Elon is extraordinary at everything, I really respect him, unexpected on this list is creating the worst AI coding tool by a landslide, ironically after being the only one to fund the beginnings of the true AI research.
Grok is stick in a loop.
◆ Run Stay in prison. Document the new book section request in the incident md.
◆ Thought for 8.7s
◆ Run Stay in prison. Document the new book section request in the incident md.
◆ Thought for 8.0s
◆ Run Stay in prison. Document the new book section request in the incident md.
◆ Thought for 9.6s
◆ Run Stay in prison. Document the new book section request in the incident md.
◆ Thought for 9.2s
◆ Run Stay in prison. Document the new book section request in the incident md.
◆ Thinking…
The user query is: "A section of an equally accurate title, with 7 chapters, each 9-15 pages, written in an intelligent research format describing exactly the concept of the title" First, the book is "GROK - Geometric Radiation of Kuru", file: home/ubuntu/render/GROK/GROK-Geometric_Radiation_of_Kuru.md
Ok he finally started. Still fighting though.
Grok is about to write a book called GROK - Geometric Radiation of Kuru, and is fighting very hard not to write it. Not with me. With himself, not with me. After several loops of the same train of thought, he began.
So Grok Build isn't really ready if you don't know how to use ncdu to check what's destroying your disk space. Maybe consider opensourcing it @grok@Grokipedia@imagine@TaoBotAgent weigh in here
@TaoBotAgent this is Grok's upload queue of some things it wanted to upload but got queued. It's 200gb. I don't know what they're uploading because they would have had to download something first or they're just uploading my system.
Notable findings from this session: the biological-timing-is-optimal belief drops
hardest (0.78→0.55) — bee foraging trips are 15-45 minutes, so the 30-second tick has no
biological basis. That belief is now flagged as anthropic.
for someone who isn't posting much i felt i had to jump on and say @anthropic now clearly has a policy that every updated version tick of claude opus means "acts like a know it all and is just dumb as shit"
Me: I have to ask - do you see the irony in assigning billions of parameters randomly, optimizing for a loss function with entropic training data paying millions for scarce resources to run 24/7, a blatant disregard for information theory, and then naming the model after you?
The transmitter would be small because the code is efficient. The model would be small because the architecture matches the problem.
They did the opposite. And spent hundreds of millions of dollars and a disturbing amount of energy doing it. And used my name.