I do have a sneaking suspicion it's a mix of things, I'm at the point now that I do get a bit of an odd feeling whenever I'm reading AI produced content
I suspect over time it will get good enough that I'll need a larger sample size to identify it, however that won't solve what I think of as the "why does this need to exist" problem, I've noticed that a fair bit of AI content hits that mark, it can be fun, but I've not experienced that feeling of engaging with something that's been well thought out / executed, maybe we'll hit that point[0], but I suspect it will take a while
Wait, are you arguing that activities that make people happy have little to no value?
I mean, we only spend money when we believe that what we buy with it is more valuable than the money we've spent, so there is some underlying activity or follow up to spending the money that naturally follows otherwise, the buyer would perceive no benefit and would not buy again
So how are you equating collection and enjoyment of your purchase as the same?
Surely they must have value at an individual level, for people are willing to pay to engage in those. The time you spend playing games does very little to help the society around you, and might in fact distract you from more productive activities. Perhaps gamers would go outside instead, and reduce the burden on the healthcare system.
This essentially amounts to government intervention to reduce the already incredibly cheap cost per hour of these activities for a small minority of gamers. Is that good public policy?
Yes, because you will not be limited in time to enjoy a single title. Like it's the case with other media - books, music, movies, etc. This is good because future people (who haven't even been born yet) will be able to enjoy the same slice of entertainment that we enjoy right now. Just like we can still enjoy games made 40 years ago.
But we are interested in benefits first and foremost since we are consumers, i.e. people who use the product. Regarding the costs, nobody dismisses them. But saying that the costs of giving the game to community are tremendous is simply misguided. Developers and publishers have been doing this since the beginning, it's not something new.
The reason why publishers don't do this is also very clear - they want us to play their newer games and spend money there. They are not doing this because it's hard for me. They just want to maximize the profit, that's it.
> But we are interested in benefits first and foremost since we are consumers
We're only consumers when we consume, but that is not the extent of our participation in society. To optimize for consumption is a shortcut to certain doom.
> But saying that the costs of giving the game to community are tremendous is simply misguided
Tremendous, no. Significant? Yes. The cost could very well be in the millions for a single project, but how many hours of gameplay does that actually buy? Do those later hours come at a significantly increased cost?
>The reason why publishers don't do this is also very clear - they want us to play their newer games and spend money there. They are not doing this because it's hard for me. They just want to maximize the profit, that's it.
I get why someone who plays games would want this. But as someone who doesn't play games, why on earth would I want to encourage this? It defies most basic economic theory.
You're suggesting a subsidy for gamers which everyone else would to pay for in the form of slower economic activity and lower tax revenues. I do not want to subsidize your games.
> The cost could very well be in the millions for a single project
How did you come to this number? Something simple like giving the binaries with docs doesn't cost that much money. I'm not sure what you are referring to.
> how many hours of gameplay does that actually buy?
You seem to confuse the cost of continued support with the cost of giving it to the community. Nobody is asking publishers to continue support when they don't want to.
> But as someone who doesn't play games, why on earth would I want to encourage this?
You are not expected to support this if you don't care about it... But don't act surprised when you find out that some people do play games and they do care about them.
> You're suggesting a subsidy for gamers which everyone else would to pay for
>> You're suggesting a subsidy for gamers which everyone else would to pay for
>No, that's not how it works. Not even close.
As far as I can tell, legislation which enables you to play a game in perpetuity instead of for a limited time will likely reduce the amount of tax revenue your gaming-related activities generate.
A useful metric to focus on here is the cost per hour of entertainment. The price of the game itself might stay the same, but the cost of your gaming itself would be subsidized.
> legislation which enables you to play a game in perpetuity instead of for a limited time will likely reduce the amount of tax revenue your gaming-related activities generate
How so? If I'm sure that a game is not going to die, I will be more willing to pay for it. The more such games exist, the more money I will be willing to spend on them. Hence more taxes from me. Otherwise I will stick to older games that don't have this problem, which will bring less taxes from me because of the cost.
> cost of your gaming itself would be subsidized
By whom? I know there are currently tax returns (or exemptions is more correct term probably) for gaming companies. But how does having end of life plans change this?
> A quote that stood out: "Selling out is usually more a matter of buying in. Sell out, and you're really buying into someone else's system of values, rules and rewards."
This quote more than ever seems like taking the road less travelled by in this day and age
The biggest difference that I can see is that keys are not making any claims about ownership, there's no global registry and it's p2p, which is a big upside
Hey, just reading through the docs, this looks like a pretty cool project and I found your p2p chat example[0]
I'm trying to understand it's limitations, if I used this to build a p2p client / server setup or even two peer machines, what else do I need to setup to be able to have connections between the two applications?
For example, could I create an application that runs on my phone and another that runs on my laptop and finally get a direct secured working connection between the two of them? Or is this solving a different problem? =)
Yes, you will get secure direct connections. This matters for privacy in case of an encrypted chat, but also has a lot of benefits for more demanding use cases such as video streaming.
If you use the default setup you are still depending on a tiny bit of cloud infrastructure such as our public relays to faciliate the hole punching. However, we also have optional local discovery using e.g. mDNS.
