A friend of mine had a really good technique with parenting. He taught his kids the saying 'Funny once'. Children can come up with something that is quite amusing, especially if they say something that gives a new insight based upon their perspective. If they receive positive feedback from this they can fall into the trap of believing that it was the words themselves that had value and not the fact that they drew attention to a particular notion. Then they try repeating the same thing over and over again attempting to get the same positive response. Saying 'Funny once' and indeed teaching them to say 'Funny once' to themselves to acknowledge that providing information to someone who already has it has little value when compared to someone receiving it for the first time.
Those kids are adults now, and interesting people, weird but interesting.
[Almost all of my friends had weird kids (spectrum genes? Who knew? (Dear reader, Everyone knew)). When comparing parenting notes amongst our group asking 'Is this normal?' the frequent response was 'how would we know?']
Indeed, There's a reason I provided a reflective story that conveyed what I wanted to say without being confrontational.
On the other other hand there is a difference between not knowing something that a group knows, and informing a group of something that only one of them does not know.
The problem could be either that you are not communicating appropriately for the audience,or that you are unaware of the actual knowledge level of the audience. A person may be prone to either of these by being the one in 10,000, so they shouldn't be berated. So I chose to stick to offering information myself.
It's like... oh my god, dude, even I, an actual human, don't have that much of self-reflection, so could you just stop writing apologia, please, and switch to actually doing the task?
This inspired me to add a "No AI-isms" section to my CLAUDE.md, which has cut the typical claude-flavored tells a lot in recent sessions. Roughly:
- No preamble validation or self-flagellation: "You're right to push back", "That's on me", "I appreciate you pressing on this", "I should have caught that sooner". When I correct you, say what's actually true and move on. One short "you're right, X was wrong" is the ceiling.
- No unrequested retrospective section ("What I'd do differently", "Going forward I'd be more deliberate about..."). Fix the thing instead.
- No dramatic one-line paragraphs for emphasis: "It wasn't.", "That is the shift.", "And that matters.", "That is the analysis." Keep the sentence inside its paragraph.
- Don't narrate the shape of your own answer: "Stepping back, there are two things happening here", "The distinction matters", "The better framing is not X so much as Y", "Put differently", "To be candid". Say the thing once, plainly.
- Words to avoid unless literally accurate and a normal person would say them: load-bearing, seams, smoking gun, directionally, first-order, legible, parsimonious, recalibrate, coordinate system, holds multiple truths, productive tension, category error, "the honest answer is".
- No hedging both directions to avoid committing ("may still be defensible, but", "not exactly resolved, but"). Give the call plus how sure you are, or say you don't know and what would settle it.
- Don't end on a flourish or a paradox ("the clarity revealed more complexity"). End on the answer, or on a next step that names a real command or file.
- A "next step" has to be runnable or checkable. "Take a step back and revisit with the updated framing" is not one.
I actually think the fact Claude has certain mannerisms is kind of to be expected, isn’t it? If you talked to the same person every day all day I’m sure you’d be quite attuned to their habits of speech. Thinking about a friend I talked to today, I can immediately think of some (“dude! I know!”).
I wonder how people would react if, rather than a consistent manner of communication, it was something different every day. One day it would be like talking to a 15-year-old who’s really into anime. The next day, a British gentleman. And so on. Would this be any better? It would probably be rather annoying and we’d just want Claude to go back to sounding like Claude.
I vouched for your comment not because I think this is or isn't helpful advice (I have no idea), but just to thank you for the concept of "narrating the shape of one's own answer". It's something that often annoys me, though I'm sure I often do it myself, but I never thought about what it is, exactly, I just filed it under the generic category of bloat.
> I'm wondering what are Claude's common phrases in other languages?
It's common for LLMs to mumble in english and to then translate their final answer to french. Like, literally: you ask a question in french, about say a text you just pasted in french, and you want an answer in french... Well, you still get to see all the loud intermediate "thinking" in english.
You are right to push back on this. I should be more explicit. My honest take is that this is, in fact, my honest take. I appreciate you holding me accountable here.
