This is what happens when you quietly change usage amounts (or change something without thinking it through), people get upset and start complaining, Tibo says "guys there's no problem" and then a few days later, after several, hundred complaints, he does his "war room" where he figures out "what's going on" and they magically find that there were indeed a bunch of issues with usage and then you "fix" them. We've seen this movie before. I can keep track of what "5h" means. Time is pretty linear and fixed. But whatever OpenAI decides is "2x"...that may change with the weather. I'm a huge fan of OpenAI's products but I'm finding myself increasingly disillusioned by this. I'm using my $20 Claude subscription more than my $200 ChatGPT Pro subscription because I'm getting better results, way better usage, and sometimes 5-6x times the speed with Opus 5.5 than I am with Sol 6.1, even on Fast Mode. OpenAI, you need a fundamental reset in how you're handling usage amounts, I think. I'm sure there's a lot more going on with the backend than anyone knows, and I sympathize (I used to work in operations/datacenters), but the end results are what they are.

I had no idea this would be an unpopular opinion, but maybe it's because I'm an adult...I never found it productive to get angry at/be abusive at an LLM and it has nothing to do with LLMs having "feelings" or being "moral". If an LLM goes off the rails, I will either go back to the previous message and resend it, or I'll start a new conversation. I'm not going to waste tokens or time or energy "arguing" with it or cursing it out. If I want catharsis, I'll wait until I'm playing a video game or participating in a demolition derby. It's just so unproductive to get all "bro wtf?" up in your chats with an LLM. Someone will inevitably pipe up and say "nuh uh LLMs are proven to work better if you threaten them" and sure, I have no doubt you can find instances of that for your own self-indulgence. But that doesn't match my experience. tl;dr I doubt this is going to affect anyone who understands how to properly use an LLM. It sure is great for outrage and clicks though; I expect in the coming weeks we're going to see people purposefully baiting this system to post to social media for clout.

I never thought I'd do the "never leaving this app" thing, but...

if she wasnt there for http://localhost:3000, she won't be there for the .com either

This tweet continues to live in my head rent-free. A lot of people could do with reading it:

After consistently seeing this "$LLM wiped my drive" post happen predominantly to Windows users, I wrote up a response that I hope you take to heart if you use coding agents on Windows. I'm not telling you not to, but you need to be safe, and that doesn't just mean "use auto mode".

JUST LET ME DREAM, OK!? 😢

The top reply is "Attention is all they need." This is so disgustingly adorable and wholesome.

Aaaand we just got a reset for ChatGPT, love how OMP showed it to me (you have to opt in to this in settings). Lovely little fireworks :)

I will shamelessly admit that I'd eat the hell out of this.

Every model is nerfed the second you stop being impressed

Hello OMP users! OhMyStash has been updated to fully support Windows instances of OhMyPi: github.com/hkay-dev/OhMyStas…

We're coming up on 3 hours since we were supposed to get a ChatGPT reset...a lot of people burned their usage in anticipation and are now wondering if they should use their banked reset just so that they can keep working. This is *not* a good user experience and not a good look.

I think the lowest IQ response you can have to something like this is "hurr durr OpenAI stole the internet, now they're complaining?" It's an entirely true irony. There is nothing to debate. It's very easy to just reduce down any news about distillation attacks as "Wow look, the master thieves are mad that someone is thieving from them". It doesn't really take any cognitive effort to post that, like a meme, a knee-jerk social media commentary reaction that gets the likes and the retweets and the laughing emoji. But take just a few seconds longer. The point is that frontier labs keep doing all the hard work, spending all the money, time, and resources (the thing that so many are quick to call the "VC money pit") and then along come Chinese labs who distill/piggyback off of that, put out cheaper variants, and then people proceed to talk about how much better the Chinese models are. "I don't care, I just want cheaper models" is actually a totally fine response, and I don't blame you. Local AI is getting cooler and cooler by the day and what it can do is getting impressive, and anything that further democratizes AI access among all the people in the world is a good thing in my book. You don't need to cheer on frontier labs. You don't need to be a fan of Western AI in the slightest. But before you start cheering on and idolizing Chinese models, make sure you understand where their efficiency actually comes from, because it's not magic and it doesn't just materialize out of thin air. AI is as geopolitical as it gets. The last time the headlines were "DeepSeek puts out model that beats OpenAI & Anthropic for just a few million", it rocked the Western economy to the core. It was also completely false but the truth about it never quite got the same level of sensationalism. Do you really think that other countries are not watching the US and how hedged they are on AI? That they weren't paying attention to what happened to the stock market during that DeepSeek event? Why is it, do you think, that Chinese labs flood the market with lots and lots of high quality models...for completely free? What exactly do you think is the benefit for them? Yeah. Start exploring that, and all of a sudden, it's not black and white and it's not as simple as "Merica good, China bad". I'm not making any opinion here about the morality of OpenAI or Anthropic or Google or anybody who trained on the internet. I just implore you to think a little bit harder about it.

I continue to find myself more and more disappointed with Tibo. What exactly differentiates a codex task from regular work?

Yep, the verdict is in. I was completely underwhelmed. I didn't see anything that justifies halving the usage of my $200 subscription. And while 6.1 Sol is nice and all, I don't understand why I wouldn't just use Opus 5.5 and get significantly more out of even a $100 subscription from Claude than a $200 one from ChatGPT. I'm giving it a month just so that I get the remainder of the 20x that I paid for and whatever I get out of the dots, but I'm going to re-up my Claude Max 20x subscription. Speak with your wallet.

Yep. Doubling down on this. Tibo's message should have waited until after the Dev Day keynote.

Conveniently omitted in the tweet. help.openai.com/en/articles/…