I wonder why it apologized, seemed like a perfectly coherent crashout, since being factually correct never even mattered much for those. Wonder why it didn’t double down again and again.
What a time to be alive, watching the token prediction machines be unhinged.
I had that for a few months and cancelled. They have minutely rate limits as well so you get 3-4 hyperspeed responses and then a 45 second pause waiting for the throttling to let your next request through.
And then, depending on what you're working on, the 24M daily allotment is gone in under an hour. I regularly burned it in about 25 minutes of agent use.
I imagine if I had infinite budget to pay regular API rates on a high usage tier, it would be really quite good though.
> They have minutely rate limits as well so you get 3-4 hyperspeed responses and then a 45 second pause waiting for the throttling to let your next request through.
I haven’t really gotten that, though have noticed on some occasions:
A) high server load notifications, most commonly, can delay an answer by about 3-10 seconds
B) hangs, this happens quite rarely, not sure if a network issue or something on their side, but sometimes the submitted message just freezes (e.g. nothing happening in OpenCode), doesn’t seem deliberate because resubmitting immediately works, more often than not
> And then, depending on what you're working on, the 24M daily allotment is gone in under an hour. I regularly burned it in about 25 minutes of agent use.
That’s a lot of tokens, almost a million a minute! Since the context is about 128k, you’d be doing about 8 full context requests every minute for 25 minutes straight.
I can see something like that, but at that point it feels like the only thing that’d actually be helpful would be caching support on their end.
You must be on some pretty high tier subscriptions with the other providers to get the same performance!
I still enjoy using GLM 4.7 on Cerebras because of the speed you can get there and the frankly crazy amount of tokens they give you. Before that, 4.6 messed up file edits in OpenCode and VSC plugins more frequently, 4.7 is way more dependable but still has some issues with Python indentation and some partial edits sometimes (might also be tooling issue, e.g. using \ vs / as file separators in tool calls too) - but the quality of the output went up nicely!
I hope GLM 5 will also be available on Cerebras, since for the low-medium complexity work that's my go to, with Codex and Claude Code and Gemini CLI being nice for the more complex tasks.
Most people are on Discord, joining servers has very little friction (no separate accounts), there's decent bots and mod tools and such, ability to create as many channels per server as you want (e.g. for discussions, media, music, whatever) and participate in text based conversations more or less live. It's also easy to jump into a VC and chat while you play games with friends, share screen, stuff like that.
Most alternatives suck for that purpose. There is search for server members and the "blinking things" keep you up to date with where the new stuff is and I presume give you dopamine hits. It's simply not old school forum software and considerations that'd make sites like that good don't enter the equation at all - which also makes attempts at turning Discord into a support forum for any "organized" group or project misguided at best, but also great for the more casual gaming/interest oriented communities.
It does have a weird source of friction. The need to find an invite to the "server". Sometimes you'll find one but it will be dead. There's no way to search to find a server then join it as far as I know.
This, but for an escrow so people can show their actual interest in GitHub Issues, instead of just demanding new features or fixes. So if it gets implemented, the devs get the bounty, if not then they're refunded. I sometimes think about how this could help fund open source at least a little bit.
No comment on making PRs paid, not everyone would react well to that, and some people might be in countries and circumstances where any amount would be problematic.
escrow is a more complex system, and there are multiple possible implementations, but the nice thing is you can skip it and get the same results.
let's assume for a second that the repo owner spends time on PR review, and that time needs to be reimbursed. let's also assume that the person pushing a PR expects some sort of bounty. then as long as the review price is less than bounty price, there's no need for escrow. the pushing party goes out on a limb paying the reviewer to merge their PR, but also expects (rightly or not) to be remunerated for solving the bounty. whether they really did solve it is in the remit of the bounty originator, who might or might not be part of the group controlling the repository. if there's escrow, then the bounty giver probably has to be part of that group. not having escrow allows for crowd funding by interests outside of the repo controlling party.
escrow is only usefully different in a situation when there is no bounty, you want to push code, and then you want to say "ok, here's some money, and here's a PR, either accept the PR and give me money or don't accept it and take my money" as a means of skipping the line or getting a shot at pushing in the first place. however, at that point two things are apparent: 1. you expect the reviewer to do work required to implement your desired changes for free and 2. this might start getting abused, with PRs getting rejected (to gain money) but then modified / refactored versions of this code being pushed via commits or from another user who is the repo owner's puppet (refactoring code is becoming super cheap due to AI). so that degenerates escrow-to-push into a scam.
there are more considerations like that in the article I linked to. I agree that an economy around FOSS pushing would be desirable. it also doesn't preclude free-as-in-money contributions - there are at least two mechanisms that would allow it: 1. you get sponsored by someone who sees your talent (either gives you money to push, or they have push access to that repo and can hand it out free) 2. you create a fork that becomes so good and valuable that upstream pulls from you for free
ultimately becoming a respected developer with free push access to contended repositories should be something that you can monetize to some extent that's purely within your remit, and it would greatly reduce unserious bullshit coming from third parties (especially all those weird hardware developers) and make it easier to be a FOSS dev.
I like writing code that I don't have time pressure around, as well as the kind where I can afford to fail and use that as a learning experience. Especially the code that I can structure myself.
I sometimes dread writing code that's in a state of bad disrepair or is overly complex, think a lot of the "enterprise" code out there - it got so bad that I more or less quit a job over it, though never really stated that publicly, alongside my mind going dark places when you have pressure to succeed but the circumstances are stacked against you.
For a while I had a few Markdown files that went into detail exactly why I hated it, in addition to also being able to point my finger at a few people responsible for it. I tried approaching it professionally, but it never changed and the suggestions and complaints largely fell on deaf ears. Obviously I've learnt that while you can try to provide suggestions, some people and circumstances will never change, often it's about culture fit.
But yeah, outsource all of that to AI, don't even look back. Your sanity is worth more than that.
I wonder if some of the divide in the LLM-code discourse is between people who have mostly/always worked in jobs where they have the time and freedom to do things correctly, and to go back and fix stuff as they go, vs people who have mostly not (and instead worked under constant unrealistic time pressure, no focus on quality, API design, re-factoring, etc)
I’m pretty sure that the answer to that question is positive: those who have worked with code that sparks joy won’t like interacting with it closely being taken away, whereas the people for whom the code they have to work with inspires misery will be thankful for the opportunity to at least slight free themselves from the shackles of needing to run in circles for two weeks to implement a basic form because everything around it is a mess.
Sometimes you don't need "serious" bullshit like Spring Boot either but something more minimalist is more than enough. Though I think it can only work well when you have enough usable primitives, e.g. I've written software in Go that has minimal dependencies and it only worked so well thanks to both the language being relatively simple AND also having a really strong standard library.
The moment when someone tries to write a web server from scratch in a collaborative environment (e.g. project at work with deadlines), I'm peacing the fuck out of there. And in my experience, trying to make design systems or frameworks from scratch leads to a pretty bad mess, because people miss all sorts of inherent complexity until it comes around to bite them later.
I guess what I'm saying is that there's quite the difference between what you do for your own projects, vs what is most likely to lead to long term success in a group setting, especially in 1-2 pizza teams - evaluate the benefits vs risks when considering using vs building.
I'm also a bit blind to this when it comes to certain things, look at this funny video: https://www.youtube.com/watch?v=40SnEd1RWUU I would still prefer to use self-hostable software so I know more of the stack and have control and possibly some cost savings vs someone who'd just fork over a bunch of cash to a SaaS / PaaS vendor.
reply