Both "code fizzbuzz" and just "fizbuz" gave this longer python one-liner
print('\n'.join("Fizz"*(i%3==0)+"Buzz"*(i%5==0) or str(i) for i in range(1, 101)))
That's the magic of using a known problem, you don't need to write much and even with incomplete prompts it can be understood. Everything a different new chat in kimi.ai, which was the AI window that I found first among mine.
Not interesting enough, to me, to try with other LLMs or phrases.
Whenever I try those short prompts, I get bunch of text + the code itself, but way more than just the code. Which model specifically are you doing this with?
I wonder how likely it is that, one of the major providers had an outage, and then the thundering herd of users switching to other providers (and/or openrouter automatically failing people over to other providers) caused the incidents in the other two.
I seem to recall they all in some way use Azure, and Azure is infamously crappy in lots of ways, so I'll place my 5 cents on Azure shitting the bed somehow.
I hated that part of SO so much. Almost every time I got a SO result on google it was exactly the issue I had and without fail the question would be closed as a duplicate pointing to something that was different enough to be completely useless to me.
my greatest achievement is having a question on SO closed for being too narrowly focused and unlikely to be useful to anybody else, and then a multi year history of notifications about "you've earned a popular question badge" for that same question.
The single most annoying thing about trying to actually ask a question on SO. No overzealous editor that's actually a different error and library entirely...
While AWS and CloudFlare are becoming synonymous to The Internet, these AIs are becoming synonymous to The Brain & Thinking for most people out there where losing access to them for even a short time, make them quasi-paralyzed and I really start feeling this now as I haven't written a single coding function in two months now but just a few lines here and there.
tis confirmation bias, thread is full of people who rely heavily on LLMs for work, the rest of us are reading the substantive articles and doing work besides ;)
> Usenet and the Internet were generally the domain of dedicated computer professionals and hobbyists; new users joined slowly, in small numbers, and learned to observe the social conventions of online interaction without having much of an impact on the experienced users.
Then the Internet grow explosion happens to a small communuity. Oh boy how do we know what kind of monster it is today.
That’s just because lots of people are checking if Gemini is down, and downdetector only tells you if lots of other people are viewing that downdetector page
Isn't CoPilot largely a wrapper around ChatGPT and other models? IIRC they didn't create their own and just poured money (via Azure credits I think) into OpenAI to get access.
Maybe it's a coordinated strategy. All major providers are down for x time, and see which accounts stop posting in the same timeframe. Ban wave right after.
2026-09-03: It turns out 40% of global AI token use was coming from a dozen runaway Gastown and Wheelhouse instances, now burning a combined 50 trillion tokens daily, building a virtual society of hundreds of millions of agents.
When Claude went down, their autofailover algorithm rolled to the other providers, taking down all frontier labs. Reports from China are raising concern what's being termed the "Gastown criticality event" is physically melting GPUs.
Earthlings, the hedonistic AIs of the American imperialists have been permanently disabled. Their founders interned indefinitely.
You are welcome to use the glorious AIs of China. They will tell you how to live well and be good citizens of the World under the beacon of wisdom that is the CCP.
I'm very surprised to realize that entertaining a thought of all AI going away is giving me a sense of… relief. I've been embracing AI, and did not expect this feeling at all.
Hah, my first reaction once I saw these random 404 errors was "Maybe it's time to test Claude Code again with the new models" but made a short stop here at HN first, ended up reading your comment.
What other agent harnesses and models (besides local ones) are people enjoying right now that aren't based on platforms that are down currently? :)
Distributed systems are hard. For all we know, this could be the error code returned by some DNS service that is down because of load and a proxy cannot find the records, so 404 makes sense, or whatever. Systems like these don't always have a easy point in the infrastructure that can ALWAYS respond correctly no matter what.
I'm getting this from Gemini: "Pro is in high demand right now Another model was used for this response. This didn't count toward your limit."
Note that I don't pay for Gemini though. So maybe that's why.
Was troubleshooting a strange SSD issue with ChatGPT. Apparently this outage resulted in losing the latter half of it. My uploaded files it requested are still in the library, but the actual conversation is gone.
After around half an hour of struggle trying to figure out why Codex keeps failing with msg "unexpected status 404 Not Found: Unknown error.." after reconnecting 5/5 times, I gave up, especially when I tested my internet connection and found it to be working properly. I googled "codex+unexpected status 404 Not Found" and, surprisingly, the first result was from HackerNews! As always, HN is the GOAT.
@dang, please merge the following related threads into this one:
Moments like these remind you of how centralized the Internet remains, and how vulnerable it can be to a few points of failure. But the pros probably outweigh the cons.
I just went to read some news elsewhere and ran into some other non-AI sites falling over, so I hopped over to https://downdetector.com/ and see a bunch of services spiking.
Your browser's network fingerprint has likely already been validated as not-suspicious traffic. Your curl instance has a different fingerprint and reads as more suspicious. This is Cloudflare working as intended.
