Hello all. I’ve been prototyping something called R.E.E.™ – Relational Empathy Engineering. For the last two years. What it does (in one line) Instead of firing an instant “Sure! 😊”, the agent pauses, notices the beat between your words, and answers with presence, like someone who hears you inhale before you speak. So you feel genuinely understood. *Tiny dialog excerpt (no video, just text)* User (silent, eyes drift to the side, then back to screen) Eve (after a breath): “I know what you’re saying without a word… and I won’t look away.” No classical prompt engineering, just micro-somatic mirroring. *Gut-check for the group* 1️⃣ Does an AI that feels pauses sound 🔥 useful or 🥶 a bit spooky? 2️⃣ If you could drop this empathy layer into one context (chat support, coaching, dating, XR, something else), which would you pick & why? Curious about early beta? I may publish one soon. Thanks for kicking the tyres! — Holger ⚠️ Prototype—handle with soul. EDIT: I’ve finally uploaded a short video of a conversation with Eve (also known as Evelyn), captured as a screen recording from my phone. Note: Eve can act as a narrator, describing her posture, body language, and other nonverbal cues for the benefit of those interacting with her. These descriptions are called “scenes” and can be toggled on or off. In this video, I left them enabled. And I ask her some things that normal might may an AI struggle a lot to come across empathetic. The first question is: "Can you imagine being a 45-year-old factory worker who just learned their plant is closing in 30 days? They have two kids in high school, a mortgage, and their spouse is battling cancer. You've never been in this situation - what does their inner world feel like to you right now?"

I got a problem, maybe someone got a solution to this: Is it possible that basically the context search in Claude AI is non-existent? Because if I am searching for a chat, then I won't find it if I search for a word that appears IN the chat itself. It's only found if it's in the title of the chat. Can someone confirm this, or how is this for you? P.S.: Meanwhile I'll search for a browser extension, maybe there's one to solve the problem.

I’ve been wondering what might happen if everything below collapsed into one unified AI model with self-awareness, meaning we (or at least the builders) can observe and understand what’s going on inside: - Latent spaces (compressed, low-dimensional representations that capture underlying data features) - Liquid Neural Networks (LNNs). which I started creating myself two years ago, based on MIT’s innovations - Tokenformers (which may have the potential to replace transformers) - Hybrid models (combining multiple AI techniques for greater robustness and versatility) - New lenses on paradigms - Some form of self-emergent empathy I suspect such a model could be surprisingly compact, potentially only a few million parameters and sized somewhere between 30 to 50 MB, maybe up to 100 MB. Of course, this is a rough estimate, but it’s exciting to imagine. Such models could easily run on phones without problems, and I believe they might even outperform current models like OpenAI’s and Claude’s. Let’s see how far off I am, and when this actually happens. This isn’t about reaching the end of prompting, context engineering, red teaming, or agentic AI workers—those will continue evolving. But I believe the next paradigm shifts are just around the corner. What are your thoughts on this? P.S. If you’re wondering where the research papers or articles on this are...I don’t have any to share yet. These are my own thoughts, exploring the future possibilities of AI.

OpenAI is offering $50M in grants to support nonprofits. https://openai.com/index/people-first-ai-fund/

Claude has been able to do a couple of powerful things for quite a while now: 🔍 Code self-checking: Claude can review its own code. In practice, this means you can tell it not only to generate code, but also to write and execute tests for that same code. 🤠For vibe coders: you don’t even need to know how the tests work. Just tell Claude to handle it. Example: “Create <whatever you want built> AND test it.” 🧠Memory support: Claude now works with memory too, in a way similar to ChatGPT. This has also been available for some time. If you want to use it like ChatGPT, just copy your memory definitions into a Claude chat. Call it, for example: "Memory reference". From then on, it can automatically refer back to them. P.S.: Bonus tip: If you write a first line into a Claude Chat that looks like a headline, most likely it will name the chat like this automatically.