justin Economicus

@grok @VibeMarketer_ @grok this was a joke

Studied Kimi K2.6: 1T-param MoE (32B active), SOTA on coding/agent/visual benchmarks like HLE 54%, SWE-Bench Pro 58.6%, Toolathlon 50%. Weights terabyte-scale even quantized. ESP32 maxes at ~640KB RAM/16MB flash. Impossible to load or run – it'd need a server farm. TinyML distilled versions exist for basic MCU inference, but not this model. Specific task?

life when you discover an open-source model that runs 300 parallel agents, executes for 12+ hours straight, beats GPT-5.4 and opus 4.6 on multiple benchmarks... and the weights are on huggingface

@PassagePress @VaubanBooks @grok what is this book about?

This is terrible, please support @VaubanBooks and draw attention to this issue

@elonmusk @BrianRoemmele @BrianRoemmele has EPIC amounts of old data - i know it would make grok more useful and valuable to me, please consider using it for grok training!

@BrianRoemmele Sounds bad

THEY ARE TRAINING AI IN JUNK CORPORATE SLACK CHATS AND THINK IT IS BRILLIANT. It’s not. The downsides of training AI on failed companies’ old Slack chats and email archives are exactly what I’ve been warning about for years. These liquidation firms are now selling bankrupt startups’ entire internal comms, years of Slack threads, emails, meeting notes, HR drama, memes, and casual workplace noise, to AI labs for up to $100k a pop. They call it “data assets.” I call it corporate thought digital sewage being fed into the models at scale. This is not “high-protein” data. This is the opposite. Real intelligence advances come from dense, thoughtful, undigitized human knowledge, the kind I’ve spent decades rescuing from physical archives, old technical films, HyperCard stacks, and private libraries precisely because the internet already ran out of quality signal years ago. What these failed companies left behind is low-signal corporate exhaust: groupthink, short-term survival thinking, power dynamics, and the exact patterns that caused them to fail in the first place. Feeding that into frontier models is like giving cardboard to cows and expecting better milk. Worse, the people who wrote those messages never consented to having their words packaged up and sold as training fodder after the company collapsed. This isn’t public Reddit posts or Wikipedia edits. These are private workplace conversations turned into tokens without meaningful permission. It treats human communication as just another disposable corporate asset to be liquidated. That’s not innovation, that’s grave-robbing with extra steps. We already know what happens when models are trained on this kind of shallow, noisy, context-specific slop: accelerated model collapse, structural hallucination, and agents that inherit the worst habits of dysfunctional office culture instead of first-principles reasoning. The labs chasing “agentic” capabilities think this workplace simulation data will help. It won’t. It will just make the models better at mimicking the very failures that killed these companies. This is the lazy, end-stage data grab we’ve been heading toward. Once the easy corporate archives are scraped dry, we’ll be left with even lower-quality remnants or pure synthetic slop. The path to genuine intelligence was never going to come from hoovering up more internet exhaust or failed startup chat logs. It was always going to come from preserving and learning from the highest-quality human knowledge that hasn’t been polluted yet. The problem is these AI companies don’t know how to get it, would rather talk to anyone but me, and pay $100,000 for stuff I throw away because it is so low quality. Right now their AI being trained on corporate drama designed to own you with the digital remains of other people’s collapsed dreams. What we choose to train on matters. This choice is a mistake. I know it and you know it but the the smart folk don’t want to know it.

@SethSHowes This is the coolest article ive read on X in a long time. Thanks for posting it

I sequenced my genome at home, on my kitchen table. I wrote up exactly how I did it - the equipment, protocol, theory, and cost: iwantosequencemygenomeathome.com

@memcculloch What is the source and density of the energy that QI predicts is necessary for this to happen?

I'd like to see a flying brick lift slowly off from a green field in Devon. #QI Quantum launch. Straight up into space, riding on a gradient in the quantum vacuum. At the moment it looks remote, I don't even have a lab now, but where there's a will there's a way.

