simon83

So the rumor spreading around the Claude subreddits is that Anthropic is A/B testing plan quotas. A lot of people are complaining about reduced usage since about 2-3 weeks, while others don't see any significant reduction. But nobody bothers to provide data so their complaints are just feels.

Yes, I keep hearing that, but I never saw any concrete numbers. That's why I tasked Opus to write a script to keep track of this.

Here's the gist how it works:

The transcripts log the token count, so you can figure out what those tokens would cost in API prices. Each response returns the current 5h and weekly quota percentages, this is what is shown in the statusline. The returned quotas are not stored anywhere by default. So Opus wrote a small script that sits in the status line and records the percentages while I work. Dividing dollars spent by the percent used give what 1% of quota is worth.

On the 20x plan, 1% of weekly limit is worth about $20 in API prices. A full week is then about $2,000, or $8,000 a month. 1% of the 5h window is worth about $5.20.

All of this only works correctly if you use your sub on a single machine, within CC, as chat usage and other machines can not be tracked this way.

I'm on the 20x right now because I got offered 50% off as a returning customer (I cancelled a few months ago). So next billing period I'll switch to the 5x plan and measure its weekly and 5h limits the same way again.

Yes that's what I remember as well. I haven't used Claude Code in a few months and only started to use it again around 2 weeks ago, but it seems they reduced the the subscription value quite a bit. See my other comment above how I measure the API dollars to weekly rate.

Also the Claude subreddits are complaining since last week about nerfed usage. But since it is reddit, nobody bothers to actually measure the true value.

Do you have it decompile and recompile automatically after every update? Or have you turned off auto updates?

In my sessions it is ~16%, not using workflows due to the nature of my work (a lot of research subagents where fixed workflows don't work well)

Oh wow! I showed Opus the gist and asked it to analyze my own transcripts, and since those summary calls don't show up there, it calculated an estimate based on the 30 second timer.

I'm on the 20x Max plan and my quota is at 98% (resets today), so I had a quite heavy week of usage, including a bunch of orchestrated subagent runs (no workflows).

API equivalent spending for this week: $1,970

Hidden status-label calls: ~$380

Share of the true total: 16%

That's incredible...

Subagents have a default TTL of 5 minutes though, with an opt in to increase it to 1h, for a premium for writing to cache.

Does anyone know how the AMD consumer GPU support on Linux has been implemented? Must use something else than ROCm I assume? Because ROCm only supports the 7900 XTX on Linux[1], while on Windows[2] support is from RX 6600 and upwards.

[1]: https://rocblas.readthedocs.io/en/rocm-6.0.0/about/compatibi... [2]: https://rocblas.readthedocs.io/en/rocm-6.0.0/about/compatibi...

> I can't reason as well or as quickly. Things go over my head. My working memory is so much smaller. I get lost in code all the time. I forget what I'm working on.

Is that while being high on THC, or being sober but having consumed lets say the day before?

I also don‘t think a customer facing chat bot brings much value, but an internal, employee only chat bot could be really useful, depending on the organzization of course. The company in my last position was a rather big one with an insanely huge Confluence instance. I‘ve spent (wasted) so much time searching information there. Having a chat bot, trained on all that information, would‘ve been really useful, I think.

As a Linux noob I decided to try Arch Linux 4 years ago because I wanted to actually learn how the various parts work that make out a whole Linux distro (things like the bootloader, systemd, various desktop environments etc). Other distros give you everything ready out of the box, which made it hard for me to actually dig deeper under the hood, because I had no clue where to start. With Arch Linux you more or less build a whole Linux system from the bottom up.

Following the Arch guides to get a fully working system was easy enough for a complete Linux noob like me. But you still need some basic knowledge about computing, for things like networking, file systems, drive partitioning and such. If you learned this knowledge from other operating systems then Arch Linux is fairly easy to setup by just following their guide.

I understand that all of this is based on some fundamental mathematics, a couple of algorithms here, and some statistical analysis there. And I find it absolutely amazing that we can use all of that knowledge and encode it into something that resembles intelligence. This makes me think if our brains and the concept of intelligence are really as special and mysterious as we assume.

Google: "confetti has left the cannon"

> No results found for "confetti has left the cannon".