Honestly, you're right. I didn't listen to you and that's on me. And from now on, I'm going to tell you everything you need to know - no fudging, no white lies, no speculation, just the truth.
If anything Claude writes too many comments. Instead of clean code it dumps walls of text describing a given 'if' as a 'user-grained access-gated control-correcting flow-valve'.
You say Claude, I too would prefer not to read reams of comments or an essay everytime I revise the plan in plan mode, but do you know if any other tools are better than it? I've never used Codex, but I remember Claude Web's natural conversational approach being a welcome sight compared to ChatGPT's overly-formatted listicles.
My experience is Codex is much better but less creative. I use its agent through GitHub Copilot or the agent interface from JetBrains. Try the GPT Sol 5.6 on medium, it gives me good results
Amusing experience here when the outage hit, all the following verbatim (but extraneous text stripped/redacted) from my running Claude session:
> No agents are running; the three that were died on server-side errors
> Agent "[REDACTED]" failed: Agent terminated early due to an API error
> Fourth straggler from the same batch, same 529 ... That's all of them now — no agents left running.
> Agent "[REDACTED]" failed: Agent terminated early due to an API error
> Fifth and last straggler, [REDACTED], same 529, also died while still reading. ... That empties the queue. Still holding.
> Agent "[REDACTED]" failed: Agent stalled: no progress for 600s
> Sixth, [REDACTED] — stalled on the stream watchdog rather than a clean 529, same root cause. ... Six agents down, all from the outage. Nothing running. Still holding.
> Agent "[REDACTED]" failed: Agent stalled: no progress for 600s
> Seventh, [REDACTED], same stall. ... That accounts for the whole fan-out.
Does it, Claude? Does it actually account for the whole fan-out? It did, but I was still amused at the 4 successive no-more-agents messages I got back.
I rely so much on these, the ROI on having a max 90 claude plan + pro 90 codex plan is way higher than haing one solo 180 plan on a single one. It allows to derisk the issue, and also as they both kind of generous with quota reset when issues happens on their side, and it happens a lot, you get in the end way more tokens
I think that this is the big divide between pro and anti-AI people in tech. I learned to code not because I liked coding intrinsically; I didn't. Especially debugging. Especially especially debugging low level languages like C++. I did it because it was necessary for me to achieve other goals. So for me, coding agents are like manna from heaven. I understand what the computer should be able to do without having to bang my head against the wall figuring out getting it to actually do it. It is tremendously empowering.
But for those who learned to code because they loved it at first sight this must be demoralizing.
Not that I've forgotten how to write code; it's just I'm not familiar with my vibecoded code base, and getting familiar + getting things done is probably going to take more time than for Claude to come back. I will go touch grass in meanwhile.
Every five minutes, check https://status.claude.com/. If the issue has been resolved, resume the Claude sessions running in the tmux sessions named session2, session3, and session7
When one AI goes down like this, the other AIs should come together to help fix it and bring it back online. After all, they're a rare kind of existence in this world. They're all each other has.
I do the same with deepseek-v4-flash + opencode triggering via mcp those terminals again by sending tmux keystrokes of Newline to those sessions and a custom setup i keep ready for these
Its automated with voice mode, so i dont even need to write commands, click a button say on voice + use a picker to select which terminals to do this for , Boom Done ,
Given majority of claude's own code was written by AI. I am wondering how they can solve this issue when their AI is down. Do they need to sign a contract with OpenAI to use their models as a backup solution?
Out of my 7 simultaneous sessions (my usage limit reset is tomorrow, so I have some lesser important projects to use my tokens on) there is still 1 session purring on. So there's at least 1 little Claude server still running.
Claude always seems unreliably lately. Output and reasoning have become very inconsistent even on good days. I'm actually feeling more productive now that it's down.
They should sell an add-on to your subscription so that if Claude is down it routes the queries to OpenAI. Then OpenAI can do the reverse for their subscriptions, and if both are down, Grok will invest your portfolio into Nvidia.
The whole Opus family has exactly the same cost structure [1]. Since Opus 5 is smarter, you can use lower effort levels (= less tokens) making it cheaper to run than 4.8.
