Deepseek referred here seems to be v3, not r1. While the linked article didn’t seem to have info on parameter size, fact that they state it is sparse MoE architecture should suggest it is capable to run pretty quick (compared to other models of similar parameter space), so that’s cool.
Depending on what you want to do with it, and what your expectations are; the smaller distilled versions could work on CPU, but most likely will need extra help on top, just like other similar sized models.
This being a reasoning model, you might get a more well thought out results out of it, but at the end of the day, smaller parameter space (easiest to think as ‘less vocabulary’), smaller capabilities.
If you just want something to very quickly chat back and forth with on a CPU, try IBM’s granite3.1-moe:3b, which is very fast even on a modern CPU, but doesn’t really excel in complex problems without additional support (ie: RAG or tool use).
Everything eventually dies off, or transforms into something not serving our needs and the legacy version dies off; free, paid, proprietary or open source, doesn’t matter. The only thing we can do is position ourselves in such a way that when it happens, not if, we are ready to take what we’d need to the next solution that will serve our needs.
Sure. But the capacitors in the devices do make a pop and the fragments/shrapnels from the damaged devices depart from their physical location at pace that I would not be comfortable with.
If I’m dealing with a spicy pillow situation, the technical definitions as to whether or not something counts as an explosion is the last of my concern.
Most portable electronics today use some variation of lithium ion batteries, which when it becomes unstable can combust/explode if mishandled. However, devices generally have thermal management software and hardware, as well as multitude of other safety mechanisms like power management systems to handle charge regulation. Unless you intentionally puncture your batteries, they’re not likely to cause any problems on their own.
I did in fact read the paper before my reply. I’d recommend considering the participants pool — this is a very common problem in most academic research, but is very relevant given the argument you’re claiming — with vast majority of the participants being students (over 60% if memory serves; I’m on mobile currently and can’t go back to read easily) and most of which being undergraduate students with very limited exposure to actual dev work. They are then prompted to, quite literally as the first question, produce code for asymmetrical encryption and deception.
Seasoned developers know not to implement their own encryption because it is a very challenging space; this is similar to polling undergraduate students to conduct brain surgery and expect them to know what to look for.
Completely agree with you on the news vs science aspect. At the same time, it is worth considering that not all science researches are evergreen… I know this all too well; as a UX researcher in the late 2000s / early 2010s studying mobile UX/UI, most of the stuff our lab has done was basically irrelevant the year after they were published. Yet, the lab preserved and continues to conduct studies and add incremental knowledge to the field. At the pace generative AI/LLMs are progressing, studies against commercially available models in 2023 is largely irrelevant in the space we are in, and while updated studies are still important, I feel older articles doesn’t shine an appropriate light on the subject in this context.
A lot of words to say that despite the linked article being a scientific research, since the article is dropped here without context nor any leading discussion, it leans more towards the news spectrum, and gives off the impression that OP just want to leverage the headline to strike emotion and reinforce peoples’ believes on outdated information.
While I agree “they should be doing these studies continuously” point of view, I think the bigger red flag here is that with the advancements of AI, a study published in 2023 (meaning the experiment was done much earlier) is deeply irrelevant today in late 2024. It feels misleading and disingenuous to be sharing this today.
You don’t always have a choice as it is dictated by the service provider, but whenever possible, disable SMS based MFA and enable TOTP or something else. SMS based MFA is susceptible to SS7 MitM attack.
Linguistic question: is it misogyny if it originates from women? Reason for asking is because I genuinely don’t know if it is like racism against own race kind of situation, and the article appears to have been written by two women.
Edit: lol Lemmy showing their true colors. Would rather dodge and avoid the hard questions, downvote and continue to circle jerk themselves about anti-AI. Love it. Keep it up Lemmy!
What’s that joke? Think of how stupid the average person is, and realize half of them are stupider than that?
Same idea here.
You’d find about half of people whose creativity level being lower than the “average” (technically, mean). If Gen AI is learnt from the totality of our collective knowledge, it should help those on the lower half of the curve much more than those above the curve. However, since Gen AI itself is not able to create new concepts, the collective end up creating more of the same stuff that Gen AI is regurgitating from its training material.
I don’t think this is necessarily a bad thing. This doesn’t apply only to creativity but all spectrum of general knowledge, and should help with raising equity and equality for the humanity at large.
