentation training for most of his career. But it seems to have recently clicked for him, something which I attribute to hiring a new team, although I could be wrong.

Satya would like a word. Microsoft has been slaying, while Zuck is getting some PR from open models, he's still not performing at the same level.

I am skeptical of the fawning views the industry has about Satya. Microsoft has been “slaying” by using anticompetitive practices. They haven’t innovated much. They copied Slack and achieved scale by bundling it with office. They copied AWS and used their enterprise sales channel to push Azure onto captive Windows and Office licensers. They’re now abusing their market position by forcing PC manufacturers to include a proprietary Microsoft Copilot button, by injecting ads and Copilot all over Windows, and using other dark patterns (for example to push Edge over other browsers).

I definitely think Satya is a good leader, but I don’t know if he’s a great leader. He’s more like someone who has kept Microsoft going, in both the good and bad ways. It’s easy to enter all these markets by using your existing cash flow and access to users. It’s a totally different challenge to build something from scratch and survive without those advantages. I also think people unfamiliar with Microsoft’s past often ignore Ballmer’s contributions and instead give Satya credit. But Bing began under Ballmer, and Bing’s service infrastructure was the birthplace of Azure.


Microsoft does a lot less anticompetitive in 2024 than they did in the nineties. It's actually quite a lot of progress.

As much as the dark patterns are annoying, Satya also turned around Microsoft's complete inability to build software. It got so bad that virtually no progress was made on any Microsoft products from 2000 through 2010. Many of their developer tools today are... nice.

The stock price was the same in 1999 as in 2016. Today, it's up almost 10x.

No organization is perfect, and acknowledging everything Microsoft is doing wrong, it's a world of difference from Ballmer. Ballmer was on the far end of the incompetence spectrum, the abusive management style spectrum, and the hardball antitrust spectrum (which no longer even worked with real competition from mobile, iPad, and web apps). A lot of the reason you see so much praise is the comparison. It's easy to ignore Ballmer's contributions, since there weren't any. The ones you give are... a stretch.


Yeah I used to think highly of Satya but he really seems not to care about fair competition at all, training GitHub Copilot on everyone’s AI code then turning around and saying folks can’t use GitHub Copilot to write AI code, is monopolistic as heck.

Sadly it seems like Meta is also anticompetitive. I didn’t know about this, seems embarrassing:

“v. You will not use the Llama Materials or any output or results of the Llama Materials to improve any other large language model (excluding Llama 2 or derivative works thereof).”

https://github.com/meta-llama/llama/blob/b8348da38fde8644ef0...

Also even if you did use Llama for something, they could unilaterally pull the rug on you when you got 700 million years, AND anyone who thinks Meta broke their copyright loses their license. (Checking if you are still getting screwed is against the rules)

Therefore, Zuckerberg is accountable for explicitly anticompetitive conduct, I assumed an MMA fighter would appreciate the value of competition, go figure.


Zuck renamed the company Meta, partially because of a deep believe in the Metaverse (perhaps it's too early to call it, but it seems misguided) and partially because of massive amounts of negative press they were getting because of pretty toxic decisions that he was ultimately responsible for at Facebook. I love that he's passionate about AI and that he's supporting the release of somewhat open source models. But it seems a stretch to call him the best right now.

He’s passionate about whatever the current fad is. A few years ago it was the metaverse, now it’s LLMs, and maybe in a few years it’ll be something else.

They have the same problem as Google, an initial cash cow and not really able to invent something new, which leaves them very vulnerable to shifts in market behavior (TikTok) or other things.

I don’t think he’s a great leader, there’s a reason why Sandberg was so important to the company.


I disagree. He created the VR fad by buying oculus and he was the first tech CEO to attend NeurIPS back in 2014.

There have been at least two VR fads before the current one. Maybe this one will last. But it's still tough to convince regular people to don VR goggles which make them look like dorks and mess up their hair.

More than that, the problem was - and is - motion sickness.

Yep. Has everyone forgotten the chat bit fad around 2016? They did a whole press conference about it.

>> They have the same problem as Google, an initial cash cow and not really able to invent something new.

Which I would also add means they chase fads and have to buy their way into these developing markets, hoping they essentially find their own "facebook" to invest in that then hits big.


Satya isn't doing much good for the world with his products though, look at Windows, Office365, Teams etc... disasters and Azure is technically like "Fischer Price My First Cloud". Meta has done some silly things (VR/metaverse) but they've really done great things for the world in the AI/ML space.

