You re an idiot. Who are you trying to do that with? Wasn't this your comment? When did this turned into condemning scam? Homeless retards all of you are.

🤣🤣

Joan Wicked

Hi @elonmusk , I see what you guys did with @grok 4.5 model. Like how could it be so smart to handle complex nuances correctly yet so fast? The team cooked. Can we get weekly Limit reset

In this LLM token based economy, do your best to stay in charge of how the agent tackles tasks. The model providers built the client interface, whether it's a CLI tool or desktop app. They own the harness and orchestrate the agent. So you're not just paying for the cost of the job, you're paying for whatever their orchestration costs to complete it. The agent can go haywire, accumulate bills, break your codebase, then attempt to fix it and introduce even more bugs. You still have to pay. You and I know this model isn't sustainable. It will burst. People who think this is some kind of miracle will eventually realize they're better off paying a developer $500 to deliver working code than paying the growing cost of LLM mistakes. You don't think this is a big issue because you've never been able to build something of note before LLM coding. Like I said, this will burst because it isn't sustainable. When that happens, those who found empowerment in this era will look for another form of empowerment to spend their money on.

People sleeping on SVG. I have been working with itbfor the past 7 years and I cant stop being amazed at the promises and possibilities it holds. Warning: Browsers cant render these natively anyway.

To whoever needs to see this: The claim that running AI is expensive for organizations like GitHub, Anthropic, and OpenAI is only a business strategy. Right now, they are taking advantage of the AI hype to sell tokens, with a system that is heavily influenced by their harness and inherited prompts. Secondly, @AnthropicAI reached its high limit at Sonnet 4.6, for @Google it is Gemini 3.1, and for @OpenAI it is GPT-5. Everything else has to do with adding more training data and prompt engineering so you can "one-shot" things. That is why you will continue to see more models until they cover all the use cases you can one-shot. But these baseline models are enough for all coding exercises if you actually understand what an LLM is and the limits it has. I understand this because I used to jump on the latest models from these three. But a recent issue with @GitHubCopilot made me go back to the drawing board. For context, I just started using a full IDE (VS Code) three years ago. That means for 14 years, I wrote code in several languages using Notepad++. I adopted AI because I realized that I have coded by hand enough to welcome anything that will ease that stress. I know my shit. I know what "done" means. I know certain requests I could make under certain contexts where an LLM will probably mess up the codebase. So, I picked the @antigravity CLI (I used to work with the Gemini CLI), chose Gemini 3.1, and changed the way I work with it. Now, I work with an LLM the same way I would work with Notepad++, except now the LLM outputs the code. This has been my method for over a month now. Another mistake people make is thinking they can prompt their way to a perfect result. The strength of an LLM lies in its non-deterministic nature. If you control it too much, you get poor results. If you allow it to work freely, you can have it do things it wasn't supposed to do according to its inherited prompt. This alone means that an LLM requires more than prompts of any kind to deliver. If you are coding with it, understand that your codebase is more important than any prompt in your .md files. Bring your experience with you. Since I brought my experience into my LLM use, allowing Gemini to do its work according to its capacity while I fill the gaps, it has been amazing to see what it can do.

Server hardening has always been part of my initial setup. And I have done this for 17 years and never cared about visualizing the logs on browser interface. Grep-ing and tail-ing has always been handy and still is. But first time I had to set up a proper dashboard page to visualize the logs and it shocks me what my firewall has got to deal with over the years.

@github should not have gone into AI. I just started a chat session and in plan mode, and I have reached 53% of weekly rate limit. Looking at the scrollbar you would see I havent even gone anywhere. Been using @antigravity and just decided to branch here since my limit was reset yesterday and here we are.

Built this Advanced Website Builder (Aiwa) for a client back in 2021. ​5 years later, the code is still solid, and it has just crossed a massive milestone: Over $2M in revenue generated. ​In an industry obsessed with the New Shiny Object, there is massive value in building robust, high-performance tools that just work and drive actual business results. ​What’s the oldest piece of code you have still running in production? #buidinpublic #SaaS