Also, if AI can write software, create art, run a business and detect cancer in scans, better than any human what exactly is left for humans to move "up" to? The transition from farm labour to software was just blue collar to white collar, physical to intellectual. But if machines have taken the physical work and AI has taken the intellectual work, what exactly is left for us to pivot towards?
Taste was the answer that we agreed on a few years ago. Being able to choose the best output or know when things are off. Having design sense and the tenacity to dogfood the output in order to fix the things AI can't see. It's like a combination of a design lead role with a QA role, you have to know every feature from the perspective of a user, sense where there is friction and confusion and know how to do it better. Maybe calling it a "director" role would be more apt since that combines creative vision with hands on, detail oriented work and directing the work of others to redo the product until it is right.
Seems like the issue happens readily at high load on a system with many cores. Once the conditions were right it seemed to impact a large proportion of their fleet. They just didnt have that combination in their test/validation pipeline. They likely have some sort of load test in their pipeline, but it's running on machine that's much smaller than their production machines... so it didn't trigger the lock contention bug.
The opposite is true running in production for the last several months. They were runnign on big production machines with enough cores, but never quite had enough load to trigger the lock contention. But once both conditions were true, big machines and high load, due to some unrelated failover elsewhere suddently bringing more load, we then had the magic combination.
It's really not unreasonable to expect an enormous hyperscaler like Microsoft to do load testing on hardware that exactly matches their production systems.
I feel exactly the same. Switchting to GLM for Le Chat without an option to prefer a Mixtral model leaves a bad taste. I hope this is temporary while the new model rolls out, but this is not filling me with confidence in either the model or the judegement of the company.
I’m not sure it speaks so poorly on the company. I’m not sure whether they’re using GLM 5.3 or 5.3 flash. The former is twice the price (probably better results?) and the latter is a quarter the price.
If they don’t end up switching models by the end of the year, I think that says a lot about the models. If they’re just waiting for an appropriately large or small model, or waiting to get large 4 put through its paces before switching, I think that’s a pretty ego-less way to go about it, and could speak pretty highly in a way.
Mistral large 3/medium 3.5/small 4 have all been pretty dated for a while so if they just needed a stopgap, I could understand. Now that large 4 is out though, we will see if that was the case.
I have absolutely no problem with them hosting and selling GLM as an offering, it just makes sense for them to do this as an option. As a long time subscriber however, I don't like at all that they have started using GLM as the default model in the Vibe app (formerly Le Chat), with seemingly no way to switch to Mistral's own model. I'm paying a monthly subscription to Mistral explicitly to use a European tool in my daily non technical LLM needs. This feels off and I'm reconsidering my subscription. If all they are giving me is a hosted Chinese model, I can probably get that cheaper elsewhere.
Exactly this. And there are few bigger targets than Chrome when it comes to finding exploits (OSes and network equipment are probably on par). I'm sure they have devoted large compute resources and human staffing at making sure that they find and fix these issues before anyone else does.
I completely agree. With everything from Out-Of-Order execution, deep pipelines, SIMD, huge CPU cache, etc... I would be surprised if the performance increase is not considerably more than 1024x.
Exactly - or encrypted payloads hiding inside standard image file formats. Basically steganography. If all you want to send is a small encrypted text message, it should be possible to hide that in a large image (encoding in the LSB or whatever) in just a small percentage of pixels so that it doesn't fail statistical tests and is indistinguishable from real noise.
I see this slightly differently. Before, if I wanted to be able to do something like flash firmware onto some device I would have to download some random C++ application and install and run it on my local machine. As well as having access to all of my USB devices, it also had access to everything else on my system's user context. I didn't have a way of running that code and only giving it access to a single USB device and nothing else. Now, I can avoid installing anything at all. I visit the project page and opt-in to some flashing flow that's running in a sandboxed env. When the app requests it, the browser asks me for permission and I get to choose exactly which USB device I want to give it access too. That's picking exactly the minimum "outside" access I want to give it, nothing more. It doesnt get to read/write other USB devices I didnt choose. I doesnt get to read/write to my filesystem. It doesnt get to call system APIs. It doesnt get to set itself to start at startup. It doesnt get to install an auto-updater. For me, this is a better security posture than installing random win32 apps.
> I didn't have a way of running that code and only giving it access to a single USB device and nothing else.
To be honest I think that's the most compelling case for webUSB today. If desktop OSes had sandboxing tools (or more granular permissions in general) that are easily usable be everyday users, there would be no need to put webUSB in a browser sandbox. The cross-pltaform nature of it is nice, but that alone is not enough IMO. I think it would be interesting to see a Linux distribution where software that is not explicitly trusted (i.e. not installed by the system package manager) has no permissions by default. Interpreters make this more complicated but for binaries it could work.
Flashing was already solved by UF2, where the device-to-flash temporarily pretends to be a USB storage device. Giving raw USB access to to random websites for that is massively overkill.
I could understand it if you were trying to do realtime configuration of or interaction with some device like a printer or a Stream Deck, but something as trivial as firmware flashing?
trivial for you maybe but many people don't know how and where to find the right firmware for their specific device, and can be in environments where the UF2 volume isn't as obvious (e.g. using a phone)
It's making a niche rarely done use case safer at the cost of making the common case (browsing the web) less safe.
And yes, I am fully aware that I can not press the button that give random sites access... But the issue is it increases the attack surface and is yet another thing that I could get tricked by on a bad day.
The OS should really be able to run code like a firmware flash utility in a sandbox that only has access to one USB device... But instead of improving the OS we keep adding features to the browser which increases the attack surface.
I have a very long list of things I am unhappy about the OS allowing just any app to do, especially app installers/uninstallers should not be a thing.
If you're worried about that, you can just disable WebUSB in the chrome settings. Any website will be denied access to that API from now on. And what's even better: you can selectively enable WebUSB for some websites.
That's what I do and that's what I suggest for any security-conscious user to do. Just explore Chromium settings, there are dozens of various APIs that could be disabled. Do you need Web MIDI? I don't. Disable.
Won't work as a default setting for average user for sure, but if you consider yourself an advanced user, do that.
reply