Oh, I have EARNED my right to knock it and berate it for being the absolute trash product that it is … this is more directed at Anthropic and open AI rather than LLMs in general models are fine is the assholes behind these companies Altaman in particular complete opportunist sociopath
Ok kids the final clean up is running for GCclean !!!
ECT doin his thing (read the details):
It will be like say anything
Is this even legal ?
time to calibrate the ride cropping algorithm,
Yes it’s going to be the same jankiness we saw when 29” mtb first came out and everyone tried to cram the bigger wheels into the 26” form factor. After a few years they’ll stop trying to make 32” inch that rides like 29” and make 32” that rides like 32”
I feel like gravel has been reactionary for awhile sort of trying to thread the needle between road and mtb, 32” is a change than looks set to be driven by gravel (what would the old monster cross originators say?)
You know in 3 body problem when one of the suns get too close and they all have to scramble and dehydrate to try to survive ? yeah that’s compaction
I am so close to finishing the filtering toe GCclean, SO MANY rabbit holes to dive down.
ha! did they produce anything interesting from it?
… so it’s really your brain doing the heavy lifting in making the LLM output seem impressive. that’s why there’s such an incredible divergence between hype (your brain accepting the vibe) versus hater when the illusion falls apart on inspection and real use.
I think what they meant by vibe coding is how LLM output creates the sensation of a thing without it actually being the thing … it’s basically playing on phenomena that the human brain fills in gaps to complete the picture …
i was working on reconciling json ride level MMP with "manual" mmp where gaps are zero filled and found disagreement between the GC json per ride MMP and the manual ride MMP off the raw data.
@liversedge.bsky.social is this correct for GC: A1 nails the catastrophic-gap rides (0.1–1% agreement). A2 wins on the micro-dropout rides (GC seems to interpolate 1-s gaps but skip
big ones). A hybrid — interpolate short gaps, skip big ones — would reproduce GC almost exactly.
Look - I’m not beyond vanity projects at this point in my life, I’ll even send @courier429.bsky.social a tshirt
talented inexperienced devs are aware learn etc. token space is more like a black box full of unheardable cats
That’s a fairly trite regurgitation of the talking points. LLM token space navigation is nothing at all like super talented inexperienced devs. Nothing like that at all. It’s about steering vectors, attractors, and paradoxical traps etc
yes silo based architecture is absolutely mandatory
Very Big. Changing all of sports performance time series modeling and analysis in one pipeline of sequential projects
part of the issue is it’s turned into several interrelated nested projects. One is effectively a PhD on token space navigation, two a software sweet to optimize token space navigation, and three the LightBox projects themselves
Full blown 100/month version with Opus. I want to make a really hard fast push to get all the lightbox projects built. I keep thinking I’m about ready to share with the group then 10 more things need to get done
I feel like the world needs a MGTHB store with stuff for purchase
I swear Claude code legitimately seems like none of the devs actually tried to use it for a real task. the compaction issue absolutely destroys contexts and there’s no warning … about the only warning is that you finally get some progress going then gone lobotomized
I wonder what the Venn diagram of people with awful data and those asking LLMs to find deep insights their data
an so negligible to non-negligible. Do you think that it’s real human users that were accessing the GC S3 or other data scraping?
What’s the cost level for S3?
those are rad!
No rush I’m just getting the research pipeline started with building GCclean, that will be the base data for a series of studies that build up the tools for the capstone project which will the encoder and learning. The season level is the moonshot project after that.
