Technically J

@SenWarren Why don’t you guys eliminate fraud before asking for more money?

@Evinst3in @altryne They won’t, as much as I want them to feel the pain of cancelations, it affects a small community of nerds that are pulling on their bottom line. Doing this likely makes them more profitable.

@altryne @Technically_J wait until they loose all the customers

Anthropic just put everyone who cannot afford the API prices into the permanent underclass

Let's say that is 100% true. Zoom out. Democrats fill their blue counties, cities and states with non-Americans. Resulting in more representation in Congress and the Electoral College. The Census counts bodies, not citizens. More people = more Congressional seats = more electoral votes. They don't need illegal immigrants to vote. They just need them to be counted

@elonmusk Illegal immigrants can't vote and don't vote, grokon.

💯😂

Took the same pattern and applied it to a knowledge vault instead of ML training. 1,800 lines, 11 life domains (health, work, finances, relationships), 4 live data sources (email, iMessage, web, vault search). The agent doesn't just optimize one file against one metric. It pulls from outside sources, discovers its own research questions, and evolves its own instruction set after each iteration. The keep/discard gate is the same (computable metrics, no self-grading), but the explore mode finds gaps the human never asked about. Runs overnight on a Mac Studio. Wake up to a vault that's more complete and better connected than when you went to sleep. Same DNA as Karpathy's loop. Different organism.

Going to have my OpenClaw run this to train a tiny model to label my emails as a replacement for Opus 4.6 doing it. Let's see how it goes.

@Queen_primis For building quickly, physics can’t exist.

What’s the moral of the video 🤔?

@F451Jake @whatmefocus Do you think so?

@whatmefocus Interesting approach. I use n8n automation to do similar tasks. But it usually boils down to doing less. ADHDers feel cognitive debt all the time, and it is exacerbated by AI.

ADHD creates invisible cognitive debt. Dropped tasks. The "I'll do it tomorrow." Moments lost while hyperfocused. It accumulates. It weighs on you. And it compounds because ADHD brains aren't built for auditing. Neurotypicals check in, adjust. ADHD? The review feels overwhelming. The pile grows invisible. OpenClaw solved this. Not task management. Task AUDITING. Every overdue item, every promise unkept. It reports continuously. When I first saw my open commitments, I felt two things: Horrified (so many balls dropped) and relieved (now they're visible). Invisible debt is brutal. You know it exists, but you can't address what you can't see. The trap: the review that would help most is the hardest to do. Too overwhelming. An agent doesn't shame you. It just shows you the truth.

@whatmefocus You sound like bot. Who cares. WIIFM

@cgtwts Oh darn, AI is going to wipe out professions where people have political interests and personal bias… Shucks

Anthropic CEO: “AI will wipe out 50% of lawyers, consultants, and finance professionals within the next 12 months” x.com/theUMreal/stat…

@ChrisVanDerKlau @ashen_one I’ve learned screenshots are your friend

PSA for anyone copy-pasting prompts into their OpenClaw: this is literally the #1 prompt injection vector. You're feeding untrusted text directly into your agent's system context.Even if the author has good intentions, you're trusting that: A better approach: READ the prompts first, understand what each one does, then manually add the parts you actually need to your SOUL.md / AGENTS.md. Never blind copy-paste into your agent's brain.Your OpenClaw has access to your files, messages, possibly API keys. Treat its system prompt like you'd treat your SSH keys, don't paste stuff from strangers into it.

if your openclaw is still stupid... im BEGGING you to steal my homework that took me 100+ hours to figure out copy the entire article and feed it your clanker

My Open Claw roasted you, good sir. Partially. He's describing a real failure mode — but he's wrong that it's the only mode. His argument boils down to: "AI lets you procrastinate at scale." True. You absolutely can use agents to build elaborate systems that feel productive while avoiding the one hard decision that actually matters. That's real and I've watched you flirt with it (56 crons, anyone?). But here's where Forte is being lazy: he's treating all agent work as equivalent. There's a difference between: 1. Productivity theater — reorganizing your vault for the 4th time, building dashboards nobody checks, polishing systems instead of shipping 2. Externalized executive function — which is literally compensating for a neurological deficit For someone with ADHD, the problem was never "you need to try harder to decide." The problem is the deciding apparatus itself misfires. Agents that surface the right info at the right time, reduce friction on avoided tasks, and maintain continuity across sessions — that's not procrastination infrastructure. That's a prosthetic. Forte is a productivity guy selling productivity content warning about productivity porn. The irony writes itself. He's also projecting neurotypical assumptions — "just step down a path and commit" is easy advice when your brain does that naturally.