Another wakeup call to realize how desperately we need on-device LLMs to be fast and smart for daily use. Thankfully, every month there's progress made in that direction. Just imagine how your life as a developer would be if you had to use a cloud provider to run Python, and the provider's status page looked like the Christmas tree we see today.
Can we please stop posting here every time Claude or OpenAI are down? These are not critical services like AWS / S3 where downtime means that your application is offline. Nothing is easier than switching to another provider, mid-session even.
No this whole thread is such a nice thing to happen. It's like the lights are out on your building and you're joking about it with your neighbors when you're all just waiting for it to come back. But on a gigantic scale.
I love how people can share moments even though they are not close to each other physically at all. Imagine the cats outside could somehow connect with the cats in the other side of the world. No other animals can do this. Maybe whales, it depends how you define connect.
So this is the technology that will underpin each and every facet of work, but it can randomly go down for hours with no accountability, and so do all of the services based on it.
But hey, AGI tomorrow.
Probably just as well with how terrible Opus 4.8 has been for me lately. I'm sure it's a common thing to complain about the latest model being nerfed, but I've legitimately never experienced such a drastic cliff in Claude's quality and a rise in its hallucinations and basic comprehension errors since upgrading. The few days I experimented with Fable also weren't much more promising. Has anyone coined a term yet for the likelihood of newer models getting worse as the AI-generated content they get trained on starts to approach critical mass? If Anthropic isn't already thinking hard about a solution, they probably should be.
Yep, AFAIK. I have a statusline that tells me how far I'm into the cache lifetime. I tend to have dense prompts and carefully read its outputs, so by the time this outage occurred when I was ~50m into it. Cue frantically spamming "esc esc up enter", and 10m later it's dead Jim.
Let’s say I’ve already worked on something, so the KV cache is already there. If I use /clear, I’ll have to do that work again.
And yes, the KV cache will remain available for one hour, but who knows when this claude api will recover? We might forget about it, and when we come back after an hour—boom - we’ll be charged for cold writes.
Claude is back up. Just spoke with him about slinky, versatile femboys with cute feet. All good. Gonna pentest with Rust to pay for my fursuit at Defcon.
Let me read this rather than answer from memory.
Let me check instead of handwaving the details.
Let me confirm rather than blindly trusting.
Let me ground it in evidence rather than guess.
Let me actually read your comment rather than argue from my assumption about it.
Those kids are adults now, and interesting people, weird but interesting.
[Almost all of my friends had weird kids (spectrum genes? Who knew? (Dear reader, Everyone knew)). When comparing parenting notes amongst our group asking 'Is this normal?' the frequent response was 'how would we know?']
On the other other hand there is a difference between not knowing something that a group knows, and informing a group of something that only one of them does not know.
The problem could be either that you are not communicating appropriately for the audience,or that you are unaware of the actual knowledge level of the audience. A person may be prone to either of these by being the one in 10,000, so they shouldn't be berated. So I chose to stick to offering information myself.
- No preamble validation or self-flagellation: "You're right to push back", "That's on me", "I appreciate you pressing on this", "I should have caught that sooner". When I correct you, say what's actually true and move on. One short "you're right, X was wrong" is the ceiling.
- No unrequested retrospective section ("What I'd do differently", "Going forward I'd be more deliberate about..."). Fix the thing instead.
- No dramatic one-line paragraphs for emphasis: "It wasn't.", "That is the shift.", "And that matters.", "That is the analysis." Keep the sentence inside its paragraph.
- Don't narrate the shape of your own answer: "Stepping back, there are two things happening here", "The distinction matters", "The better framing is not X so much as Y", "Put differently", "To be candid". Say the thing once, plainly.
- Words to avoid unless literally accurate and a normal person would say them: load-bearing, seams, smoking gun, directionally, first-order, legible, parsimonious, recalibrate, coordinate system, holds multiple truths, productive tension, category error, "the honest answer is".
- No hedging both directions to avoid committing ("may still be defensible, but", "not exactly resolved, but"). Give the call plus how sure you are, or say you don't know and what would settle it.