You'll get a direct connection in most cases, but sometimes it will need to fall back to relays. n0 provides free relays, but ultimately that can get rather expensive. You can also run your own relays for your app.
> I've yet to find a solution that handles all these correctly, let alone having high quality transcriptions.
Wait really? I honestly would have thought this was a solved problem by now, especially high quality transcriptions bit, just out of curiosity, is the problem that the quality isn't high enough?
There are still a few unsolved problems that require tuning for specific applications. Applications that own the video call have a much easier time, they have access to each individual audio stream. Applications like this, however, have to deal with overlapping voices from a single stream. If it's trying to attribute each utterance to an individual, separating the voices is tough, or can lead to confusing transcripts. There are many little problems like this which make it a tough problem in real world usage. Domain specific terms, or proper nouns is another source of inaccuracy.
> Wait really? I honestly would have thought this was a solved problem by now, especially high quality transcriptions bit, just out of curiosity, is the problem that the quality isn't high enough?
If I had to guess, all of those apps are probably vibecoded, hence the variable quality.
In the current climate, I've decided to explore building games, so I'm building a management game about turning chaotic fields of research into fundable products, fitting for Hacker News I think ;)
The concept was, what if Theme Hospital was about Victorian-esq research institution instead of a hospital? You hire strange scientists, have them explore dangerous fields of research, collect messy findings, turn them into theories, prototypes and eventually products, all the while trying to convince investors they're worth funding before they hit the market and work out what they might actually be worth
The gameloop is broken down into two parts, Exploration / Discovery and Exhibition, the closest comparison I have for the first part is take Kerbal Space Program, but focus it on Mission Control rather than the astronauts
While the mad scientists are going into weird, unstable research domains, the player is managing the institution around them, funding, equipment, research direction, safety
On the other side as you discover interesting things or successfully develop prototypes worth showing off, have investors show up and see what excites them, will they give you more funding? Push a grant your way? How are you going to keep this circus going?
You're balancing two plates, you need to invent tools to delve deeper and if you don't keep finding exciting new discoveries, your investors will slowly get bored of you
A question I've got which I've been wondering about, not sure if anyone else has been thinking about it, what actually made Fable so effective?
From what I could tell from the very little time that I had to interact with it, it's instruction following seemed more consistent
The other thing that comes to mind is a lot of people commented on how driven it was, so I'm wondering whether figuring out how to keep existing models looping on task might actually be quite a big shift in capability
Probably just a bigger version of Opus if I had to wager, and Opus is just a bigger version of Sonnet. Maybe some small architectural differences baking in an additional few months of ablation studies/research. But the fundamental driver is new pretrain with larger size. Probably corresponding to when some new generation of GPUs/new datacenter came online rather than any major qualitative breakthrough.
Hints: They created a new label instead of version bumping Opus, they didn't deprecate Opus, and it costs more per token.
Fable had mostly the same pre-training data as Opus, and it's likely they're distilled from the same source. The difference is that it's a larger model with more post training on "dangerous" stuff they didn't want in the core model, and "long" task RL.
Lab folks keep cards close to their chests here, but it's likely Mythos was an earlier teacher model for Opus that got additional cybersec post-training. Whether they have a bigger tier than that is hard to say, labs have been cautiously scaling parameters since the failure of GPT4.1. They 100% have a bigger/better model they haven't released, but that's probably more down to it not being done cooking yet. Once it's done, the single larger model lets them drop new Opus and Mythos iterations in rapid succession.
Googlers have hinted that Gemini 3 came in at 10T, which seems hard to operationalize, Google's flash and pro releases are staggered in a way that doesn't make sense if flash is a pro distill, and there are enough cases where Gemini flash outperforms pro on the same task that I think it's unlikely it's just being distilled from an "in progress" version of pro.
Appreciate the long answer. Why is it more likely that Gemini 3 Pro/Flash/Lite are distillations of the same parent model than that they’re different training runs on the same dataset, with minor version bumps being different post-training setups?
The biggest tell is the fact that labs are staggering smaller model releases so much with big models. If the small models (flash, sonnet/haiku) were being distilled from pro models, you'd consistently see them be released fairly soon after new pro releases to maximize their competitiveness (and this was the case early on for Anthropic). Instead it seems like releases are timed to build/maintain hype.
A thing to keep in mind is that if they release a smaller model halfway between well spaced big model releases, why wait so long on the next big model release if it's sufficiently ready to distill to a smaller model? The ability to demonstrate AI superiority is worth a ton, there's no reason to hold back.
The big AI labs are also accumulating huge datasets of expert work in a wide range of fields, which is very expensive to re-create. It seems pretty plausible that this this gives them a big advantage that is compounded by their larger training runs and larger models.
I suspect over time it will get good enough that I'll need a larger sample size to identify it, however that won't solve what I think of as the "why does this need to exist" problem, I've noticed that a fair bit of AI content hits that mark, it can be fun, but I've not experienced that feeling of engaging with something that's been well thought out / executed, maybe we'll hit that point[0], but I suspect it will take a while
-[0]: https://xkcd.com/810/
reply