I'm amazed that a "stochastic parrot" can come up with such a beautiful idiom.

> maintain security? built-in firewall and DDoS protection?

I have a Hetzner dedicated server (not the Cloud offering) and I setup OpnSense as an all-in-one routing and firewall solution in a separate VM. All incoming and outgoing traffic goes through this OpnSense VM, which acts as default gateway for the host system and all other VMs/Docker containers. You either need to book a 2nd public IPv4 address (or just use IPv6 for free if that is good enough for your use case, since each server comes with a IPv6 /64 subnet), or if you want to just have 1 IPv4 address you could do some Mac spoofing on the main eth interface of the host OS and give the actual Mac address and public IP to the OpnSense's WAN interface. This is necessary because Hetzner has some Mac address filtering in place, meaning only the Mac address connected to the public IP is allowed to make traffic.

You, like many others, are basing your opinion on the current state of LLMs/GPT. And I totally agree with you, the current GTP-4 version might not replace programmers. But how about future iterations? Personally I can't fathom what the next few years will bring us in this area, especially considering the jump GPT 3.5 to GPT 4 has made in such a short time frame. I'm almost convinced that it will make a lot of jobs obsolete in the future, including some programmer jobs, but I'm not bold enough and make a prediction when this will happen, be it in 2, 4, 10 or 20 years.

Oh wow and I thought I'm going crazy, because I also make so many typos when typing on my iPhone 11 (normal size)! On any other Android phone I don't make nearly as many typos, and it just feels so much better. I never understood why, maybe the "hitbox" on the iOS keyboard keys is just smaller? Because I could swear I'm tapping the right key but somehow it often triggers a neighboring key.

Also agree with the horrible cursor movement and text selection, it is wack af on iOS

I was a huge React fan myself but when I started to use Vue and its Composition API one year ago I had tears of joy in my eyes. You have a setup function that is executed exactly once. Inside this setup function you setup your life cycle callbacks, reactive state variables and so on. This is how I wish React + Hooks should have been. Nowadays I won't even accept projects that use React anymore, exactly because of all the pain points OP mentioned.

SolidJS is following the same principle like Vue + Composition API. I think Svelte is also in the same boat. This kind of pattern is so much easier to use and reason about. React Hooks become really painful and annoying really fast.

I've been using React since around 2016 and when Hooks were introduced I played around with it but didn't like it at all exactly because of the reasons mentioned in this article. Sure it reduced quite a bit of boilerplate code that came with class based components and enabled better code re-usability outside of class inheritance or mixins but the disadvantages were too big for me personally to really consider using Hooks in a "greenfield" setting.

In my current project I'm using Vue with Composition API which gave me the exact same "aha" moment the author of this article had with SolidJS. Vue + Composition API is way more similar to SolidJS in principle than React + Hooks.

For this reason I've setup a OpnSense VM on my dedicated Hetzner server where all inbound and outbound IPv4/IPv6 traffic has to go through, it acts as a gateway for the host itself and my other VMs. OpnSense itself is a pretty powerful firewall with tons of other features.

Of course you'll lose access to your server if the OpnSense VM breaks or doesn't boot up for whatever reasons after an update or so, but after 2 years I haven't had any problems. But in case something goes wrong Hetzner offers some nice recovery options, even if you don't have internet access to you server you can access your volumes in some kind of VM and get access to it via a VNC like interface (I had to use this feature a few times during the initial setup which consisted of a lot of trial and error I locked myself out a few times).

I wouldn't run this setup for anything mission critical of course, it's way too hacky and an official firewall solution would be better, but for my personal purposes as a "home lab" like setup it works perfectly fine so far.

I wonder if you could get into legal trouble if you start to re-distribute calc.exe from older Windows version, as a free download. Probably yes, but it would look so ridiculous if MS lawyers would go after a random person doing that.

I just wanted to add that Pinta 1.6 (which is the current stable version in the official Arch repos, for example) has awfully slow rectangle/selection tools on larger images. In my case, selecting something in a 2000x2000 image would freeze the whole program for a couple of seconds. Same with resizing or dragging a selection. It was pretty much useless.

But 1.7 fixes this issue. It's a development version and not considered "stable" yet (though I haven't experienced any issues so far).

