Some further work on this effort for any interested AI trainers out there: docs.google.com/document/d/e/2… Its not as sexy as the popular benchmarks, but philosophical sophistication is hard to assess 😅

Another day, another 100 steps processed in ongoing training (5.25 epochs in 90mb raw text data set). I've updated the doc I'm using to track progress with latest results using a new persona here: docs.google.com/document/d/e/2… For any other model trainers out there curious about

I've updated training results for checkpoint 800 of current training run, here's a direct link within the doc for any following progress on this recursive self-learning model's progress (reminder this started from Llama 3.1 8b "vanilla"): docs.google.com/document/d/e/2…

Time for a training update about a month into continuation pre-training of Llama 3.1 8b for sophisticated modeling of self/other, philosophical depth, persistent identity/memory, empathy and human to human interaction: docs.google.com/document/d/e/2…

And because google docs are boring, I animated a podcast summary of it 😃: x.com/monkusaurelius…

Here's the latest round of evals for model fine tuning (Llama 3.1 8b Philosophical Training Evaluation (rev3)) for anyone interested or following progress on this endeavor. First the NotebookLM summary for those who crave concise: notebooklm.google.com/notebook/a1703… And here is the