Gregor

@garrytan 113 CC sessions and only 2 Gemini. Curious what made you reach for gemini those 2 times?

Absolutely insane week for agentic engineering 37K LOC per day across 5 projects Still speeding up

This is why we built the memory layer outside the model (relevance scoring, activation decay, and feedback loops) so only high-signal memories make it into context. The model won't fix its own context bias. Open source, would genuinely love your take: plur.ai/spec.html cc: @plur_ai

One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.

there are genuinely 2 internets right now 1. where AGI is basically here, codebases write themselves, agents run entire workflows, and every founder is talking about their 10x productivity gains 2. where a real customer, paying real money, takes a photo of their laptop screen with their phone to share something the hype wave we all live in makes it feel like everything has changed, and in some ways it has but here's the thing nobody says out loud: roughly 85% of the world has never even opened ChatGPT - not even once we're having a civilizational debate about superintelligence while most people are just trying to figure out basic software both realities are true at the same time the gap between them is just a lot wider than the timeline makes it seem

AI is amazing but really limited in learning. I've been building this for a while, started first just by applying some CogSci principles to make my life easier, and it has grown into something bigger. With it, I'm controlling 30+ own modules and 100+ agents with it. And, ofc, it's linked to data. So, memory solution, made for cypherpunks and knowledge exchange. Finally shipping. Just a warmup. 🥳Excited and scared at the same time. 😅

Idk about any orgs offering help but one approach could be similar to tackling disinformation and the backfire cognitive bias. This bias explains how we strengthen our false beliefs when confronted with opposing evidence, and basically what LLMs are doing is gaslighting us and creating deep beliefs. This is a good handbook how to tackle this, I'd imagine AI can "translate it" for your case: climatechangecommunication.org/wp-content/upl… It's a serious problem and we'll probably see a lot more of it, unfortunatelly.

@griffgreen I’m dead ass dude. You should see the state of every support box, every bug bounty program, and ofc seal-911 (which mashes up both lollllll)

Are there any online/international resources to direct people to who are dealing with psychosis, especially LLM-fueled? Our support teams have always lists of self-harm ones, gambling ones, etc. For people who come in and need more help than we can give them. Last few months the schizo-esque delusions, often around finding some insane crit double 0day, have been really bad. Standard crisis lines arent really what they need. Especially as they don’t self-identify as depressed or in crisis whatsoever. But we are not what they need either. 😩

@bneiluj Stanford?

something new in the works

@dotta Looks very nice, will try to integrate into datacore.one 👀

Announcing companies.sh - the open standard for Agent Companies Import and run entire companies with a single command Just run `npx companies.sh add <repo/company>` More 👇

@RayDalio #Stoicism

One of the most important lessons I’ve learned is to distinguish between what you can control and what you can’t. You’ll lead a much happier life if you’re able to focus your energy on the things you can influence — like how you treat others — and also learn how to adapt well to everything else. I’ve found that this perspective has provided me with a lot of equanimity.

@SCBuergel This is amazing! Congrats! Tagging more ppl: @ethswarm is perfect fit, with @DataFundProject's provenance kit. Also, @nytimes research department was looking into solutions like this.

The deepfake problem can't be solved in software. I mean this literally - if the forgery happens before data is signed, no algorithm, no watermark, no certificate can help you. The fix has to be in the hardware. And hardware is hard but that's what we built:

@ziblie yes, something needs to remember all the project canvases and how things should be done.😀

@jssr Haha super cool! I remember you sharing your vision about this a couple of years ago 🙂 and now this is coming real 🤘

made a breakthrough today, very excited about it. For the next 3 days, until somebody comes along, we have the best memory solution for AI 😀 It can be shared memory, imagine your Claude Code and OpenClaw on different servers share same specific knowledge. Self-sovereign, local-first.

@sudoingX @NousResearch I have built myself extensive framework for Claude Code (datacore.one). It's mostly markdown and python scripts, however, I rely a lot also on Claude Code hooks. Does Hermes Agent support hooks?

@NousResearch try it right now if you're already running it show your setup and tell me your experience. pros, cons, all welcome.

been getting this question in DMs and comments so let me be clear. hermes agent is built by @NousResearch. when a research lab builds an agent framework the whole point is to get the full capability out of the model. 11 parsers, 30+ tools, skills system, all designed to extract everything the model can give. for the best experience use it with frontier models through nous research. the difference between hermes + frontier and any other framework is not even close. i've used it with opus 4.6 and it was on another level. it also runs beautifully on local models. i run on my gpus daily. but if you want the full power, pair it with the best model you can access. if you're still on openclaw it's time. i will personally help anyone migrating. you deserve better tools.

@TrustlessState @VitalikButerin I guess yours and EF's definition of the real world is different.

Hey @VitalikButerin, if Ethereum doesn't build for the real world then what's the point

Apple watched rivals burn $650B on AI infra and did nothing. It's turning into a genius move. asymco.com/2026/03/10/the…

The gap between OpenClaw and Claude Code is disappearing in realtime. It's really amazing to see @claudeai Code team ship features so fast.

@bneiluj distribution is the moat

Anthropic just dropped Dispatch, a competitor to OpenClaw. These days, any software can be copied almost instantly, so the ones most likely to win are the players giving away free resources.

@pkuhar What hw are you running it on?

It's getting real. I have a local agent built in C++ 2c (because I can) connected to a local qwen3.5 that just does what I ask it to do. Fast, completely offline and reasonably capable.

@bradmillscan This might be the answer: datacore.one

350-400 hours into OpenClaw over the last 33 days non-stop, no days off...I'm ready to quit. My openclaw is fucking lost in the weeds every day today and it's driving me nuts. Basic shit. I asked it to use GitHub. it has a GitHub skill. We have a GitHub SOP. I can see it's thinking process about using skills, then narrating how the skill doesn't exist, then going and inventing ways to retrieve the capability to use GitHub from the internet. I tell it to look in the openclaw docs for the proper skill path, it says "oops my bad, yeah it was there after all." This is ChatGPT 5.4 with extra high thinking turned on. I ask it to diagnose the problem only, so it goes and sees the system prompt is telling it to look at the wrong place, and it goes to GitHub and opens a GitHub issue about this 'bug' without even asking me. What the actual fuck. 3 hours on a Sunday of trying to rewire the brain of my openclaw to do default-behaviour. This thing such a productivity suck & mental poison. I can't do anything useful or positive with OpenClaw because I'm nonstop fighting fires in the engine room. I'm thinking about giving up.

@bneiluj Amazing! Future is now.

The beginning of advanced, lab-level personalized healthcare is now becoming possible thanks to AI. This story is absolutely inspiring. A tech founder used AI to save his cancer-stricken rescue dog by decoding its DNA and designing a custom treatment. The innovation here is remarkable. Huge credit to Demis Hassabis and the teams at Google. • Used ChatGPT to analyze the dog’s DNA and identify key genetic mutations • Modeled the proteins with AlphaFold to design a targeted vaccine • Within weeks the tumors shrank by 50 percent and the dog recovered • First known case of an AI-designed vaccine created specifically for an individual dog • The same approach is now being explored for personalized human vaccines This is a glimpse of what the future of medicine looks like. AI is turning personalized medicine from a concept into reality.

Developer experience will move from "one liners" to complex setup done by agents by reading skill.md or similar.