scoff manifesto
@andimgladofit.bsky.social
parody. non actionable. all posts are financial advice
at the same time, different channels will have different overall power spectra (that a full rank representation preserves) and so good latents must be doing some sort of spatial mixing directly, and the diffusion models must untangle that and step *down* in dimensionality while increasing dof
August 27, 2025 at 1:06 AM
at the same time, different channels will have different overall power spectra (that a full rank representation preserves) and so good latents must be doing some sort of spatial mixing directly, and the diffusion models must untangle that and step *down* in dimensionality while increasing dof
because any information noised in the forward process cannot be seen later, these models always encode a hierarchical series of representations. but latent space is much closer to full rank than the target data manifold (a perfect one would be exactly full rank)
August 27, 2025 at 1:02 AM
because any information noised in the forward process cannot be seen later, these models always encode a hierarchical series of representations. but latent space is much closer to full rank than the target data manifold (a perfect one would be exactly full rank)
there is. in the continuous limit the models learn the target score of the conditional distribution. but the forward process is a gaussian perturbation kernel so the step between any two diffusion times is a white noise, so high frequency modes must drop (exponentially) faster
August 27, 2025 at 12:58 AM
there is. in the continuous limit the models learn the target score of the conditional distribution. but the forward process is a gaussian perturbation kernel so the step between any two diffusion times is a white noise, so high frequency modes must drop (exponentially) faster
lmao that these might print
June 5, 2025 at 7:46 PM
lmao that these might print
this is easily one of the top 3 worst trades in nba history. david stern is furiously fighting his way out of hell to stop this
February 2, 2025 at 11:55 AM
this is easily one of the top 3 worst trades in nba history. david stern is furiously fighting his way out of hell to stop this
shoutout to deepseek, showing you can just bolt on CoT with direct rl if your base model is good enough
January 27, 2025 at 3:56 AM
shoutout to deepseek, showing you can just bolt on CoT with direct rl if your base model is good enough
california basically needs to remove its entire regulatory state at this point or the people are going to elect democrat hitler
January 9, 2025 at 6:59 PM
california basically needs to remove its entire regulatory state at this point or the people are going to elect democrat hitler
hard to say with insurance its very regulated and im not an insurance guy. the broad issues are fraud and states making it unprofitable to service. definitely more parametric structures in the policies, but idk if those are even legal to offer to consumers (mb bypasses california's idiot laws?)
January 9, 2025 at 6:46 PM
hard to say with insurance its very regulated and im not an insurance guy. the broad issues are fraud and states making it unprofitable to service. definitely more parametric structures in the policies, but idk if those are even legal to offer to consumers (mb bypasses california's idiot laws?)
there are already hurricane binary options, but there are some otc parametric structures (so like wind pressure, rainfall). i'm sure somebody has something similar for fire, but that is still otc. the big volume rn are temperature based contracts (LNG hedge)
January 9, 2025 at 6:41 PM
there are already hurricane binary options, but there are some otc parametric structures (so like wind pressure, rainfall). i'm sure somebody has something similar for fire, but that is still otc. the big volume rn are temperature based contracts (LNG hedge)
unrelated: you see this paper openreview.net/pdf?id=gojL6...
if this works on fluid dynamics im gonna lose my shit
if this works on fluid dynamics im gonna lose my shit
openreview.net
January 8, 2025 at 7:18 AM
unrelated: you see this paper openreview.net/pdf?id=gojL6...
if this works on fluid dynamics im gonna lose my shit
if this works on fluid dynamics im gonna lose my shit
resolving to reply to more posts with lol in 2025
January 8, 2025 at 6:32 AM
resolving to reply to more posts with lol in 2025
this person is going to the reeducation camps when i take control
December 31, 2024 at 2:30 AM
this person is going to the reeducation camps when i take control
i thought about doing something like this for fusion operations with triton or MLIR, but i think that's actually just a full phd topic of work because i'd need to develop some sort of proof engine for it
December 29, 2024 at 5:02 AM
i thought about doing something like this for fusion operations with triton or MLIR, but i think that's actually just a full phd topic of work because i'd need to develop some sort of proof engine for it
some lab needs to give me 50000 h200s so i can implement an implicit runge kutta token sampler that costs 3.5 million dollars per inference run and outputs "i don't feel like doing that right now" 50% of the time
December 28, 2024 at 6:01 PM
some lab needs to give me 50000 h200s so i can implement an implicit runge kutta token sampler that costs 3.5 million dollars per inference run and outputs "i don't feel like doing that right now" 50% of the time
if your children don't venerate Urkel thought theyre ngmi
December 28, 2024 at 5:55 PM
if your children don't venerate Urkel thought theyre ngmi
yeah it's basically greenfield and its the sort of problem where throwing money into a furnace gets you better solutions for a while
December 28, 2024 at 5:52 PM
yeah it's basically greenfield and its the sort of problem where throwing money into a furnace gets you better solutions for a while
not going to speak for him, but at least in terms of "make my llm bigger and deeper" it's unlikely that going from 600bn to 6T model size with autoregressive LLMs gets you even a 20% better model
there's a lot of room in inference compute though. imo mindless scaling isn't dead for at least 5 years
there's a lot of room in inference compute though. imo mindless scaling isn't dead for at least 5 years
December 28, 2024 at 9:20 AM
not going to speak for him, but at least in terms of "make my llm bigger and deeper" it's unlikely that going from 600bn to 6T model size with autoregressive LLMs gets you even a 20% better model
there's a lot of room in inference compute though. imo mindless scaling isn't dead for at least 5 years
there's a lot of room in inference compute though. imo mindless scaling isn't dead for at least 5 years
its also a pretty good book! not as a reference manual, but a good introduction
December 27, 2024 at 8:50 PM
its also a pretty good book! not as a reference manual, but a good introduction
i will stop flaming you when you read this book
December 27, 2024 at 8:02 PM
i will stop flaming you when you read this book
genuinely shocked you didnt have it already lmao. my family loves making fun of me because i can pack all my material possessions into 4 boxes and move within 4 hours, and i still had one of their dutch ovens
December 25, 2024 at 9:08 PM
genuinely shocked you didnt have it already lmao. my family loves making fun of me because i can pack all my material possessions into 4 boxes and move within 4 hours, and i still had one of their dutch ovens
o3 arc just being optimal control style value iteration over token trajectories is a really funny way to blow up the agi foom cranks though
December 25, 2024 at 1:02 AM
o3 arc just being optimal control style value iteration over token trajectories is a really funny way to blow up the agi foom cranks though
the only mistake i made by being stochastic control theory brained was not leaning into it 50x harder
December 25, 2024 at 12:58 AM
the only mistake i made by being stochastic control theory brained was not leaning into it 50x harder
i know i have insane bay area brain when im looking at a 30y 800k 7% and thinking 'huh that's reasonable'
December 23, 2024 at 12:06 AM
i know i have insane bay area brain when im looking at a 30y 800k 7% and thinking 'huh that's reasonable'
they cant really even install cuda they just pip install torch, huggingface and pray the installation doesnt detonate
December 22, 2024 at 11:57 PM
they cant really even install cuda they just pip install torch, huggingface and pray the installation doesnt detonate