Started makemore today (Karpathy's series). First model: a basic bigram.
It counts how often one character follows another, turns those counts into probabilities, and samples new names. That's the entire model.
Code + actual output from my run 👇
@kzitouni1 I’m studying to be an AI engineer. I know the competition is tough but I’m trying to go past just vibe coding and harness engineering and actually solve real problems with machine learning and every set of eyes on my work is another reason for me to keep going.🤖
Productive to an extent
The easy access to deep research that used to take hours now condenses into single outputs. That’s made my learning drive 100x more curious.
Organizing thoughts into complex projects is the next biggest boost.
But if you’re not focused, it’s easy to fall into an architect loop: endlessly improving the system in concept, without grounding in the foundations. That usually leads to one of three traps:
•Rushed builds → imposter syndrome when it inevitably breaks on day 3 or at scale (consequences hit later, big productivity loss)
•FOMO in the agent arms race → constant anxiety that someone will ship first, leading to overthinking, skepticism, and becoming a follower chasing other people’s projects (consequences now)
•Endless learning without execution → slow but steady progress while feeling left behind, like you’re studying for nothing (consequences now)
Stay disciplined or AI will amplify your worst habits.
Thank you for attending my Ted talk
@ramxcodes The thing is you can learn at the same time an agent builds at but you need to the neuro plasticity and ground work to know what questions to ask and how to ingest the information