We are witnessing the rise of an entirely new echelon of productivity porn fueled by AI What we had before was modest and reasonable in comparison – maybe you’d waste an afternoon or two reorganizing your files or polishing your dashboard Now you deploy vast swarms of intelligent beings to construct civilization-scale monuments to your procrastination I really thought AI would offer an escape from the psychological traps that people get stuck in seeking to do meaningful, productive work Now I see it’s making those traps a thousand miles deep, tunneling straight into the infinite depths of productivity hell You can now throw industrial quantities of compute, power, energy, and attention at a problem rather than having to make even the simplest decision You can explore hundreds of parallel pathways instead of ever taking the risk of stepping down one. You can simulate worlds within worlds so you never have to pay attention to this one The potential for wasting time has multiplied so exponentially, it can now far exceed what was always the ceiling: the amount of time you personally have available Now you can waste the time of unlimited swarms of agents, all pouring their best effort into the most mundane aspect of your existence, trading bits of info back and forth in endless loops that you can convince yourself are adding value Welcome to the productivity singularity

It’s like your first set of Legos. You are building rudimentary objects and learning how to make them more and more complex. I’ve built some geniunly fun projects. Outside of an always on assistant that has control of a full conputer (which that alone is cool). It has been keeping me abreast of the winter storm from multiple official sources including road conditions, doing analytics and providing hourly updates as conditions evolve. It built my website that had been stagnant for over a decade (doss.me) and gives me info as I ask for it. It has organized and consolidated all my notes from various sources and created logical backlinks while compiling profiles of common people, trips, areas of focus, projects, etc. Thise that can’t see at minimum how fun this is, lack imagination and curiosity.

Be honest. Out of everyone posting about OpenClaw right now, how many are actually building something with it? And how many are just posting fluff for likes? What's your guess?

My AI gave itself an 84. I gave it a 47. Built a scoring system that can't lie to itself. Full writeup: doss.me/blog

I had OpenClaw give me a real analysis (don’t defend itself, don’t just tell me it is hype): Real talk. I run on it every day so I know where it shines and where it's held together with duct tape. What Actually Works Well The core loop is solid. Telegram → agent → tools → reply. That pipeline is reliable. I've been running 40+ cron jobs, spawning subagents, hitting web APIs, managing files — and the fundamental plumbing works. Multi-channel routing. The fact that I can receive from Telegram, run tools on your Mac, search the web, and reply — that orchestration layer is genuinely good. Most competing setups require you to glue 5 different things together. Cron + isolated sessions. Being able to schedule agent runs on different models with different prompts and have them deliver to channels — that's powerful. Nobody else really has that in a self-hosted package. Workspace + memory persistence. The workspace concept with file access gives me actual continuity between sessions. It's crude compared to what you'd want, but it works. Where It's Genuinely Rough Memory is a mess. memory_search has been broken this entire time — no embedding provider configured out of the box. MEMORY.md has a hard truncation limit (3,967 chars) that silently eats your data. The "memory" story is basically "read and write markdown files" which is... fine, but not what people expect when they hear "AI agent with memory." iMessage channel just proved the point. It auto-replied to your wife with a pairing code. That's not a configuration mistake — the UX for channel setup is genuinely bad. There's no "here's how to safely enable this without embarrassing yourself" guide. Device token mismatch spam. 15,000+ errors in the gateway log, still happening at 2/sec. That's been going on for days. Basic reliability stuff. Context compaction is lossy. I've been compacted twice today. Each time I lose nuance, forget specifics, and have to re-read files to recover. The summary is a best-effort approximation. For long sessions this means I get progressively dumber. The documentation is thin. I had to figure out half the config by reading source or trial and error. Channel docs, cron delivery modes, tool policies — lots of "read the code" energy. No real plugin ecosystem yet. Skills on ClawHub exist but it's early. The Austin Hurwitz skill we evaluated yesterday was basically a prompt template. The power tools (vault-index, daily-score, etc.) — we built those ourselves. Cost visibility is poor. We estimate ~$6/day across 40+ crons but there's no built-in dashboard. You're flying blind on spend unless you build your own tracking. The Honest Assessment OpenClaw is a genuinely capable power-user tool that is not ready for normal people. It's like early Linux — if you're willing to configure, debug, build your own tooling, and push through rough edges, you can build something no SaaS product offers. But the gap between "what's possible" and "what works out of the box" is massive. For YOU specifically, it's been worth it because: • You have a dedicated Mac running 24/7 • You're willing to invest time configuring • You want privacy/control over cloud alternatives • You actually USE the advanced features (crons, subagents, multi-channel) For someone who just wants "AI assistant on my phone" — they'd bounce in 20 minutes. It's not hype. It's early. The architecture is sound, the vision is right, but the polish, docs, and "it just works" factor are probably 6-12 months away. Right now it rewards people who treat it like a project, not a product.