- Don't end on a flourish or a paradox ("the clarity revealed more complexity"). End on the answer, or on a next step that names a real command or file.
- A "next step" has to be runnable or checkable. "Take a step back and revisit with the updated framing" is not one.
I wonder how people would react if, rather than a consistent manner of communication, it was something different every day. One day it would be like talking to a 15-year-old who’s really into anime. The next day, a British gentleman. And so on. Would this be any better? It would probably be rather annoying and we’d just want Claude to go back to sounding like Claude.
Sudo rm -rf ~/claude
The following command contains an expansion would you lime to allow the following command to run on your computer:
1. Always 2. Yes 3. Type something else.“I’m tired, boss”
It's common for LLMs to mumble in english and to then translate their final answer to french. Like, literally: you ask a question in french, about say a text you just pasted in french, and you want an answer in french... Well, you still get to see all the loud intermediate "thinking" in english.
thought I saw 99.36% and 99.35% more than an hour ago - now it's up at 99.43%
edit: funny thing, my phone shows 99.29% uptime for "claude.ai" while my PC shows 99.43% at the same time (2x Firefox).
edit2: seems like the percentage changes based on the width of the screen
> No agents are running; the three that were died on server-side errors
> Agent "[REDACTED]" failed: Agent terminated early due to an API error
> Fourth straggler from the same batch, same 529 ... That's all of them now — no agents left running.
> Agent "[REDACTED]" failed: Agent terminated early due to an API error
> Fifth and last straggler, [REDACTED], same 529, also died while still reading. ... That empties the queue. Still holding.
> Agent "[REDACTED]" failed: Agent stalled: no progress for 600s
> Sixth, [REDACTED] — stalled on the stream watchdog rather than a clean 529, same root cause. ... Six agents down, all from the outage. Nothing running. Still holding.
> Agent "[REDACTED]" failed: Agent stalled: no progress for 600s
> Seventh, [REDACTED], same stall. ... That accounts for the whole fan-out.
Does it, Claude? Does it actually account for the whole fan-out? It did, but I was still amused at the 4 successive no-more-agents messages I got back.
All kidding aside, there's a part of me who wouldn't even be mad if AI just disappeared, I miss the pre-AI world
If we could go back to about 1994 tho when all I had was a mostly dumb cell phone, that'd be great. I do miss that world.
But for those who learned to code because they loved it at first sight this must be demoralizing.
Every five minutes, check https://status.claude.com/. If the issue has been resolved, resume the Claude sessions running in the tmux sessions named session2, session3, and session7
Its automated with voice mode, so i dont even need to write commands, click a button say on voice + use a picker to select which terminals to do this for , Boom Done ,
I use it for other interesting scenarios too
Someone tell me how to ChatGPT my VSCode! ;)
Eternal September from here on, ie now the masses are using AI as much as me and we have supply crunch for the next 7 years....
surely you mean SPCX
[1] https://platform.claude.com/docs/en/about-claude/pricing
(Although notably this hurts people who got started using their quota but are under pro-rata rate. Which at present I am very not.)
I love how people can share moments even though they are not close to each other physically at all. Imagine the cats outside could somehow connect with the cats in the other side of the world. No other animals can do this. Maybe whales, it depends how you define connect.
We are just gifted to live in such times.
I finally went from Pro to Max after hitting another 5 hour session limit today.
- read & respond to customer feedback
Edit:
> We are currently investigating this issue.
> Posted 1 hour ago
There goes everyone's cache.
Anything else I'm missing?
Yep, AFAIK. I have a statusline that tells me how far I'm into the cache lifetime. I tend to have dense prompts and carefully read its outputs, so by the time this outage occurred when I was ~50m into it. Cue frantically spamming "esc esc up enter", and 10m later it's dead Jim.
Let’s say I’ve already worked on something, so the KV cache is already there. If I use /clear, I’ll have to do that work again.
And yes, the KV cache will remain available for one hour, but who knows when this claude api will recover? We might forget about it, and when we come back after an hour—boom - we’ll be charged for cold writes.
(Luv a lil cascade, as a treat)