hm initially thought grok3 was pretty good, but have been throwing some different questions with thinky mode and research today and really impressed! just cancelled chatgpt sub.
one of the best educations you can give your kiddos is to buy a bunch of raspberry pi's and a 3d printer and teach them to make stuff. kids should be excited about space, building their own tools and learning how things work.
its becoming clear that OSS AI is going to crush closed source and i have to say its a relief. it doesnt matter what country it comes from, the people should own AGI, not one single country or corporation.
deepseek r1 is batting near 100% on inferring speaker names from long form audio. transcribed with whisper + diarization from pyannote audio. interesting use case and quite handy
in redteaming, we learned a long time ago its far better to have the threat be known vs. unknown. hence open source tooling proliferation. there is significantly greater risk to only a few having AGI.
with all the fear of AGI incoming, i think about the old genie tale. moral of the story: it doesnt matter if you have a superpower, the most important thing is how you use it. knowing the right way is not trivial.
man i feel so in tune with claude. it one-shots some pretty complicated things. it extrapolates things i didnt mention, like its reading my mind. what a time to be alive.
SmallThinker 3B seems pretty good. Small thinking model similar to o1. FP16 is only 6.8GB! If you think about where open models were early last year vs. where we are today, it's pretty incredible and I am grateful to all the awesome people who release this stuff for free. OSS will win.
@kitanivan@yacineMTB I started out by having him ask a question, i'd type it out, then would read the response back to him. Graduated to teaching him to type his own questions, reading the responses himself.