First important step is the need to make your application faster.
Second step is to actually make it faster.
And I would say 99% of the time you can make applications faster without going down so deep on this low-level (which is still not easy) but sometimes you can only get faster via low-level optimizations.
Coincidentally a few days ago, also for a sorting algorithm, I stumbled over a situation where replacing the branchless cmov with branching instructions actually made the code 30% faster:
https://github.com/graphhopper/graphhopper/pull/3380
Or at least the AI explained it this way to me, but I'm unsure if this is correct.
Not useless. It is like the missing printer driver for Linux Desktop. It makes the experience ugly, but this is not the fault of the Linux OSes.
Also the bank should not require apps (instead they can offer hardware key support or desktop apps) and in fact some - at least in Germany - offer a different authentication possibility. Also the app for the German ID is published on fdroid and does not rely on Google services.
Even if these models are so strong that they become a security threat if in the wrong hands this arbitrary and intransparent regulation worries me. Not for the current SOTA model vendors but for the open weight models.
I have always wondered how AI industry is not in a (partial) bubble because open weight models will substantially eat into their revenue (rather sooner than later). But with this ('bro') regulation it seems that an easy strategy could be to regulate this problem away, especially easy to convince everyone that those chines models are truly evil.
The current costs do not have to be sustainable for the SOTA model providers as they grow their user base. But I really wonder about the future as the costs have to increase at some point (to be sustainable) but at the same time the competition and local models get better and better.
I think the German Personalausweis has an electronic ID function that does this: age verification without giving out other information like name etc. But the application chooses which information is requested and could/will request more.
Unfortunately few know and use this information although the usage via NFC is not that complicated. Have used it successfully already a few times for banks, but including name information etc.
> and it takes more than 600kcal to create 600kcal plus transportation into your home
This is not a strong argument or at least a different area of discussion. Because you can say exactly the same for the electricity: you need more than the power. You need the coal+transportation, or power plant, or solar panels ...
I'm pretty sure humans are much more energy efficient ... for certain tasks ;)
Wow. This is crazy. And not only much more expensive but the standardization means you have to buy more disk if you e.g. just need more RAM or similar.
And does the standardization mean that I can no longer buy extra hardware?
1. the difference for the (public) elder care (Pflegeversicherung) is already there: a fee of 2.4% vs. 1.8% with kids (or much less with more kids). And it was now proposed to slightly increase the fee to 2.5%, i.e. +0.1%. But as it was recently increased several times (nearly every year) this is a strange suggestion and looks helpless.
2. this small increase (<1 billion €) does not fix the existing financial gap of approx 4 billions. The gap will increase a lot for the next years.
A bit off-topic, but I (as a non-native reader) find the "anti" in the name off-putting. Has it a special meaning beyond the (meta)physics or is Google just bad with naming?
This. But OP wanted special requirements to open a PR. I.e. if those requirements are not met the PR is never visible to all and so admins can reject spam PRs without giving them a platform.
Your suggestion would help a bit but I would prefer the opposite: before someone can 'pollute' my pull request space and draw attention from subscribers I would prefer an acceptance step (just like a moderator on a forum) instead of having to archive the PRs.
This is especially important as (AI) spam increases and just because I am away for a few days or weeks I don't want those PRs lurking around.
This is what I do not understand as well and advertising the knowledge and more advanced model is also the only thing that comes to my mind.
Since a month I am using gemma4 locally successfully on a MBP M2 for many search queries (wikipedia style questions) and it is really good, fast enough (30-40t/s) and feels nice as it keeps these queries private. But I don't understand why Google does this and so I think "we" need to find a better solution where the entire pipeline is open and the compute somehow crowdfunded. Because there will be a time when these local models will get more closed like Android is closing down. One restriction they might enforce in the future could be that they cripple the models down for "sensitive" topics like cybersecurity or health topics. Or the government could even feel the need to force them to do so.
Does somebody know the power values for idle and sleep for this release? When I tried Remix 43 on a m2 mb pro 1 month ago the idle power usage was above 5W (max 10h) and sleep roughly 3W (~20h).