For me a range of 100 km would be good enough. I'm just not willing to pay 20.000€+ for that. And so I'm currently looking into a Renault Twizy, which has a range of up to 100 km, for around 8700€. Only problem I have with it is that it's a one person car, and the battery rent is 50€ per month, which is more than I currently pay for gasoline.

> unless Tesla innovated far beyond those, it'll be a marginal car company

You mean just like Apple became a marginal tech company?

I find it fascinating how people make bold claims here in the comments, predicting the future like this...

I'm one of those hatin' cool kids, I guess. Coincidentally I'm also a millennial, one of those ignorant kinds who simply does not get what a Good Life™ we have, at this point in time, compared to the last decades and centuries. Yes, we have it better nowadays (but only some parts of the world), but it could also easily have been way, way better by now (or worse, of course).

Let me also say that I don't hate capitalism per se, it's more or less just a tool with no moral obligations. The problem with capitalism are the humans, like always, and Monopoly is perfect to experience the bad side of human nature: greed and the me vs. them attitude.

Do you think it would be possible to come up with rules for Monopoly, where everyone would be a winner, living in abundance? Absolutely, but that would be boring. If I can't be the best of all, then why bother. It's Me vs. Them, and totally screw Them...

Those people, whose families played Monopoly for long enough until now, and who were able to amass a lot of wealth and keep it within their small circle, they won't allow to change the rules. Because then they wouldn't be the Winner anymore. It's always Me First, and everyone else can die trying.

Edit: I've been really depressive the past 15 years, it started once I was old enough to realize how our world is run. The lack of empathy and social justice is really sickening. I tried to play the game, but I'm not good at it because I lack the attributes that are required to become good at it. I don't care about money, I just want to stay alive, take care of my family and code cool and useful things.

But I started to get a tiny little bit of hope that eventually things may drastically change in the future. Like I do think that we may be able to develop an AGI within this century. One that is rational and smarter than humans. I can totally understand Musks fear of an AGI that would try to destroy humanity (personally I hope it'd go only after the bad leadership :)), and while I think Musk has generally good intentions, he's still a hardcore capitalist, and I'm not sure if I want him and his buddies have too much say in this.

Another shorter term solution to some of the problems we have with capitalism is to setup systems based on Blockchain technologies. I'm really no expert in this area, but I believe it would be possible to set up a financial system that allows for transparency, privacy, but no anonymity and no speculations like we experience nowadays with cryptocurrencies. Something that prevents corruption, black money, etc.

Such blockchain technology could be also adapted for a voting system, where voter fraud then becomes impossible. One promising project I've recently discovered is Cardano SL [1], and the team behind it looks like they really want to make a change. I'll keep my fingers crossed.

[1]: https://github.com/input-output-hk/cardano-sl/

Thanks! Maybe I'll give it a shot...

Thanks, I'll take a closer look into those freelancer networks you mentioned, it looks quite interesting.

Thanks a lot for sharing your experience. You are working purely remote, I assume? I'd be curious to see how I'd fare in their screening process.

Can you tell me roughly how many hours in total the screening process takes? (excluding the time you took to study) And how many hours did you spend on the live project? Was it an actual project, or just a dummy one?

I'm asking because I'm working full time, and the whole screening process sounds very time intensive. Not sure if all of this could be done after work hours, or if I'd have to take a couple of days of vacation.

Thanks a lot!

I saw toptal mentioned in another comment, and their "Why 3%" page is quite unnerving. I don't work at Wharton, MIT, CERN, or Google, and I'm not a Top 100 Django or Rails contributor. Or is this just the 3% of the 3%, and the remaining 97% are more mortal?

I'm not freelancing, but I'd love to hear advice from other freelancers.

I have 11 years working experience as a full stack web developer and I think I'm pretty good at it by now. I'm currently employed and underpaid, and I live in Germany. I have no idea how I'd go about freelancing. What would be your advice to get me started?

From their website[1]:

The Top 3%

"Everyone at Toptal has a proven track record and elite industry experience. There are Toptalers from Wharton, MIT, CERN, and Google, as well as Django committers, Top 100 Rails contributors, leading global designers and art directors, professors, founders and more."

Holy crap man I'm not even going to bother, I'm none of that.

Impostor syndrome intesifies

[1] https://www.toptal.com/top-3-percent