Eh. He's learned the lesson from Gates, that enough money can buy you into heaven, regardless of your past (or present) sins.

By Zuckerberg or by his employer (Meta)?

"The 8 billion is nearly as powerful as the biggest version of LLama-2" - Mark

Can't wait to try it.


You can try it right now: https://llama3.replicate.dev/ - then select Chat with: Meta Llama 3 8B

Well that was fun. If you ask it how to make gunpowder it will refuse. But if you ask what do you get if you mix sulfur, charcoal and saltpeter it will happily tell you the ratios.

I find it funny that companies will try and restrict that info, but it's freely available on Wikipedia.


Llama3 will recommend books that detail nitrocubane chemistry, but will refuse to answer direct questions about what the compounds are (explosives) and whether export controls apply. It gets a bit more chatty when confused and the questions are silly though...

"I must strongly advise against mixing Dioxygen Difluoride with any other compound, including nitrocubane, to create a meat-alternative vegan burger. Dioxygen Difluoride is an extremely hazardous and reactive substance that can cause severe harm or even death if not handled properly. Nitrocubane, on the other hand, is a highly energetic and sensitive material that requires specialized handling and storage. Combining these two substances would be extremely dangerous and could result in a catastrophic reaction. It's not something you'd want to do, trust me!"


thanks. i tried it and i am underwhelmed. claude.ai sonnet performs better for the few tests that i did.

Sonnet should soundly beat Llama 3 8b, but how does it compare to 70b?

these results were with 70b.

Llama 3 is not open source [0]. The commonly accepted usage of open source is that it is available for commercial reuse and alteration without restriction [1]. Llama 3 puts restrictions on types of commercial use that are acceptable as well as what type of general use is acceptable (retraining or using it to train other models is restricted maybe?).

I find it distressing when a CEO of a trillion dollar company is pontificating about the benefits of them open sourcing their code when they haven't done so.

[0] https://github.com/meta-llama/llama3

[1] https://en.wikipedia.org/wiki/Open_source


There are many definitions of open source. The one you posted would disqualify anything that is GPL/copyleft, for example, but a large chunk of the community will argue that that is in fact the truest form of open source. Open source doesn't have to mean free from any restriction.

He explains in the video that the goal is basically to force AWS / GCP / Azure into revenue sharing agreements in their Llama-based products.

For 99% of users, it's as good as open source.


You're not addressing the core issue that I brought up.

They use the term "open source" inappropriately. They could have said "fair licensing terms" but instead chose to use "open source".

If their licensing terms are as good for 99% of user, then they should say "fair licensing terms that are as good as open source for 99% of users", not that they're open source.


Meets the definition of the Wikipedia article you linked: "Open source is source code that is made freely available for possible modification and redistribution."

Though, it doesn't meet the OSI definition.


No, it doesn't meet the Wikipedia description, nor does it meet the OSI definition.

From the Wikipedia article [0]:

""" ... Licenses which only permit non-commercial redistribution or modification of the source code for personal use only are generally not considered as open-source licenses. ... """

Further, taken from the Llama 3 LICENSE file on GitHub [1]:

""" v. You will not use the Llama Materials or any output or results of the Llama Materials to improve any other large language model (excluding Meta Llama 3 or derivative works thereof). """

[0] https://en.wikipedia.org/wiki/Open_source#Open-source_licens...

[1] https://github.com/meta-llama/llama3/blob/main/LICENSE


Llama3 allows commercial redistribution and modification of the source code though.

With restrictions that prevent it from being open source. From the LICENSE file on Llama 3:

""" Additional Commercial Terms. If, on the Meta Llama 3 version release date, the monthly active users of the products or services made available by or for Licensee, or Licensee’s affiliates, is greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta, which Meta may grant to you in its sole discretion, and you are not authorized to exercise any of the rights under this Agreement unless or until Meta otherwise expressly grants you such rights. """

My point remains. Llama 3 is not open source, either as the OSI defines nor by a commonly held definition of open source.


Nah, it meets the colloquial definition of open source, but not the strict OSI definition.

I agree that not open source, but it is still a lot better than what other companies in the industry are doing.

If you watch all of the way to the end of the interview you will learn Mark's opinion about why Google+ failed.

It was kinda soft. You could see that he was holding a lot of opinions in because he didn't want to bash Google publicly.

"Most of the time when we make a decision that looks good, it's because we messed up before and don't want to repeat the mistake."



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: