thinking about machine learning models as a form of extremely out-of-order serial publication
"we're going to take your blog post and break it up into byte pairs then republish those byte pairs in algorithmic order, interspersed with and alongside other people's byte pairs. cool with you?"
"no"
"too bad!"
@aparrish stringing together a text from titles of a chain of youtube recommendations
@phooky @aparrish i'm gonna go ahead and try that, always using the first recommendation.
Why Car Companies Still Spend Thousands On Clay Models The massive Fatigue Carousel helps keep roads safe Why are there no bridges in East London? What happened to Old London Bridge? Tower Bridge could have looked very different English counties explained Why does London have 32 boroughs? What's wrong with London's boroughs?
@phooky @aparrish ok one more try, this time picking the 3rd recommendation that isn't the same channel
Coming Home The Unknown Master of Restoration Hands: A Dublin Bookbinder A DIY Synthesizer You Built From A Magazine Are KAMOKOTO knives a SCAM? Day in the Life of a Japanese Natto Maker 屋台ラーメン A Day in the Life of a Ramen Chef Nishiki Market, Kyoto, Japan／錦市場、京都、日本 Cologne (Köln), Germany Walking Tour
@ranjit @aparrish ok, here goes:
Making A Depth Stop For The Straight-Line Engine, Industrial grade CNC with DIY resources -- Impossible?! The rest of the story, what happens to YouTube projects, This Beat Up, Non-running Omega Seamaster Has Big Potential!, 1940s Metal Lathe Restoration, 1920s Paint Sprayer Restoration, Antique Check Perforator Restoration, Antique Beam Drill Restoration [descends into endless loop of Hand Tool Rescue videos]
@aparrish "recommendation algorithm"? you mean my central nervous system?
@aparrish This fits my understanding of formal grammars theory including and especially Backus-Naur Form. We’ve used them for parsers ever since, but a lot of the early work was more trying to reverse engineer the brain’s recommendation algos for language
@aparrish
In essence the way our brains surface relevance and meaning from all sensory input fighting for attention by pattern recognition and deciding what is signal and what is noise?
@aparrish im going to train a neutral net on cracked warez and then force it to watch a thousand hours of proprietary software (fair use) and see what it thinks
now i'm thinking about all language models as essentially a form of recommendation algorithm. "if you liked that sequence of tokens, you might also like: ' the'"