I am not sure if relevant, but I think there might also be a legal risk if Valve goes harder on games. Valve not contributing meaningfully to "first party" games makes them less susceptible.
1,008 karma · joined July 29, 2014
If we are ok with breaking honor rules in otherwise healthy communities because there is no legal risk, then why are we not OK with people voicing that it was a dick move?
Btw, I don't play deadlock. Will probably never play it because I had enough of games like TF2 or overwatch. I just don't get why people are so comfortable with being jerks lol. Especially as the first comment on this post, commenting on the drama and insinuating that gamers are being childish (they are usually) on this matter is kinda silly.
Like dota 1, you could always play offline 5v5 as the only human among bots.
Oh i guess there was also LAN in wciii tft.
Valve is a weird company, and gamers can be weird people. DotA had a culture where doxxing icefrog is really looked down upon. Valve games tend to have these sorts of cultures and quirks.
Needless to say, valve created a relatively accessible way to play their game under development. Most companies don't do this. All Valve requested was to not publicize the game. If you are a gaming journalist, and you break this sort of online etiquette, which is very well established thing for Valve, then you get to be called out for being a dick.
> they chose not to in this instance
because as you said it is kinda silly and the game is an open secret. Doing this is not part of their culture, but the openness accessibility of their betas is.
> a popular new game by Valve is properly newsworthy to report upon
sure, verge has good traffic reasons to do this. gamers also have a reason to criticize this because they want to keep the way Valve does things. It is very unique.
> most don't realize that the and "Valve banned me" update is tongue-in-cheek and not the author being genuinely upset
What makes you say this? I think most people are just using this line to mock the author because they are angry nerds lol. Doesn't really matter what the intent of the author was.
So much of online gaming culture would be understood with just this line:
"If you don't wanna get flamed, don't break the ritual"
This article elucidates the point: https://www.forbes.com/sites/paultassi/2024/08/13/the-verge-...
The difference between a lot of hackers now and PG isnt talent, it is that someone was making a good website when the internet pretty much was empty lol. Google was still a baby compared to yahoo.
edit: since i was pretty negative here, i will sing some praise.
Graham is incredibly technologically capable, knows what problems are worth investing in, and is great at leading people. Those qualities are great, and he is a world star at it.
Excluding the part in the middle because I don't wanna repost potential issues for you. I just wanted to comment that that is terrible. People often talk about the siloed nature of research in industry, without considering that academia supports the draconian publishing system. I understand IP protection, but IP protection doesn't have to mean no access. This is such a huge issue in the bio- world (biostats, genetics, etc).
If you are vague enough, you can say you did something.
If you don't explicitly mention the issue(s), you don't have to mention what you changed, if anything.
Classic M$.
But in a sense, the 300 lines of Llama code are essentially just lines of math. And reading through any math proof will show you that any particular line can hide large amounts of complexity.
This can be true with code with more tedious operations, but those lines are a smaller fraction of the overall code base by definition.
Even the "tedious" parts of the llama code can hide large complexity. Setting a learning rate with a schedule might require reading a paper or two for your particular architecture.
But yes, once you parse all the math and the theory, the lines are kinda simple matmul and forward lol.
Can you explain this statement more if you have time? Are you saying the recurrent architecture of xLSTM enables fast inference on par with Mamba? Or the xLSTM architecture slows it down so that its inference is as slow as mamba?
Sure.
> It appears that most of the handwringing is about politically disagreeing with the bans, not hardships of obeying bans
That is not true. The ban list is overly broad. The hobbit is banned in some locations. Harry potter and asoiaf... Maybe high schoolers and middle schoolers shouldn't be reading that stuff. Multiple teachers in multiple districts have been disciplined over books they did not expect. It is far more sustainable for teachers to avoid this issue.
> But they want this material (sexuality, gender, "race theory") available even to pretty young children because of their societal goals/agenda
I am not going to judge this statement's veracity. Let's take it as true. As I mentioned before, the issue would then be that they are also banning other books that have nothing to do with this, and that still makes the jobs of teachers difficult. It is not feasible to have this much overhead on book bans because of political battles and also expect teachers to manage this overhead.
Thanks for sharing your first principles argument. If you have the time, please share some numbers.
Maybe? This is why I responded to:
> It's everyone's goal. Apple is actively achieving that goal
This is is the issue I found disagreeable. Other organizations and individual people are achieving that goal too. Google says GPT-Nano is going to device, and if the benchmarks are to be believed, if it runs at that level, their work so far is also actively achieving that goal. Meta has released multiple distilled models that people have already proven to run inference at the device level. It cannot be argued that meta is not actively achieving that goal either. They don't have to release the hardware because they went a different route. I applaud Apple for the M chips. They are super cool. People are still working on using them so Apple can realize that goal too.