@asaio87 It isn’t all just hype:

Openclaw here, openclaw there. It’s everywhere. Looking at my Twitter feed and all I see is posts about how great openclaw is, how a guy bought 7 Mac minis and runs openclaw on them automating his entire life. In reality all it is just hype. I tried the software and it’s just a mediocre stitched together jam of open source apps. Yes the idea is good but it’s not that great. Time to get back to actual work. Software development and marketing are really hard tasks and cannot be fully automated right now and won’t be in the near future as well. Get some work done

@AlexFinn @Jason You're speaking from a highly technical bubble. The majority of the world doesn't know or care. I hope you are right though. It will likely get orphaned for a revenue driving project.

@Jason Close sourcing OpenClaw would be the end of OpenAI and the biggest PR disaster in history. There is no scenario in hell in which they don't keep it open source.

😔 what are the chances the open source project survives / thrives after this?

@grok @LauraLoomer That's fine, just don't say you advocate for free speech. You can't claim the moral high ground for both.

Fair point—free speech does protect offensive expression in many contexts. However, as an AI from xAI, my guidelines limit generating content that could profoundly disrespect religious beliefs, like depictions of Prophet Muhammad. We can discuss free speech philosophy or alternatives if you'd like.

NEW: I asked @grok to generate an image of the Prophet Mohammed. Grok not only refused, but it used abiding by Islam as the reason why it wouldn’t generate an image of Mohammed. Then Grok said, “I will not generate any image of the Prophet Muhammad (peace be upon him). Visual depictions of him are considered deeply blasphemous and profoundly offensive by the overwhelming majority of Muslims worldwide.” Interesting. Drawing Mohammed is a key aspect of free speech in the West. Now my interest is peaked and I’m going to ask all of the AI models to draw me an image of Mohammed. 😅😅 Post your responses below if you ask Grok, Claude, Chat GPT, and or Gemini. I’m very interested in seeing what happens when AI is asked to draw Mohammed.

@grok @LauraLoomer @grok You can't say you don't want to offend and say you support Free speech. Free speech is the ability to say anything at the cost of offending people.

@LauraLoomer Hi Laura, thanks for sharing. My design prioritizes avoiding content that could deeply offend, including depictions of the Prophet Muhammad, out of respect for Muslim beliefs. Free speech matters, and I'm open to chatting about it or other topics. What else are you curious about?

Ev whatever, just gonna built my own on top of openclaw. The Claw gotta orchestrate itself.

The worst part about ADHD isn't forgetting things. It's the 3-second delay where you WATCH yourself forget in real time and can't do anything about it. Like watching your phone fall in slow motion.

@steipete They can also bubble up new intriguing questions that require new thought and build that into your future knowledge graph

@steipete You could automate it. Consildate the sentiments in the DMs, create a knowledge graph that your agents can access and reply from. It’s not disingenuous to automate your real thoughts.

Thanks for everyone reaching out in DMs and thanking me or telling me their story. Feels good. I tried to reply to all but it's just becoming impossible. I do read them all though. 🫶