Thanks @modal for the compute credits, we'll do good with them
English
Unsuspicious Industries
27 posts
Unsuspicious Industries
@unsuspiciousind
Non-profit formally verified open research hub
Katılım Mart 2024
43 Takip Edilen16 Takipçiler
Unsuspicious Industries retweetledi
Finally releasing an alpha version of the Engine powering the Proposition 7 constrained generation project !
The currently system supports Simply Typed functionals, context-mutating imperative features, a bit of subtyping and soon refinement types
STLC tested against Coq impl
English
Slowly our website is starting to look more and more like LessWrong and there is nothing we can do about it
English
this feeling when you read a cool paper but it's obviously written in google docs
English
@__tinygrad__ Do you think measuring the "quantity of information" necessary for a task (through Komolgorov complexity approximation maybe) could be used to set a smart LOC threshold ?
English
Software that looks like tinygrad is the future, even more so with agentic coding getting decent.
It wasn't written like this in the past because it needed to be factorized so companies could increase headcount so managers could feel important.
dogecahedron@dogecahedron
wow @VitalikButerin proposed to follow the lead of @__tinygrad__ and establish goal of max lines of code for ethereum spec
English
Unsuspicious Industries retweetledi
amazing. here did not yet spend anything from the money we didn’t raise 1.5 years ago.
Amjad Masad@amasad
Have *not* yet spent a single dollar from the money we raised 2.5 years ago!
English
Unsuspicious Industries retweetledi
USI never sleeps: new website design and updates on Proposition 7 !
English
At Unsuspicious, we encourage such transparent practices. Research for humanity first !
Z.ai@Zai_org
Since GLM-4.1V, we've addressed many community-reported issues. GLM-4.5V resolves common problems like repetitive thinking and incorrect output formatting. However, some limitations remain: - Frontend code reproduction: May output raw HTML without markdown wrapping or have character escaping issues causing rendering errors. We provide a patch to fix most cases: github.com/zai-org/GLM-V/… - Text Q&A: Room for improvement as this release prioritized multimodal scenarios. - Complex prompts: May cause overthinking or content repetition. - Answer restatement: Occasionally repeats answers at the end. We welcome feedback in our GitHub Issues section and will address problems promptly.
English
Unsuspicious Industries retweetledi
Unsuspicious Industries retweetledi
USI is recruiting a Formal Systems & Grammars engineer for type-constrained ML - Join us to build Witt's dream machine
English
@Dorialexander @__tinygrad__ is really good, it can reach very high speed, usually much higher than anything based on AMD's official drivers. It can be a bit of a pain to setup though
English
Any recommendation to run LLM inference on AMD? Really struggled with vllm last year and since I see more alternatives wondering if there is anything else.
English
Certains aiment écrire le symbole ⇒ à toutes les lignes d’un raisonnement : c’est mal. Déjà, c’est laid. Et puis, il a une signification précise, qui n’est pas celle qu’on veut parfois lui prêter : il ne signifie pas « donc ». Si on pense «donc», pourquoi ne pas écrire «donc» ?
Français
economics are nothing more than simplified physics applied to humans instead of atoms and we at USI will write the first thermodynamics of the masses
English
Every engineer at USI has to play the daily #Leetle
⚠️⚠️⚠️⚠️⚠️⚠️
⚠️⚠️🟩🟩🟩⚠️
🟩🟩🟩🟩🟩🟩
English