aurelia
@aurelia.nullvoid.space
she/they, ancient (32)
i make computers do my bidding for a living
https://social.treehouse.systems/@aurelia
i make computers do my bidding for a living
https://social.treehouse.systems/@aurelia
kene orte an denen kannst du das nur falsch machen, angerotzt wirst du ohnehin
January 13, 2026 at 4:25 PM
kene orte an denen kannst du das nur falsch machen, angerotzt wirst du ohnehin
yeah, or just tell it to fetch explicitly. never seen this happen automatically.
January 13, 2026 at 4:22 PM
yeah, or just tell it to fetch explicitly. never seen this happen automatically.
Nope. They do an OAuth flow.
January 12, 2026 at 7:14 PM
Nope. They do an OAuth flow.
but they don't because the claude plans on the consumer side rely on a bunch of users not actually using their full quota every time.
January 12, 2026 at 7:09 PM
but they don't because the claude plans on the consumer side rely on a bunch of users not actually using their full quota every time.
if they wanted to actually solve this problem they'd just give you an API key with an allowance and not have three code paths for using claude code (subscription, subscription with fallback credit on the consumer console, developer console keys)
January 12, 2026 at 7:09 PM
if they wanted to actually solve this problem they'd just give you an API key with an allowance and not have three code paths for using claude code (subscription, subscription with fallback credit on the consumer console, developer console keys)
At the end of the day an Opus jailbreak is still less than 200 tokens and Anthropic's countermeasure being prompt injecting on specific API keys is a hilarious misunderstanding of the limits of your own tech.
January 12, 2026 at 5:21 PM
At the end of the day an Opus jailbreak is still less than 200 tokens and Anthropic's countermeasure being prompt injecting on specific API keys is a hilarious misunderstanding of the limits of your own tech.
Not a word. I think I was pretty clear on that when I said I think the entire "constitutional AI" and fake safety concern thing is a bit to fuck with their competition.
January 12, 2026 at 5:21 PM
Not a word. I think I was pretty clear on that when I said I think the entire "constitutional AI" and fake safety concern thing is a bit to fuck with their competition.
the cost to anthropic is kind of irrelevant. the fact that they have a monopoly on the most useful model as of right now and can essentially charge whatever they want is the concerning part. that's why open models that compete with them are so important.
January 12, 2026 at 5:17 PM
the cost to anthropic is kind of irrelevant. the fact that they have a monopoly on the most useful model as of right now and can essentially charge whatever they want is the concerning part. that's why open models that compete with them are so important.
Because we're in the loss leader / market capture phase of LLMs where everyone is throwing VC money at the problem and the true cost is not exactly known yet.
Paying OpenRouter for open weight models is not without problems either, but a lot less factors involved.
Paying OpenRouter for open weight models is not without problems either, but a lot less factors involved.
January 12, 2026 at 5:07 PM
Because we're in the loss leader / market capture phase of LLMs where everyone is throwing VC money at the problem and the true cost is not exactly known yet.
Paying OpenRouter for open weight models is not without problems either, but a lot less factors involved.
Paying OpenRouter for open weight models is not without problems either, but a lot less factors involved.
be careful filling out surveys!111
*logs in to the 7th data harvesting verification bot this week*
*logs in to the 7th data harvesting verification bot this week*
January 12, 2026 at 3:56 PM
be careful filling out surveys!111
*logs in to the 7th data harvesting verification bot this week*
*logs in to the 7th data harvesting verification bot this week*
Anthropic specifically seems to be going for regulatory capture aimed at open models too. That "Claude can send THOUSANDS of HTTP requests per second when hacking!11" series of social media posts from a few months back was amusing and telling at the same time.
January 12, 2026 at 12:52 PM
Anthropic specifically seems to be going for regulatory capture aimed at open models too. That "Claude can send THOUSANDS of HTTP requests per second when hacking!11" series of social media posts from a few months back was amusing and telling at the same time.