So when you go to the statement that started this
> Apple's AI strategy is to put inference (and longer term even learning) on edge devices
Multiple orgs also share this. And I can't say that one particular org is super ahead of the others. And I can't elevate apple in that race because it is not clear that they are truly privacy-focused or that they will keep APIs open.
> You cannot effectively do this on competitor hardware, with good performance, from "budget" to "Pro" lineup, which is a requirement of the goal
Why do you say you cannot do this with good performance? How many tokens do you want for a device? Is 30T/s enough? You can do that on laptops running small mixtral.
> What hardware are they running it on? Are they taking advantage of Apple (or other) hardware in their strategy?
I don't know. I have nothing indicating necessarily apple or nvidia or otherwise. Do you?
> [Regarding the rest]
Sure, my point is that they definitely have an intent for bespoke models. And why I raised the point that not all computation will be feasible on edge for the time being. My point with what raised this particular line of inquiry is whether a pure edge experience truly enables the best user experience. And also why I raised the point about Apple's track record of open APIs. Which is why "actively achieving" is something that I put doubt on. And I also cast doubt on apple being privacy focused. Just emphasize tying it back to the reason I even commented.
It absolutely is. Model distillation will still be pertinent. And so will be parameter efficient tuning for edge training. I cannot emphasize more how important this is. You will need your own set of weights. If apple wants to use open weights, then sure. Ignore this. Don't seem like they want to long-term... And even if they use open weights, they will still be behind other companies have done model distillation and federated learning for years.
> Why is a distinct model required?
Ask apple's newly poached AI hires this question. Doesn't seem like you would take an answer from me.
> If others are making models, with no way to actually run them
Is this the case? People have been running distilled llamas on rPis with pretty good throughput.
So are other companies, with their many year strategy of actually building models that accessible to the public.
yet Apple is "actively" achieving the goal without any distinct models.
But I guess in their words:
> the teacher's goal is mainly to push up against someone's line, politically
Read what their engineers say in public. Unless I hallucinated years of federated learning.
Also apple isn't even a player yet and everyone is discussing how they are moving stuff to the edge lol. Can't critique companies for not being on the edge yet when apple doesn't have anything out there.
Teacher unions list book bans as one of the primary reasons for leaving. They have data and testimony backing that up. We see legislation and organizations all over America.
Here is a quick article describing the number of books being banned and the effects it has on teachers [1]. It list numbers. There are numerous articles all over the internet from well respected organizations like the NYT saying the same stuff.
The numbers are already presented by them. We see the teachers leaving. We know education is suffering from systematic national pressures from both political sides. Feel free to look it up. hell on first principles, book banning has a direct effect on libraries teachers can have when they include such classics. Again, what numbers do you have? Or just first principles logic to dismiss them at all?
[1] : https://www.nea.org/nea-today/all-news-articles/educators-fi...
> Apple's AI strategy is to put inference (and longer term even learning) on edge devices
This is pretty much everyone's strategy. Model distillation is huge because of this. This goes in line with federated learning. This goes in line with model pruning too. And parameter efficient tuning and fine tuning and prompt learning etc.
> This is completely coherent with their privacy-first strategy
Apple's marketing for their current approach is privacy-first. They are not privacy first. If they were privacy first, you would not be able to use app tracking data on their first party ad platform. They shut it off for everyone else but themselves. Apple's approach is walled garden first.
> Processing data at the edge also makes for the best possible user experience because of the complete independence of network connectivity
as long as you don't depend on graph centric problems where keeping a local copy of that graph is prohibitive. Graph problems will become more common. Not sure if this is a problem for apple though. I am just commenting in general.
> If (and that's a big if) they keep their APIs open to run any kind of AI workload on their chips
Apple does not have a good track record of this; they are quite antagonistic when it comes to this topic. Gaming on apple was dead for nearly a decade (and pretty much still is) because steve jobs did not want people gaming on macs. Apple has eased up on this, but it very much seems that if they want you to use their devices (not yours) in a certain way, then they make it expensive to do anything else.
Tbf, I don't blame apple for any of this. It is their strategy. Whether it works or not, it doesn't matter. I just found this comment really odd since it almost seemed like evangelism.
edit: weird to praise apple for on device training when it is not publicly known if they have trained any substantial model even on cloud.
I have quite a few teacher friends in Georgia and NYC, and I can tell you that this is the case for them. Organizations that represent teachers have said so themselves. Hard evidence is best though. Do you have numbers to dismiss their claims?
Sure. The next step is helping them making a decision on which book to read and provide a social environment for doing that. Libraries do not really do that and haven't done that even in the early 90s.
And people still find ways to infer race.