From my experience at our startup, AI is still pretty shit at Rust. It largely fails to understanding lifetime, Pin, async, etc. Basically anything moderately complex. It hallucinates a lot more in general than JS for comparable codebase size (in the 250k lines range).
Are hallucinations in code generation still a problem? I thought with linters, type checkers, and compilers especially as strict as Rust, LLM agents easily catch their own mistakes. At least that's my experience: the agent writes code, runs linters and compilers, fixes whatever it hallucinated, and I probably get a working solution. I tell it to write unit tests and integration tests and it catches even more of its own mistakes. Not saying that it will always produce code free of bugs, but hallucinations haven't been an issue for me anymore.
We will see when the attacks are public, a lot of the malicious server attacks we have seen in the past were kinda of overblown. Not discounting OP but it is very easy to get into clickbait territory.
Forced savings like done in Quebec, Canada is likely the best model for most people even though I dont like it as an individual that knows how to manage its portfolio. It also has the benefit of creating a sovereign wealth fund that can invest locally and be an economic driver but independent from the government.
I actually like the forced saving of Québec. I also have a defined benefit pension plan, a TFSA and RRSP but I am happy to be forced to contribute the RRQ for the general welfare of the province even though I know how manage my portfolio.
Considering that they also have to consider economic development in their investment decisions, the RRQ funds are well managed by the CDPQ.
This is all good and well wishes as long as investors are willing to pour money into the bubble. When the music stops is where we will see the true colors. Corporations are optimized to make money, governments should be optimized to protect people.
It is interesting to see yjs with hoccuspocus being used. I am currently considering our options for real time document editing + full text search.
Seems like a common approach is something like using yjs for sync with a temporary LSM storage like rocksdb for updates and then periodically snapshot to postgres for full text search and compaction.
I am pushing myself to learn nix and get rid of base images altogether.
The syntax is hard without a functional background but I strongly believe this is the next logical step to harden containers and have reproducible builds.
As for the US, having the laws on the books appropriately applied, resulting in a breaking up of the company would make me much more likely to opt for Azure.
For the remaining 96% of the world population that isn't the US, there's not much you can do, as the ICC case shows you to be an adversary. You'd have to show through big actions that you no longer are one.
I'm sure someone wants to reply "why so aggressive, they're doing their best, they don't have anything to do with the above". Almost certainly someone who wouldn't write this if I were replying to a Flock, ClearView, Paragon [0] or Palantir employee on here, despite Microsoft realistically being a much bigger societal threat - and top enabler of the former companies - in every way imaginable.
You have a core team member of the Azure Linux group invite you to ask/tell them what you want to see in what they are working on and this is what you choose to say? Smh
They referred to Microsoft’s known practice of embrace, extend, extinguish a “conspiracy theory” in a sibling thread, so they’ve essentially lost credibility. I don’t think genuine feedback is going anywhere useful.
“We” feels a little insincere when you’re speaking on behalf of such a large corporation. I’m sure the comment had more to do with weaknesses of Azure as a whole rather than your team’s piece.
When I said "we" I meant the group of folks who work on and care about Linux and open source at Microsoft and in a position to help affect change that comes in via feedback from the community.
I would have to write a book on it, but start with allowing people to create an azure account for an organization without having to buy O365. I kid you not I had to find a sidedoor portal in a Reddit post to do it otherwise it's simply not possible.
Every interaction with Azure is a pain. Just 3 weeks ago I was trying to use Artifact Signing, after spending one hour on outdated doc on how to set it up I get hit with Identity validation. I did all steps and still "in progress" still to this day. You charge 40$/month for "support" on Microsoft Q&A which we all know is a joke otherwise its 100USD just to get a ticket in to know why your process is so broken.
At this point I get better support on GCP which is telling.
But it's true. Microsoft's reputation is in the toilet. After everything from the ICC sanctions to the AI spam in Windows to this month's Patch Tuesday incident, everyone knows to avoid Microsoft products like their life depends on it.
But if you want an actionable idea here's one: make it a hundred times cheaper, or free. People use Oracle Cloud because it's free, even though Oracle is even worse than Microsoft. If you want people to use it, you know what to do.
My company picked Azure. So I work with it every day and it is extremely painful to deploy anything that’s not a dotnet application on azure dev ops. One time the app service deployment pipeline just silently failed while trying to build our app. We only found out our new code didn’t deploy when someone asked about the new features expected to go out.
The management portal is super slow, every time you click a button it’s basically a roll of the dice whether the action will work or not.
And as with most things Microsoft these days there are reams of docs detailing every single feature, and none of it fucking works as described.
I will say, if you just want to deploy a quick app from VSCode from your local machine or whatever, it works great. But if you need anything off the golden path it quickly becomes frustrating.
I have worked with AWS, Google and Azure. Google Cloud has the worst UI of them, it slow, broken and just horrible. UI in AWS may be faster than Azure, but overal layout and organization feels a lot better in Azure. I would strongly recommend clearly separating builds from deployments, if you don't want bad surprises. In the age of containers there should really be no difference in how, where or what you deploy.
Don't forget the part where blades will often be different from what's described in the docs, because Microsoft loves changing/renaming shit for no reason.
I do have to give them credit. The cli is pretty good. And Azure Storage Explorer is probably the best Microsoft app I’ve ever used. So props to the team who made that.
Thanks, sorry for my tone yesterday. It was the end of a long frustrating work day.
Azure Linux does look interesting, thank you for working on it. Fedora is a great choice as a base image. Having a Fedora based distro designed to work well with WSL would be amazing! As a base image for apps though I'm curious how you manage the 6month release cycle. Are you planning on expended support, or would people using it need to upgrade every 6 months. I think the appeal of a Debian base is we only need to think about big upgrades every 2 years.
A few bits of Azure feedback I can think of now. Probably not directly related to what you work on, but just some of my experiences working with Azure for the last year.
1. The CLI is good, I think maintaining feature parity between the CLI and portal is really helpful and allows us to integrate with our internal infra more easily. Azure CLI is really the best part working with the service.
2. The management portal is really flaky. Like unknown error messages pop up when clicking on deployment logs. Sometimes the SSH or log tail functions just don't load at all and overall the experience just feels sluggish. I'm really not sure what can be done about this but I've been moving to the CLI just because the web interface is frustrating to work with.
3. The Microsoft documentation is really verbose and difficult to navigate in my opinion. Like we were looking in to hosting a Teams bot and those docs are full of emoji and full page articles like 'why did we make an SDK?'. I have to jump around several pages to get to what I need and even then the code examples in the docs are not actually in sync with the current version of the SDK library. It feels like AI was just set loose to write as much as possible. I think the problem is the information density of much of the documentation is very low. Maybe that's something that can be addressed going forward.
Just doesn’t match my experience at all. AWS isnsuper complex but stuff works. GCP has clearly the nicest interface but not every feature that AWS has. Azure is complex, slow, hard to use and incredibly opaque. No way I’ll use it again out of my own free will.
The key to get better quality AI PR is to add high quality Agents.md file to tell the LLM what are the patterns, conventions, etc.
We do that internally and I cant overstate how much better the output is even with small prompts.
IMO things like "dont put abusive comments" as a policy is better in that file, you will never see comment again instead of fighting with dozen of bad contributions.
reply