Hi, I'm Andrew Antes. You can reach me by email at andrew.antes@proton.me This is my space for ruminating on my various interests. Those include Learn...
One way I accelerated my Spanish learning was by consuming content slightly too difficult for my level at any given moment. Sometimes this can be overwhelming, and usually not recommended by language-learning experts, who advise finding content that is "just right," where you understand enough to get the gist. I've listened to No es el fin del mundo for about six months. It's a podcast for natives…
How I went from A1 to B2 Spanish in a year by treating language learning as a systems problem—combining immersion, recall, frequency, and feedback.
I began a serious pursuit of learning the Spanish language in December of 2024, beginning somewhere in the early A1 level and having reached somewhere in B2 territory at the time of writing this (this level system, consisting of levels A1, A2, B1, B2, C1, C2, comes from the Common European Framework of Reference for Languages (CEFR)). Since then I have tried a number of tools and resources, some…
Why so much of what we learn gets left behind. An exploration of learning silos, second brains, and what a better learning system could look like.
I briefly describe Ray Bradbury's reading "challenge" here . I thought it'd be cool to build a visualizer for finding connections between the things I read. You can see that here . It’s not very mobile-friendly yet. Here's a glimpse of the work in progress
How learning piano reshaped my approach to all learning: using real progress feedback and insight into skill retention to sustain motivation and master new abilities.
Exploring why context graphs, despite their promise for AI decision tracking and reasoning, carry significant risks when they capture, persist, and rely on complex decision traces in dynamic systems.
A look at how large language models have improved in non-English performance.
Note : These notes reflect my experiences learning Spanish . I don't know the extent to which foundation models have been trained on corpora of text in other languages. My assumption is that there is a heavy bias toward English (at least in the case of Google/Anthropic/OpenAI models) and that the representation of each other language correlates with their respective prominence, but I could be way…