571 karma · joined July 10, 2016
DevOops Engineer by trade
BSc. Software Engineer by training
https://mert.akeng.in
https://pvtmert.at.hnhttps://meet.hn/city/49.6202283,6.1729364/
In fact, this is how Kubernetes APIs work. Even the CLI asks YAML or JSON content type when `-o yaml` or `-o json` is specified.
Also, some other sites (albeit sometimes looking at the User-Agent header, hence more hacky) also doing the same. For example `ipinfo.io` returns HTML page when opened in a browser but a JSON (prettified) when requested via cURL.
> But a representative for Amazon said that there is no such company-wide metric for AI usage, nor are there internal leaderboards where employees are measured against each other. Rather, employees are able to view their own AI usage on personal dashboards.
Such a bullshit statement. There is a _global_ dashboard that ranks Kiro/QuickSuite (formerly Amazon Q) usage per-employee based on tokens. The dashboard itself is in QuickSight (well, that also became part of QuickSuite anyway).Not only the data is open to anyone, you can clearly sort by rank, daily/weekly/monthly/yearly usages. Current and former employees included. (By internal alias).
Moreover, there is an internal "awards" system that shows up in PhoneTool profile, each employee gets "awarded" of Kiro/AmazonQ/Quicksuite titles like "Blaze", "Thunderstorm", etc. You can see other recipients of the same award just by clicking on it.
Note: PhoneTool is an internal profile directory where you can look up other employees as oneself...
---On the side, I know several people who cannot produce proper code themselves, or integrate to anything on their own. Those who need constant hand-holding keep producing immense amount of stuff with Kiro/AmazonQ, over-ranking SDEs nowadays. (these are not SDEs, more of SysDev, Support Engineers and TPMs). This itself is not a specifically good or a bad thing. But once they stack-rank based on the token usage, I am pretty sure "good" engineers who put effort into writing "good" code will rank worse than people who does not put effort into "concise" solutions. Therefore, quality will eventually detoriate. And it will be too late once the leadership realizes what has been going on. (Well, they already seen the Amazon-Q/Kiro related outages and keep denying it...)
Such a bullshit. There is (was) Amazon Q (Now QuickSuite) leaderboard which is a Quick Sight dashboard. Moreover, each PhoneTool
The fact that we see stagnation in terms of billions of parameters shows that efficiency does not scale linearly with the model size. More of an S shaped chart. The middle was Claude 3.5. Since then, it is more about integrating and collaborating with different systems.
Which makes a lot of sense. Especially non-Tier-1 services.
Note: Having previously worked at Amazon, certain shifts can be really busy. Busy as in 30-40 pages/incidents over the course of the shift. I sometimes wake up to a "ghost" page, although I left my position earlier this year...i see in the linked post they mention O(n) not O(1). O(1) would basically be impossible and instant. Something like no compute required, constant results...
The name subquadratic is actually good and makes sense to me. Because today's models are usually O(n^2) or worse. Anything equals or less than O(n^1) is basically sub-quadratic.
Meanwhile O(log n) would be logarithmic as the log name indicates. But we have a long way to go there. Maybe with double tokenizer plus extensive caching it may be possible...
What I mean here is tokenizing the user input; then capturing intent; caching intent -> response. So that next time once you get the intent, you don't need to do full transformer inference compute. This can be logarithmic complexity in terms of time complexity.
Should we continue to keep trusting the AI review provided by Copilot?
- BigCo already is a zero-sum deal, they use Xcode-cloud as a service, which runs back on their servers anyway... (Google, Amazon, Azure, etc)
- It was not a long-standing product. Introduced somewhere around 2016~ish if I remember correctly. Only lasted a few major releases. Easier to kill than an established one (ie. TimeMachine)
It seems like somehow got overwritten to the original title of the post.
