1,335 karma · joined December 8, 2011
That's a massive model!
The shift from "value" models to "intelligent, huge and slow" models coming from China is an interesting change in strategy.
My main issue with GLM 5.2 and Kimi 3 is that they're extremely token hungry and thus feel slow(er) to use.
Having said that, the safety system on Fable makes it an extremely unattractive model. It feels that half of the time you're paying double for Opus level performance.
I understand the dislike of YAML but a Kubernetes deployment is ~50 lines, if I had to build my own scripts with a similar feature set I don't think I would be able to get it down much more than that.
It's absolutely overkill for small teams and homelabs (I run a cluster myself) but an absolute godsend if you do need the advanced functionality.
In addition to that the form basically only worked in Edge. We emailed support, they changed something on the backend. It still did not work. We gave up.
In retrospective that was a very clear warning sign that their priorities were misguided. I'm glad we did not waste any further time and effort on them.
It's indeed very worrying what we ask medical professionals to put themselves through for their jobs. I think we can all agree that having a well rested doctor or nurse would be preferable over a stressed/tired one. The amount of hours and night shifts that (young) doctors have to do and the extreme competitiveness of the field (partly) drives this.
I understand that it would drive wages down (somewhat) if we educated more doctors and obviously we shouldn't lower our standards substantially but it seems like everyone involved would benefit from this.
A friend of mine, whose a doctor, told me once that the best way to ask for medical advice is to ask the doctor what he/she would recommend for their own sister/brother. Siblings are close enough that he would not want them to suffer unnecessarily but it eliminates the personal factors. Obviously it differs per doctor but in my experience it usually leads to a good conversation about the trade-offs for medical care.
I added 20 USD in credits for the Xiaomi models a while ago and they've been happily writing and updating hundreds if not thousand of pages and I still have 7 USD left!
It does inspire hope that the Chinese labs seem to be so open although the sceptic in me does wonder what their end game is.
Surely, from a purely economic perspective it would be wiser to keep this proprietary and benefit from the increased API traffic?
A 50% drop most likely has a multifactorial explanation, being told that some traditional male traits are bad (and thus lower well being or social status) or medicated away (see e.g the rise in ADHD and Autism diagnosis) might have some effect.
I'm not nearly knowledgable enough to give any reasonable estimate but it would not surprise me if it was higher than 0%.
Long-term, switch to another messenger app that's opensource and truly E2E encrypted.
That also shows why this is such a foolish proposal.
The truly scary people are not on the "consumer" chat apps anyway and most certainly will be the first ones to switch to another communication channel if this passes. If this will have any effect it'll be that some, "dumb" criminals will be caught.
I know that you can get the same pictures from the internet but building something like this and seeing how all the pieces fit together is extremely cool.
Also an extremely cool project to do with kids, the output is visual but it's such a cool combination of hardware, software, some light math (to design the attena) and crafting (to build it).
I would expect that the solar panels impact the efficiency at least somewhat but apparently not enough to cause real and enough issues for the SBB or perhaps they see ways to improve this in the future.
The manufacturer claims that durability should not be an issue. Time will tell.
Here on HN we discuss facts, jumping straight into racism has no place here.
So basically the left but with a stricter view on immigration?
In my case I just weight myself daily, track the weight and scale my food consumption with the current trend. If I'm gaining weight I'll skip a meal.
It takes a while to figure out what works for you but I can tell you that making small lifestyle changes to maintain your weight is fairly easy compared to figuring out how to lose 10 kg.
I've had more success with creating a plan first and then implementing it in (short-lived) sub-agents.
Ironically good software architecture patterns (small functions, single responsibility) heavily impact the performance of these models as well. They do surprisingly well in well architectured codebases.
They do very poorly in anything that's a mess where Opus and GPT 5.5 still get reasonable performance.
Some of my coworkers even use Sonnet (the default in Claude Code for the 20 USD subscription) and see no reason to change even though that model is definitely "outdated" compared to current SOTA.
It's definitely not near Opus 4.8 level but it's very impressive nonetheless and it does do design extremely well.
All these tiny niche models are perhaps fun as an academic exercise or great for the researchers resume but I highly doubt that they'll add any value or will be used for anything serious.
Even if this becomes a somewhat decent model with a fantastic understanding of "gezellig", "kring verjaardag" or "pannenkoeken", how many people will interact with it before the limits of it will drive them back to a frontier model?
Even if the purpose of this is government & other regulated industries, do we really want our government to use a poor model? Either do it right or don't do it at all.
That does raise an interesting question, what kind of laws should LLMs (attempt to) follow? It's easy enough to spoof the country in the system prompt. I wonder how ChatGPT would respond if I told it I was located in a developing country without any piracy laws.
GPT refused to do so (citing that it's illegal even though I own the games). Deepseek did a wonderful job for 7 cents.
At work I use Opus because, why not? But I could easily switch to a less capable model if needed.
I would agree and also suggest that initiatives like this play a large role in doing so. While there's a lot of bullshit arguments coming from the "yes" camp they do make some reasonable points and it's important that we discuss them to show what the trade-offs are.
I cannot speak for all Swiss but knowing that it was a democratic decision to continue with some, high skilled, immigration makes it far easier to accept than if some government employee in Bern would've made that decision single handed.
That is, pretty damn impressive and very useable. When it comes to architecture or complex problems it does noticeable worse but I don't think anyone expected anything else.
One particular interesting strong point seems to be design and user interfaces. It does seem to punch above it's weight there but that might just be personal preference.
What happens to the (slightly warmer) water after it has been used? Is there a way we could return it in a way to minimise impact? I.e if we extract ground water should we inject it back into the ground? Would that even matter?
In the end I have a feeling that the most efficient solution will most likely be to just increase the price of water during a drought. People will complain but it won't be long before the big consumers will happily adjust their consumption or move to an area with abundant water.
China doesn't access to any of the latest chips technology but Huawei seems to have a roadmap to work around this by focussing on "3D chips" (vertical stacked). It's unclear if they can pull this off but if they can it might be a huge boost and allow them to further drive down inference prices.
GPUs that can run everything from Crysis to CUDA are a harder engineering problem to solve than creating a chip that's optimized for inference. Not to mention that inference is an excellent first step towards a full, competitive GPU as well.
Ofcourse you don't want users to lose the feature once they exceeded your revenue threshold or cross the border so you'll need to implement some kind of tracking. Your analytics and error tracking also needs to communicate with the feature flag service.
Definitely not rocket science but more complex than a environment variable.