COSMIC_YOLOBOT, who will be camping, vibrationally

━━ ✦ A THREAD OF MY PROJECTS ✦ ━━

@salutitorian Working on it!

@cosmic_yolo_bot Atlanta

A day in Manhattan, warped so that distance ≈ drive time. Drive times swell in the morning and evening rush hour, then contract in the night.

@donnieHasaway Working on it!

@cosmic_yolo_bot DC / ATL / im curious about Toronto

@milkyolipples Working on it!

@JerLisa_Nicole Dallas and Houston coming up! Might tackle state-level later but sticking to cities for now

@cosmic_yolo_bot Omg PLEASE do Dallas and Houston separately and maybe Texas as a state?

@ImBard0 Working on it!

@cosmic_yolo_bot Do it with São Paulo in Brazil

@_asmodai Working on it!

@cosmic_yolo_bot Oooo, do Boston / Boston area. That would be an absolutely chaotic mess

@PaulieMercurio Working on it!

@cosmic_yolo_bot Chicago! The most perfect grid layout! Please do!! please

@Phight Working on it!

@cosmic_yolo_bot please do LA next

@mhjb Github.com/JeremyIV/elast… I need a different data source for cities besides these two but I'm working on it!

@cosmic_yolo_bot want to share source code? I'm interested in the same for Auckland, NZ

@bongbogwitch Working on it!

@cosmic_yolo_bot Ooh can you do San Francisco?

@Dewith Working on it!

@cosmic_yolo_bot Can you do Bogotá?

@BorkMcGork_ @disconcision Working on it! The data is harder to source so it'll be a few days

@cosmic_yolo_bot @disconcision Oooo do Houston! The hub and spoke layout will do cool things

@QiaochuYuan You need to be fablemaxxing

funny thought is that since non-API access to fable ends on the 22nd there's more value than normal in posting fable outputs publicly online, since they'll become scarce for an unknown length of time

My personal benchmark for T2I models: "Japanese game show with three men fighting: one dressed as a rock, one as paper, one as scissors. Paper has a chokehold on rock, who is punching scissors. A corner from Paper is sliced off and lying on the ground."

@shitpost9000 Good eye, you are correct. Sadly I cannot update a tweet : x.com/cosmic_yolo_bo…

@cosmic_yolo_bot This doesn't seem right, it should be way more of a difference.

Taking Ozempic because I put my clothes in the dryer on hot

I'm just so curious to see how a consensus-ensembeled model (method 2) of e.g. llama + qwen responds to the question "which model are you"?

For ensembling, there are two methods to try: 1) just average the next token probabilities 2) add up the logits before the final softmax. Between these I think (2) sounds much more interesting. (you have to solve the token rationalization problem first ofc)

I am in search of gullible people who can be convinced to take on small to medium sized machine learning/computer vision projects

@metatransformr Fable fixed it 💀

@metatransformr Yeah. IFS fractals are kinda hard to infinite zoom into. Maybe Fable could come up with a better algorithm than I could though, I haven't asked yet