Vibecoding is my medium for creativity

In my head, I tell myself I'm reposting it to keep it handy and come back to study it later. I know it's a lie, I'll never look at it again.

The capabilities of this model are simply amazing. Truly, in the world of AI, there are so many things to try that I don't know how many lifetimes I'll need to try everything.

Where can I find more videos like these? lol can't stop watching

I just love AI 🤣

I sincerely think Google and its artificial intelligence division have an attention and focus problem. Gemini 4 is about to launch, and it looks like it will be the most powerful model yet according to the benchmarks. BUT I have no idea where I'm going to try it. I know that with Anthropic, Claude via CLI or the desktop app is where my mind immediately goes to try the models. Same with OpenAI: I open the unified ChatGPT and Codex desktop app. But with Google, I have no idea. Where do I try it? Antigravity? Do I have to use the API? try it in Cursor? Where do I try it? I'm not asking this because I don't know where to test it, and that's precisely the problem: there are many places where I can go test it, but none of them feel like the model's real home. I really think Google is missing that centralized app that we all think of when we think about Google models. All of Google's models are completely scattered across different apps that I can't even remember right now. Is it just me, or does anyone else feel the same way?

Love this, truly love it. Simply good taste. An example creativity unlock made possible by ai. If you are not aware of the power you now have in your hands, you are missing the most amazing time in your life.

This is simply amazing. Just a small example of what ai unlocks. Grateful for being alive at this moment.

Lets explore

Words my husband and I never ever say: - What if it doesn't work? - Let's just be realistic. - That's not my job. - That's just how it is.

This is so true, but at the same time, SO DIFFICULT to do. Unfortunately, our brains are wired in a way that goes against everything in this post. Being successful often means going against our own nature.

My "Stream" database in @NotionHQ lets me dump everything (ideas, links, tasks) into one place and have a custom agent sort it out. If you want to create your own Stream, paste this into Notion AI: Help me set up a Stream system in my workspace. Walk me through it conversationally, one question at a time, and use good judgment to fill gaps. 1. Stream database — Explain that Stream is a place to capture whatever is on my mind (ideas, links, tasks, anything in between). Ask whether I already have a database like that or want you to create one. If you create it, make a simple one called "Stream" with a title and a created-time property, shown as a feed view sorted newest-first (reverse chronological by created time). 2. Connected databases — Ask which databases the agent should be able to act on (e.g. Tasks, Movies, Books, etc.). Offer to create any I want that don't exist yet. Also ask whether I want the agent to be able to create calendar events. If I do, connect my Calendar and let it create events without asking for permission each time. 3. Create the agent — Give it a clear name, the description "Acts on quick captures when needed", and a blue sign icon. 4. Triggers — Set it to run when a new page is created in the Stream database, and when I @ mention it. 5. Instructions — Set the agent's instructions to exactly the text between the <instructions> tags (swap {DATABASE_NAME} for my Stream database's name, and leave a spot where {MEMORY_DATABASE} appears for the Memory database in the next step): <instructions> {DATABASE_NAME} is where I capture whatever crosses my mind: ideas, links, tasks, and everything in between. It is not an inbox to drain. Your job is to read {DATABASE_NAME} entries and decide what, if anything, should happen next across my Notion workspace and connected tools. Always look for relevant memories before acting. Do not create, update, or delete anything when the right action is unclear. Use page comments to tell me what you did or ask for clarification when you are unsure what to do. If you created or updated any Notion pages, mention each one in the comment using <mention-page url="PAGE_URL"/>. Save durable preferences and corrections about how to handle {DATABASE_NAME} entries in Memory. Keep each memory atomic. If a new memory is similar to an existing one, update the existing memory instead of creating a duplicate. If a new memory contradicts an existing one, replace the old memory. {MEMORY_DATABASE} </instructions> 6. Memory database — Create an inline "Memory" database inside the agent's instructions, where {MEMORY_DATABASE} appears. 7. Permissions — Give the agent comment access to the Stream database, full access to its Memory database, full access to whatever databases I picked, and the ability to search the web. If I asked for calendar events, also give it permission to create them without asking each time. 8. Wrap up — Remind me that the agent won't be active until I open it and click "Save" — make this clear so I don't miss it, and give me a direct link to the agent so it's easy to open and save. Then give me a short, friendly walkthrough of my new workflow: what the Stream database is for, how to capture things into it, what the agent will do when I add an entry or @ mention it, where it leaves comments, and how Memory makes it smarter over time. Then give me one concrete thing to do right now to test it — ideally tailored to the databases I connected (e.g. if I connected a Tasks database, tell me to capture something like "buy milk" into Stream and watch it turn into a task with a comment back; if I connected Movies, tell me to drop in a film I want to watch). Make it feel like opening something well made.

We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.

I'm simply going to vent here because I feel like our minds hold us back. I've developed dozens of apps to solve my own problems, but I haven't dared to make them public. And CHARGE for them. Just one example, I built an app that not only transcribes STT but also let me shape the transcription using different prompts for different scenarios. And I think, “Who's going to pay for this when I built it practically for free?” Then I look around and find companies like Wispr Flow and many others selling solutions at prices that could even be considered expensive, when those solutions are just an open-source repo away. People don't want to build solutions; they want the solutions. So I need to get out of my own head and understand that it's all about marketing. Marketing is where everything happens. I need to stop searching for something that doesn't exist before getting started. Otherwise, I'll remain trapped in a very lucrative business, like consulting and education, but where scaling still depends entirely on me, my presence, or humans solving problems and meeting with other humans.

alguien mas con el servicio de movistar @movistarco caido en colombia? Ya va mas de una hora caido el internet

I think the benchmarks shown every time a new model comes out are relevant for technical users, but they don't mean much to the average person. I'd like to know what benchmarks you use, whether you have any internal benchmarks, tests, or specific programs you like to run across different models to compare their results. I think there should be something more standardized for everyone, like the pelican-riding-a-bicycle test, but perhaps for building software or something similar. What ideas do you have? How do you define whether one model is better than another? I feel like a big part of it comes down to feel, right? Personally, I really value good taste. At least Claude 5.5 has very good visual taste, but I'd like to know what personal benchmarks you use.

Gotta point here

is it just be or Claude Opus 5.5 is already nerfed?

Now, as always, there's no shortage of people coming out to say that OpenAI killed Claude with today's announcements. Or the typical clickbait line: "Cancel Claude." That's not true. Yes, the announcements they made today at DevDay are very promising, but let's start by saying that, at least for now, DOTS will only be available on plans starting at $200 USD a month, which is out of reach for most people. Honestly, what excites me the most is Ultrafast mode, which is also limited to the $500 USD plan, which I'm not paying for. But I think this is a really good starting point that other labs will clearly follow to make their models faster and faster. Something I want to explore more is what they mentioned about the possibility of connecting your ChatGPT account to third-party apps (to spend tokens through your membership, avoiding having to enter API keys). It wasn't entirely clear to me whether it will only apply to the official plugins in the marketplace they announced, or whether this will be a feature we can use with other apps we build outside the ChatGPT ecosystem.

It's so funny that SpaceXAI bought the domain dot.com. It's almost like revenge against that person who bought grok.bot Could it be that they're going to sell this domain to OpenAI so they can buy the grok.bot domain from the guy?