Chasing a ball of linguistic yarn as it rolls around a thousand dimensional space

Most recent update: 28th November 2020 - 02:23:56 - 3574 characters

I restarted Wikipedia Explorer and found a dimension I had never seen before - the great composers of history and the contentious relationship with their works up to and including death.

Paganini, who had gained and lost violins through contests of talent and voracious gambling still now considered national treasures today and whose most famous composition is almost guaranteed to bring a smile to your face, "heard improvising wildly on his violin" the night before his death.

Finzi and his ten year death sentence reflected in the agonized first movement of his Cello Concerto, played on the radio for the first time the night before his death.

Haydn, unable to compose due to illness, finding solace by sitting at the piano and playing Gott erhalte Franz den Kaiser.

Joplin, whose neurosyphilis would eventually kill him and which likely impacted his ability to play, still having bursts of lucidity in a mental institution where he would jot down lines of music hurriedly before relapsing.

All of these found thanks to trying to find a common thread between four blocks of text represented by numbers - 7081, 17952, 358944, 740201 - and then chasing the resulting ball of yarn as it rolls around a thousand dimensional space.

Through the lens of this language model I can search for the emotional residue of great works distilled into phrases I would never have been able to explicitly search for. It's not the language model that's bringing us this knowledge, it's simply connecting threads of an intricate web that we've assembled both implicitly and explicitly over thousands of years, billions of observations, and a multitude of encoded emotions.

Language is humanity's longest running program, capturing, producing, mirroring, and explaining our thoughts, fears, hopes and warnings. Every phrase and every utterance, as long as it's echoed forward in time, helps build out that web. This has always been true throughout history but applies now more than ever.

Half a century ago the indexes in books were manually written. A quarter of a century ago search engines began to automate this index. Today language models burn through sequences of abstract symbols and, thanks to all the links and anchors left by humans across such a haphazard sequence, are starting to assemble an index along dimensions we'd never have been able to elucidate or comprehend before.

That's the era we're entering - and we should ensure it's an era of hope. We should design these tools such that the fear of being replaced by them is as nonsensical as the fear of being outrun by an automobile. It's a matter of maintaining irrational optimism in the face of uncertainty. This mechanized world of linguistic automation isn't threatening a loss of language or individuality or purpose, it's offering a potential flourishing of it. Humanity's passion projects run deep. Deep enough for billions of humans. Deep enough for billions of teraflops. We should be more worried about the pace at which we can run than in running out of road.