Thomas Willberger
@Thomas_ensc
Followers
1K
Following
3K
Media
238
Statuses
2K
AI, Entrepreneurship & Computer graphics. š Karlsruhe University (KIT). Founded Enscape.
Karlsruhe, Germany
Joined May 2016
The left statements are basically the human LLM prompts, engraved in firmware.
Psychologists have posited hundreds of cognitive biases over the years. A fascinating paper argues that they all boil down to one of a handful of fundamental beliefs coupled with confirmation bias. [Link below.]
0
0
0
These renderings have been made with Enscape š„ i can feel the shader code and rendering hacks when looking at these images š Always funny.
BREAKING: @PressSec announces the construction of a new 90,000 sq ft ballroom at the White House ā with construction to begin in September. Visit for more.
0
0
3
Great mathematical analysis of the transformer. Reminds me of the EDM work by Tero Karras, which is āoutdatedā now that everyone uses flow matching? š¤.
Deep learning training is a mathematical dumpster fire. But it turns out that if you *fix* the math, everything kinda just worksā¦fp8 training, hyperparameter transfer, training stability, and more. [1/n]
0
0
3
RT @fintech_shark: This guy predicted a year ago that Trump would win 2024 election and attack Iran, instigated by Netanyahu. Bro teachesā¦.
0
33K
0
Half of me celebrates this tech, the other half grieves about the handcrafted artisanal art and narrative craftsmanship that may become increasingly worthless.
Hunyuan-GameCraft. High-dynamic Interactive Game Video Generation with Hybrid History Condition
0
0
5
The ultimate test if youāre mission driven is whether or not youād grant your success to your worst enemy without envy.
1
0
2
RT @LechMazur: @karpathy This is cool as a demo but a terrible idea for real-world use. Consistency is extremely important in UI. Nobody waā¦.
0
1
0
_Drug dealers: Should NOT use their product. _Chefs: Should use their product. Be a chef startup founder, not a drug dealer.
0
0
4
Why is ComfyUI apparently only used by people with an anime fetish? :D These guys are the real early adopters.
0
0
1
RT @hisspikeness: 1/6 Why does the brain maintain such precise excitatory-inhibitory balance? .Our new preprint explores a provocative ideaā¦.
0
39
0
Still waiting for an AI-generated video that's genuinely interesting, not just a tech demo. Are we entering the era of unlimited creativity or "mass produced garbage"?.
0
0
2
RT @Michael_J_Black: If you're an international PhD student at Harvard studying computer vision and your visa is cancelled, reach out to meā¦.
0
94
0
LLMs go into an endless loop if the task requires "true" working memory.
2
0
1
4/ Probably, this conceptual shortcoming is visible in jokes, but is present in other domains as well. Are jokes that much OOD? Is it because they play with the readers extrapolation of the most likely path and diverge from it? But even then: Shouldn't an LLM learn that from many.
0
0
2