Local is cool, but open models so you have inference provider choice is probably good enough for now. I'm more concerned with how many people seem to be dependent on Claude for when they increase their pricing. Not that their API is cheap right now.
January 12, 2026 at 12:52 PM
Local is cool, but open models so you have inference provider choice is probably good enough for now. I'm more concerned with how many people seem to be dependent on Claude for when they increase their pricing. Not that their API is cheap right now.
I do that with kobold on my 7900 xtx, but it only fits a quanted small model and you really do notice the difference. I think local is very cool and important, i just wish it was more accessible.
January 12, 2026 at 1:40 AM
I do that with kobold on my 7900 xtx, but it only fits a quanted small model and you really do notice the difference. I think local is very cool and important, i just wish it was more accessible.
i'm sad the amount of calls i can make to openrouter before a spark or similar hardware would make sense financially is way too far up there :(
January 12, 2026 at 1:34 AM
i'm sad the amount of calls i can make to openrouter before a spark or similar hardware would make sense financially is way too far up there :(
good god what have they done to my poor ecmascript, did hiding data in closures go out of fashion, do i need to get my copy of the good parts
January 11, 2026 at 3:34 PM
good god what have they done to my poor ecmascript, did hiding data in closures go out of fashion, do i need to get my copy of the good parts
my guess is that it's related to this
www.bloomberg.com/news/article...
slapping "AI" on your product (no matter what's actually in your product) is the meta to get funding right now
www.bloomberg.com/news/article...
slapping "AI" on your product (no matter what's actually in your product) is the meta to get funding right now
Chat Platform Discord Files Confidentially for IPO
Discord Inc. filed confidentially for an initial public offering, according to people familiar with the matter, adding to a rapidly growing pipeline of venture capital-backed tech listings.
www.bloomberg.com
January 10, 2026 at 3:14 PM
my guess is that it's related to this
www.bloomberg.com/news/article...
slapping "AI" on your product (no matter what's actually in your product) is the meta to get funding right now
www.bloomberg.com/news/article...
slapping "AI" on your product (no matter what's actually in your product) is the meta to get funding right now
honestly, it's great, either they do it and a ton of people will move on, or they don't do it and it's still a large signal for growing anti-genai sentiment.
(let's be honest here, with rumors of an IPO, they're gonna do it)
(let's be honest here, with rumors of an IPO, they're gonna do it)
January 10, 2026 at 3:08 PM
honestly, it's great, either they do it and a ton of people will move on, or they don't do it and it's still a large signal for growing anti-genai sentiment.
(let's be honest here, with rumors of an IPO, they're gonna do it)
(let's be honest here, with rumors of an IPO, they're gonna do it)
that's still in there? wow, i'm pretty sure it hasn't been touched for years even in itunes.
January 10, 2026 at 2:54 PM
that's still in there? wow, i'm pretty sure it hasn't been touched for years even in itunes.
good news, it looks like frontier models are stalling in gains, glm 4.7 actually pretty competitive with sonnet
also bad news i guess
but also good news because we can stop throwing money at training now?
also bad news i guess
but also good news because we can stop throwing money at training now?
January 10, 2026 at 10:56 AM
good news, it looks like frontier models are stalling in gains, glm 4.7 actually pretty competitive with sonnet
also bad news i guess
but also good news because we can stop throwing money at training now?
also bad news i guess
but also good news because we can stop throwing money at training now?
vlad you better get on this right now I KNOW YOU'RE READING THIS
January 9, 2026 at 7:22 PM
vlad you better get on this right now I KNOW YOU'RE READING THIS
absolutely not
and they never were
(people here get nostalgic before they renamed the german version to match the international one)
just drink some actual juice you'll have a better time
and they never were
(people here get nostalgic before they renamed the german version to match the international one)
just drink some actual juice you'll have a better time
January 9, 2026 at 7:18 PM
absolutely not
and they never were
(people here get nostalgic before they renamed the german version to match the international one)
just drink some actual juice you'll have a better time
and they never were
(people here get nostalgic before they renamed the german version to match the international one)
just drink some actual juice you'll have a better time