Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Meta Llama 3

llama.meta.com

Post reply on HN
Print view View original

841–850 of 965 posts

Re: Meta Llama 3

#841

Post by glenstein »

gliched_robot wrote: post #721
salesynerd wrote: post #650

Earlier quoted context omitted.

Strange. The Llama 3 model card mentions that the knowledge cutoff dates are March 2023 for the 8B version and December 2023 for the 70B version ( https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md )

Maybe a typo?

I suppose it could be hallucinations about itself.

I suppose it's perfectly fair for large language models not necessarily to know these things, but as far as manual fine tuning, I think it would be reasonable to build models that are capable of answering questions about which model they are, their training date, their number of training parameters, and how they are different from other models, etc. Seems like it would be helpful for it to know and not have to try to do its best guess and potentially hallucinate. Although in my experience Llama 3 seemed to know what it was, but generally speaking it seems like this is not necessarily always the case.

Re: Meta Llama 3

#842

Post by DresdenNick »

8n4vidtmkvmk wrote: post #655
geepytee wrote: post #339

Earlier quoted context omitted.

Also added Llama 3 70B to our coding copilot https://www.double.bot if anyone wants to try it for coding within their IDE and not just chat in the console

Can we stop referring to VS Code as "their IDE"? Do you support any other editors? If the list is small, just name them. Not everyone uses or likes VS Code.

Done. Anything else?

Re: Meta Llama 3

#843

Post by cornholio »

baby wrote: post #832
cornholio wrote: post #786

Earlier quoted context omitted.

The only thing the Reels algorithm is showing me are videos of ladies with fat butts. Now, I must admit, I may have clicked once on such a video. Should I now be damned to spend an eternity in ass hell?

It’s easy to populate your feed with things you specifically want to watch: watch the stuff you’re interested in and swipe on the things that don’t interest you.

Reels don't interest me, they are just showed in my face whenever I use Facebook (or should I say Face-butt?). It's impossible to hide without using a custom script/adblock, which I ended up doing, but the only long term, cross device solution is to simply to delete the Facebook account.

Re: Meta Llama 3

#844

Post by glenstein »

bigstrat2003 wrote: post #482
bearjaws wrote: post #423

Earlier quoted context omitted.

I feel like Lex has gone full 'both sides' at this point, waiting for him to have Alex Jones on at this point. There is no real commentary to pull from his interviews, at best you get some interesting stories but not the truth.

That is a strength, not a weakness. It's valuable to see why people, even those with whom we disagree, think the way they do. There's already far too much of a tendency to expel heretics in today's society, so the fact that Lex just patiently listens to people is a breath of fresh air.

>That is a strength, not a weakness

The trouble is self-styled "both sides" types believe that since they take the both sides approach, they have insulated themselves from the kinds of politicization that compromises the extremes. But the manner in which you position yourself relative to those extremes is every bit as politicized and every bit as liable to the same cognitive biases and rationalizations.

Misinformed climate skeptics often regard themselves in this way, as not taking one side or the other on global warming. They mistakenly believe that this orientation has elevated them above equivalently offensive extremes, but in truth they have compromised their own media literacy by orienting themselves in that manner.

There are numerous instances of this all over the political spectrum, Cornell West talking to left-wing academics in left-wing academic language about how "nobody" thinks Obama is truly left-wing. Journalists during the Iraq war had a both sides approach that cashed out as extremely hawkish and apologetic in defense of the Iraq war.

The Lex Friedman version is a "centrist" in a specific kind of media environment that lends disproportionate visibility towards its own set of boutique topics. The combination of optimism about technology and trends especially around AI and crypto and some libertarian leaning politics surrounding it, which at its periphery finds itself disproportionately saturated by right-wing memeing and politics. And so it's a form of centerism that's in the center of a world as described by those things. But for him and his viewers it's something they consider a perfectly neutral state of nature that's free of any adornment of ideology.

Re: Meta Llama 3

#845

Post by naasking »

cornholio wrote: post #784
typpo wrote: post #142

Public benchmarks are broadly indicative, but devs really should run custom benchmarks on their own use cases. Replicate created a Llama 3 API [0] very quickly. This can be used to run simple benchmarks with promptfoo [1] comparing Llama 3 vs Mixtral, GPT, Claude, and others: prompts: - 'Answer this programming question concisely: {{ask}}' providers: - replicate:meta/meta-llama-3-8b-instruct - replicate:meta/meta-lla…

I'd be vary wary using tests that can are likely already solved in their entirety somewhere online. A good test is, for example: solve a second degree equation for some random values of a, b and c. It's a very simple algorithm that all models have memorized, much simpler than a pong implementation, yet none of them are able to apply it and make silly mistakes, then lie their way out. Here's what LLAMA 3 had to say ab…

Have you used any of the prompt modifiers that tend to improve accuracy, like chain of thought, review last output for errors, etc.?

Re: Meta Llama 3

#846

Post by lannisterstark »

xpe wrote: post #681
eigenvalue wrote: post #231

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…

> but also to not use pessimistic AI "doomerism" as an excuse to hide the crown jewels and put it behind a centralized API with a gatekeeper because of "AI safety risks." AI safety risk is substantial. It is also testable. (There are prediction markets on it, for example.) Of course, some companies may latch onto various valid arguments for insincere reasons. I'd challenge everyone to closely compare ideas such as "o…

I too like making up hypothetical insane scenarios in my head. The difference is that they stay with me in the shower.

Re: Meta Llama 3

#847

Post by lannisterstark »

asadotzler wrote: post #678
999900000999 wrote: post #393

Earlier quoted context omitted.

The Quest is the top selling VR headset by a very large margin. He's well positioned to take that market when it eventually matures a bit. Once the tech gets there, say in a decade we might see most people primarily consume content via VR and phones. That's movies, games, TV, sporting events, concerts.

Give me $50 billion dollars and I'll bet I could get 8 million MAU on a headset. It's a massive failure because Zuck's a nerd and not a product guy.

Asking for an impossible hypothetical and then claiming something equally impossible. stay classy hackernews. Chances are that you would take the 8 million and run.

Re: Meta Llama 3

#848

Post by lannisterstark »

asadotzler wrote: post #677
freedomben wrote: post #351

Earlier quoted context omitted.

It's a bit too early IMHO to declare the metaverse a failure. But that said, I don't think it matters. I don't know anybody who hasn't been wrong about something , or made a bad bet at times. Even if he is wrong about everything else (which he's not, because plenty of important open source has come out of facebook), that doesn't change the extreme importance that is Llama and Meta's willingness to open things up. It'…

$50B for <10M MAU is absolutely a failure, today, as I'm typing this.

You're everywhere in this thread man. Did zuck steal your lunch or something?

Re: Meta Llama 3

#849

Post by lannisterstark »

RamblingCTO wrote: post #763
tmalsburg2 wrote: post #324

Earlier quoted context omitted.

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?

I swear, this feels like people get paid to write positive stuff about him? Have you forgotten his shitty leadership and practices around data and lock-ins?

Yes how dare different people have different opinions about different people? It's almost as if we all should be a monolithic voice that agrees with you.

Re: Meta Llama 3

#850

Post by lannisterstark »

nojs wrote: post #522
modeless wrote: post #237

Earlier quoted context omitted.

Seems like a year or two of MMA has done way more for his charisma than whatever media training he's done over the years. He's a lot more natural in interviews now.

Alternatively, he’s completely relaxed here because he knows what he’s doing is genuinely good and people will support it. That’s gotta be a lot less stressful than, say, a senate hearing.

You'd be surprised with some of the reee comments you're getting here.
Post reply on HN