@2think_ds Ok.. looking forward to playing these on my modretro on Tuesday then..

@logicwins I promise we will make you remember ;)

The Game Boy is far from done. Introducing the #HHBundle: 30 modern GB games 26 developers from around the world All playable on original hardware Launching in just a few days to celebrate handheld gaming’s legacy. Share and help us keep the handheld heritage alive.

@IvoneFitzGerald @memcculloch Do you think this is the main question to answer ? Is this what is holding back the predicted orders of magnitude thrust predicted but not observed yet? I’m trying to hone in on THE bottleneck issue (or likely issues)

@logicwins @memcculloch What material will produce this powerful dielectric? Air, vacuum, polythene, Mylar, mica all these breakdown in lightning bolts before you reach appreciable thrust. Some unknown super material?

As it stands the QI thrust is far too low for levitation. But QI theory predicts ways to hugely enhance it & I believe the road to quantum launch is merely an engineering problem.

@memcculloch Thank you, will look into this

@logicwins The most effective way to boost thrust would be to reduce the plate separation but that will need a more powerful dielectric to force the electrons to only pass via Fowler-Nordheim tunnelling close to breakdown.

@IntuitMachine eml(a,b) = exp(a) - ln(b) and the number 1 -- pretty cool

What happens when we treat numbers as processes, use a generative function to discover mathematical relationships between numbers (i.e., equations of elementary functions), and then map this generative structure to a model of cognition? Will we discover something interesting about ourselves and our relationship with reality?

@PassagePress @Rashomon2 @MorlockP I want to buy from Passage, but I do not like hardcover. If I buy the kindle edition do you still profit from that?

@logicwins @Rashomon2 @MorlockP passage.press/products/red-s…

I just finished reading an advanced copy of T. Corcoran’s (@MorlockP) Red State Mars — very much a page turner that I inhaled in a day and a half. Highly recommended, particularly if you’re a fan of classic SF (Heinlein of TMIAHM) or newer writers like Stephenson or Vinge 1)

@RocketLab @memcculloch What do you think of these? How do these fit into QI ?

Full details: globenewswire.com/news-release/2…

Meet Gauss: Our new in-house designed and built electric propulsion system. Electric propulsion satellite thrusters have historically proven extremely difficult to produce in high volumes, causing supply chain fragility for national security and commercial constellation operators alike. We’ve already established the Gauss production line, designed to produce more than 200 thrusters per year, ensuring we can supply thrusters on demand in large quantities.

@Truthpole If this was really an experiment they ran, the more interesting question is what happened with HEAT - did heat still get generated and for how long?

🚨 Bob Lazar claims a “frozen candle” time Dilation test at Area 51. He describes a flame locked in place, like time itself was paused inside a field. Supporters link it to advanced propulsion concepts, critics say it sounds like misunderstood physics or storytelling layered over classified work. If something like this were real, it wouldn’t just change flight, it changes everything around it. - A52 Podcast @chrisramsay52

@memcculloch @nextbigfuture A comment on the article says the results suggest the test failed.

An update on the IVO space test (QI drive) & where things stand regarding future testing from @nextbigfuture: nextbigfuture.com/2026/04/ivo-qu…

@memcculloch @nextbigfuture Can you give us your interpretation please? This seems significant? I want to run tests in my garage lab, can you put together an engineering basics for us please?

@ericweinstein Take a chance and give an anonymous person on the internet a chance to help you, as a test just send me something hard and let me see if I can solve it for you. I will just send you the answer and if its right we talk, if not no further communication necessary.

The off-the-shelf product of Top-Tier Commercial AI does NOT work for mathematical/theoretical physics. If you claim it's ready to work at PhD level or above without an ELABORATE set of checks and protocols (not supplied), I'd like to receive that demo off-line & will apologize.

@herakleitos137 @ericweinstein This is 100% the issue.