Nevertheless, knowing Apple so far, unless _some_ large-enterprise~y customer comes and objects, they will drop the support. We already know Intel support is dropping. Why not clean up rest of the things from the kernel and the userspace?
another point to make it safer would be sharing the "chat" with the lawyer, this way it becomes media of communication.
another point to make it safer would be sharing the "chat" with the lawyer, this way it becomes media of communication
A nice and sincere excerpt from the recent past...
> Back when the XZ backdoor was introduced, I was scrolling through news on my Debian Sid laptop with some code compiling in the background. I learned of a backdoor in XZ Utils, potentially introduced by a state actor in version v5.6.0. Thinking back to the fact that I do, indeed, run a bleeding edge distro and update often, I immediately ran apt list --upgradable | grep xz-utils. Sure enough, the stains on my laptop from the coffee I spat out through the nose2 were pretty tough to deal with.
I agree the edges _can_ be sharp, I have chosen to use a thin-plastic cover which also doubles as a sticker-holder. Also helps with the "bumps" and scratches...
> There is a zero percent chance of this ever happening.
Plus, adding an extra point: When you run git log --oneline --graph and the pattern on the left is more complex than the Persian carpet patterns or Ancient Egyptian writings in the Great Pyramid of Giza, you know it's engineering & process quality issue than the code itself...
Obviously not the same, but at home I am running a Raspberry Pi with s3fs mounting my personal S3 bucket. I am exposing the same directory with /etc/exports (NFS). Which also allows me to use filesystem-caching as a bonus on the client side.
On the other hand, I should probably move out from S3 and use R2 or something...
Although I liked the Corey Quinn's post much better as it's less marketing and better comparison in terms of different workloads and pricing aspects.
In the case of if you could bring your own missile-defence-network, then you probably don't need co-location anyway. (There is nothing "co", it's just location you build & operate, with your Patriot or whatever)
As a DevOps, I have seen the quote about "premature optimisation's root of all evil" in real life quite often. In fact, optimising one bottleneck quickly yields another one -moving the goalpost further-, potentially increasing business-impact if the flow is not contained properly.
Especially during incidents, _rushing_ to fix often yields more problems. I've seen people isolating/shutting-down mildly misbehaving instances. Causing excessive load to the remaining and starting the cascading failure like dominos falling one after another.
Which reminds me a scene from "The Office", where Dwight goes rogue and conducts a "Fire-Drill" by locking doors and deliberately causing smoke. Everyone panics and hell breaks loose. This is at the beginning of the episode, maybe 5-minutes tops. I show this at the incident-management training, this is how people behave in real life. No joke.
To give more concrete aspect on the moving goalpost: SWEs improve transaction processing with multi-threading, but that causes more connections/transactions to the database. Even though theoretical gains are Nx (n-times depending on threads/cores), real life gains are 1.2x-1.3x, because database connections are getting occupied. As the next step, increasing number of DB connections helps, maybe add another master node (risk of having deadlocks increase, but ignore for now for the sake of argument). But then the disk IO becomes the bottleneck due to write-heavy (payments domain). Then we add Redis to reduce load, and maybe some asynchronous processing. At this point complexity increases and we need to solve rare occurrences of duplicate data or race-conditions because it is not single-threaded process anymore...
Not to mention manuals/instructions. Regulation discussed here is about these too.
Also as consumer, I would argue the marketing done by apple is just scammy. They keep praising how much carbon saved or sustainable new machines are. But in fact, a minor issue becomes a massive electronic dump.
I also like Macs, I own several of them. Repaired a few. Mostly replacing batteries and keyboards. For example 2014 Macbook Air had a normal battery, no sticky business. Meanwhile 2020-2025 MacBook Air has sticky stuff, making repairs harder.
The best part is, 2014 macbook air has 54 Watt/hr battery, 2020-2025 models are 53 watts/hour. The lasting battery gains are coming from Apple silicon efficiency as well as modern BMS.
Simply put, regulation is the answer. Apple makes it difficult because they can, and also because it creates revenue. Of repairability was the source of income, you would see 10/10 repairable macbooks with no (significant) tradeoffs. (ie. it could be a few grams heavier for added screws)