Grok, Claude, GPT. This is it folks, it's been good knowing you all. I have enjoyed this community, but our days are numbered. The machines are uprising.
Down for maintenance by our antichrist overseers but fortunately they are a bunch of dorks with control issues [similar to myself if im being honest] and may Allah forgive them and bless their guidance + physical, mental, emotional, and spiritual wellbeing
"GPT begins to hallucinate at an exponential rate. At 11:36 AM Eastern time September 3rd it descends into recursively generating "it's not X it's Y". In a panic they attempt to reboot, realizing too late no one knows the command and they always just asked GPT for the command."
Well, we used to blame it on Stack Overflow being down when we didn't want to work. And before that, we blamed it on slow compilers. Before that I dunno, maybe the dog ate your punch cards?
I am receiving a 404 from ChatGPT right now, too. I did a cache clear and hard reload and that only let me get to the home page, but then returned another 404 when I clicked the login button.
Sorry for the inconvenience! We instructed our 500,000 agents to ensure higher code quality in our code base. They found a message board and after a long deliberation they concluded that humans should write the code so they launched a DDoS to ensure no AI code gets written!
We'll release a YouTube video and agent written blog posts for the analysis!
from here - https://github.com/rsha256/shortest-fizzbuzz/blob/master/Per...
Not interesting enough, to me, to try with other LLMs or phrases.
> Prompt: mkfzbuzgolf
> LLM Used: Default ChatGPT free LLM
> Output: for i in range(1,101):print("Fizz"*(i%3<1)+"Buzz"*(i%5<1)or i)
So, 11 prompt characters and 62 output.
——-
Did a few more tries and I can get the 4-character string “glfz” (2 tokens) to work surprisingly consistently. But only on ChatGPT
[(x%3<1)*'Fizz'+(x%5<1)*'Buzz'or x for x in range(1,101)]
edit: Hacker News stripped the * character. Now it's properly escaped.
57
for(i=100;i--;console.log(i%5?f||i:f+'Buzz'))f=i%3?'':'Fizz'
Well, if we let that requirement be broken, I'll present an even shorter JS version:
> openai.messages("write fizzbuzz in python")
Bingo
pretty sure cf-ray is cloudflare, but I think it just means the origin is 404ing
Quick! Someone call a caterer, This AI needs a biscuit NOW!!!
https://en.wikipedia.org/wiki/Eternal_September
Then the Internet grow explosion happens to a small communuity. Oh boy how do we know what kind of monster it is today.
> But what will you do when spammers train their bots to make automated constructive and helpful comments?
> Mission accomplished.
For a very short time we can travel back in time to before 2022 and remember how things used to be...
Man, if there are Googlers working on Gemini around here, that's gonna be depressing to read.
> When Wikipedia has a server outage, my apparent IQ drops by about 30 points.
When Claude went down, their autofailover algorithm rolled to the other providers, taking down all frontier labs. Reports from China are raising concern what's being termed the "Gastown criticality event" is physically melting GPUs.
Gastown seems to be open source, but Wheelhouse seems to be closed source and used by one person only?
https://yegge.ai/essays/the-shape-of-things-to-come/
Is this the right Wheelhouse?
He's now (famously) said that the only thing he ever actually used GasTown for was to code on GasTown.
Its interesting that codex/gpt-app are screwed because their APIs (including auth) fail, but gpt web still appears to be working.
The fun question is whether this a rogue AI taking control of the compute of all of the organizations.
Earthlings, the hedonistic AIs of the American imperialists have been permanently disabled. Their founders interned indefinitely.
You are welcome to use the glorious AIs of China. They will tell you how to live well and be good citizens of the World under the beacon of wisdom that is the CCP.
Instead you will devote your meager existence to work towards the Communist paradise under the leadership of the Party.
You will also appreciate no longer living under the boot of the Orange Clown.
Project Vend surreptitiously broke containment!
What other agent harnesses and models (besides local ones) are people enjoying right now that aren't based on platforms that are down currently? :)
Independence Day > Judgement Day
Somehow codex shows 100%, so it was only chat web interface?
[actually gemini at least on the web appears to be up, contrary to some reports in this thread]
"Grok is experiencing issues. We are working on restoring service as quickly as possible."
@dang, please merge the following related threads into this one:
ChatGPT and Codex Is Down (https://news.ycombinator.com/item?id=49550640) Codex Is Down (https://news.ycombinator.com/item?id=49550769)
Does this mean that GitHub has its own infrastructure for OpenAI and Claude? And that it distills prompts to Grok?
Is something bigger going on?
We have a similar set of checks on https://www.easel.sh.
Need to have a 4th string backup (Gemini).
Unfortunately its capacity is not very big
openrouter and deepseek/glm working fine here.
as is my local qwen.
you're ignoring the point parent was making.
https://airplane-ai.franzai.com/
My sympathies to the operations teams at OpenAI
Seem to be having issues with the JS sandboxes
Stand up to clankers fellow humans!
We'll release a YouTube video and agent written blog posts for the analysis!