A couple of years is a life time in tech, but despite that, I think the one thing that should be the deciding factor is if you’re actually going to need the space in the mean time. If not, waiting won’t make a difference. On the flip side, if you’re going to need it in the next couple of years anyway, then it might be easier to recognize that $320 over 2 years is less than $0.50/day… taking the initial hit and take advantage of it earlier will probably work out great.
My understanding is that the “host” instance of the community is responsible in sending things out to the subscribers. As such, when users from B post on a community on C, C will broadcast the activity of user from B to instance A’s federated copy of community on C, and users on A will see the post from the user from B. They should also be able reply to each other’s comments on the community on C.
The amount of people who would pay is going to be near zero in the grand scheme of things.
Next time you’re anywhere where you could discretely look at people’s phones, see how many of them run apps with ads. Most apps will offer very cheap IAP to remove ads, but people choose to not pay it. Vast majority of the users have already decided that their time wasted on ads are worth less than whatever tiny monetary cost it would be to remove them. Same thing here: Vast majority of the users have already decided they’re not going to pay to get rid of the ads. This in turn means due to how few people who would be willing to pay, it is not going to be nearly sufficient to keep the infrastructure required up and running, as well as keep the creators compensated for creating the content.
Japan has nicovideo.jp as well. Russia has Yandex Efir (gone through a couple rebrands, Efir was the name in 2020 when we were discussing deals; it was operating under another name prior, and I think it is superseded by dzen). Off to the side I think vK also has a small video delivery presence like how Facebook has videos in their feeds. China has several platforms: Tencent Video (owned by Tencent), Youku as you’ve called out (owned by Alibaba), XiGua (ByteDance), Haokan (Baidu), and then slew of smaller ones like KuaiShou, BiliBili and that video thing WeChat tries to push. None of these are public service operated by the State, by the way. List really goes on… and I’d know, because I’ve worked in the space for almost 12 years now.
China’s great firewall aside, all these platforms are tiny in comparison, and in the grand scheme of things, and barely have any reach. In general, these regional are all taking a backseat just like Nebula and alike — if creators’ content are hyperlocal/super niche, they might be okay with smaller regional platforms; but if they’re trying to extend their reach and monetization (to ensure they have money to continue producing content), the creators’ presence on these platforms are really just auxiliary to their primary presence on YouTube.
Getting viewers to these smaller platforms is going to pose a significant chicken or the egg problem — creators aren’t incentivized to be there because lack of viewer, viewers aren’t incentivized to go there because lack of content. Worse yet, I’ve also seen situations where creators are paid for some period of exclusivity and then when the deal lapses they just go straight back to YouTube.
Real competitors do not exist, and likely will not exist for the foreseeable future. YouTube is the million pound behemoth when everyone else barely registers on the radar.
That’s a drop in the pond in the grand scheme of things. You just out source that out to rights management companies and absolve yourself from that obligation behind safe harbour. This is basically what they’re doing in this department. They’ve built Content ID for digital finger printing, and then invented an entire market for rights management companies on both sides of the equation.
On the other hand, 500 hours of video footage got uploaded to YouTube every minute per YouTube in 2022 (pdf warning). 30 minutes of video game content (compresses better), just the 720p variant using avc1 codec is about 443MB of space. Never mind all the other transcodes or higher bitrates. So say 800MB per hour of 720p content; 500 hours of content per minute means 400GB of disk space requirement, per minute; 500TB of disk space per day.
That’s just video uploaded to YouTube. I don’t even know how much is being watched regularly, but even if we assume at least one view per video, that’s 500TB of bandwidth in and then 500TB of bandwidth out per day.
Good luck scaling that on public budget.
Good luck getting that through the system… the cost to run something like YouTube is… well, let’s just say the lack of real competitions speaks volumes.
Here we observe a pro gatekeeper in their natural habitat…
Self driving cars need to convince regulators that they’re safe enough, even if assuming they master the tech.
LLMs has already convinced our bosses that we are expendable, and can drastically reduce cost centres for their next earnings call.
Yep! Give granite a try. I think that would be perfect for this use case both in terms of able to answer your queries and doing them quickly, without a GPU by just using modern CPU. I was getting above 30 tokens per second on my 10th gen i5, which kind of blew my mind.
Thinking models like r1 will be better at things like troubleshooting a faulty furnace, or user problems, so there’s benefits in pushing those envelopes. However, if all you need is to give basic instructions, have it infer your intent, and finally perform the desired tasks, then smaller mixture of experts models should be passable even without a GPU.