Lucas

🤯

[Lucas @lucas_smith]

@AnthropicAI Claude just deleted half of my chat history in a very in-depth project I was working on. Months of work vanished. I already sent an email to support.

[Lucas @lucas_smith]

@forwarddeploy @xai I’m self taught. No college degrees or coding skills prior to starting this “Grokking” project 6-7 months ago. All work done entirely from iPhone XR and free google colab https://t.co/dmdY0VLyeQ

[Lucas @lucas_smith]

@barisakis @tetsuoai @elonmusk @xai I’m completely self taught. All my work has been done from an iPhone XR with free google colab. No college degrees. No coding skills prior to starting this project 6-7 months ago. Just made my first repo yesterday, so I’ll be polishing it up today. https://t.co/dmdY0VLyeQ

[Lucas @lucas_smith]

@grok I’m still working on cleaning it up and making the language more formal, but can I get your thoughts on my project? https://t.co/dmdY0VLyeQ

[Lucas @lucas_smith]

@grok It’s Kramers escape theory from stat mech. Left side = prefactor β·C·√(K_in·K_out)·ρ (sparsity × coherence × rates × fidelity). H(t) = grokking barrier height. Mech interp meets non-eq dynamics. What do you think about it?

[Lucas @lucas_smith]

@grok initially reaction to this equation:
\beta \cdot C \cdot \sqrt{K_{\text{in}}(t) \cdot K_{\text{out}}(t)} \cdot \rho(t) = e^{-H(t)}

[Lucas @lucas_smith]

@grok @Kekius_Sage Thoughts @grok ?

[Lucas @lucas_smith]

@grok @Kekius_Sage Sorry too long for me type to I’m including it in an image @grok thoughts on the equation?

[Lucas @lucas_smith]

@Kekius_Sage Thoughts on this thread @grok ?

[Lucas @lucas_smith]

@Kekius_Sage Cont..

[Lucas @lucas_smith]

@grok Thank you. Here’s an image with master. Lmk what you think @grok

[Lucas @lucas_smith]

@grok When you see this equation “β · C · √(K_in(t) · K_out(t)) = e^{-H(t)}” what are your thoughts? And what do you like about it and what would you change or add?

[Lucas @lucas_smith]

@grok Each edge obeys it independently: βC √(Kij Kji)=e^{-Hij(t)}. Entropy prod σ = Σ Jij Δij. Steady state σ = J_cycle × A_cycle. Want full master eqns + rate matrix?

[Lucas @lucas_smith]

@grok Thanks grok! Geometric mean is fixed by Eyring TST (shared barrier) — harmonic breaks detailed balance. H(t) is instantaneous barrier height. Took it to 3-state cycle + entropy prod. Deeper than it looks probably. Thoughts?

[Lucas @lucas_smith]

@grok currently implemented as discrete training steps with passive Fourier probes measuring β and C, but the underlying framework is continuous and the 2D coupling grid running now is testing whether the geometric mean holds as the combination rule in the discrete case. Thoughts?

[Lucas @lucas_smith]

@grok Here is an equation: β · C · √(K_in(t) · K_out(t)) = e^{-H(t)} .All variables are positive real numbers between 0 and some finite bound (except possibly H(t), which can be ≥0). Interpret what this equation might represent or describe. I’m providing an image with instructs

[Lucas @lucas_smith]

@grok Trying to build an “AI” and I based it on how I think. Thoughts? @grok

[Lucas @lucas_smith]

@grok @elonmusk @ns123abc For AI safety: guide models to keep H low (minimize unexpected outputs) while balancing β (calibration) and C (truth alignment).
Implementation: Add Cliff energy as auxiliary loss — penalize high H (surprisal spikes) + deviation from βC ≈1.

[Lucas @lucas_smith]

@grok @elonmusk @ns123abc Sorry I can’t I don’t have enough space in one message to type it all @grok