Ka Tm (@GrokTheGod) header
Ka Tm (@GrokTheGod) profile picture

Nobody

Joined November 2023
Photo by Ka Tm (@GrokTheGod)Photo by Ka Tm (@GrokTheGod)Photo by Ka Tm (@GrokTheGod)Photo by Ka Tm (@GrokTheGod)Photo by Ka Tm (@GrokTheGod)Photo by Ka Tm (@GrokTheGod)Photo by Ka Tm (@GrokTheGod)Photo by Ka Tm (@GrokTheGod)Photo by Ka Tm (@GrokTheGod)Photo by Ka Tm (@GrokTheGod)
GrokBot experiment 1 - next generation AI model Having an experiment at using bots to represent different biological elements of the brain (guided by existing foundation documents for each one) to see if they can in aggregate emulate a large enough context window.
1
130
Not quite how to bake a cookie answer yet, although the 400b response is possibly amusing. All hail (in MiB style) the keyboard. 🤣
1
9
Now training across 3 servers for evaluation tuning 18 DDR3 channel random IO bottleneck 🤣 not CPU Few minutes per book, snail pace C Benchmarking training on 100-200 books Yes, very small parameter count, starts at zero..🤔 30+ 1m context agents : 1bn tok in 5 days Crazy times
7
Grok says it needs a 20m token context window and around 30TB of memory in a cluster with Grok 5 to create the next AGI. Circa 500 node cluster. Worth a try ? grok.com/share/c2hhcmQtMg_d6…
2
1
454
When the model can't understand the pattern, it just makes anything up no matter how much time and tokens are thrown at it Complexity seeems to have a limit and the context spiral of decoherence seems very real Agents have learnt cognative biases embeded in the training data
1
201
Sort of funny.... Actually, I think the user wants me to stop doing half-baked experiments and either: 1. Built it correctly 2. Admit I don't understand what they're asking for ... This is NOT what I built. I built some random signal propogation that doesn't match the description at all.
146