prxtml.bsky.social - Profile | ThreadSky | a Reddit-style client for Bluesky

LLMs That Don't Gaslight You A new language model uses diffusion instead of next-token prediction. That means the text it can back out of a hallucination before it commits. This is a big win for areas like law & contracts, where global consistency is valued timkellogg.me/blog/2025/02...

submitted 9 days ago • 7 comments

"𝐑𝐚𝐝𝐢𝐚𝐧𝐭 𝐅𝐨𝐚𝐦: Real-Time Differentiable Ray Tracing" A mesh-based 3D represention for training radiance fields from collections of images. radfoam.github.io arxiv.org/abs/2502.01157 Project co-lead by my PhD students Shrisudhan Govindarajan and Daniel Rebain, and w/ co-advisor Kwang Moo Yi

submitted 21 days ago • 2 comments

RadFoam source code has arrived! (Apache-v2) github.com/theialab/rad... Belated happy Valentine's day 🥰

submitted 11 days ago • 1 comment

dolphin-r1: a dataset for training R1-style models - 800k total samples dataset similar in composition to the data used to train DeepSeek-R1 Distill models. - 300k from DeepSeek-R1 - 300k from Gemini 2.0 flash thinking - 200k from Dolphin chat huggingface.co/datasets/cog...

submitted 28 days ago • 0 comments

Why choose between strong #LLM reasoning and efficient models? Use DeepSeek to generate high-quality training data, then distil that knowledge into ModernBERT for fast, efficient classification. New blog post: danielvanstrien.xyz/posts/2025/d...

submitted 29 days ago • 2 comments

This is one of the first few posts I've seen that uses Deepseek model to generate high quality datasets, which then can be used to train the ModernBERT models. Really neat stuff! Once can easily replace the slower, expensive 3rd party LLM router with a fast, cheap & local model.

submitted 28 days ago • 0 comments

some little bluesky tips 🦋 your blocks, likes, lists, and just about everything except chats are PUBLIC you can pin custom feeds; i like quiet posters, best of follows, mutuals, mentions if your chronological feed is overwhelming, you can make and pin make a personal list of "unmissable" people

submitted 99 days ago • 13 comments

A technical paper for DeepSeek Janus-Pro: Unified Multimodal Understanding and Generation with Data and Model Scaling Janus-Pro incorporates (1) an optimized training strategy, (2) expanded training data, and (3) scaling to larger model size. arxiv.org/abs/2501.17811

submitted 28 days ago • 2 comments

Deepseek-R1 thread to gather thoughts and reactions Nice to see the technical details and MIT license for something that looks at o1 level 🥳

submitted 37 days ago • 2 comments

Deepseek R1 이 오픈 ㄷㄷㄷ huggingface.co/deepseek-ai/...

submitted 37 days ago • 1 comment

huggingface.co/papers/2501.... A recent paper I liked. I have a soft-spot for GANs and (relatively) simple architectures.

submitted 41 days ago • 0 comments

Collaboration Characters (KR/JP in comments) Following the new characters, characters collaborated with UTAU producers will join soon! Who do you think they could be? Share your thoughts in the comments ;) #voxfactory #update #characters #synthesizer #musicproduction #follow

submitted 41 days ago • 4 comments

📣Join @alexhanna.bsky.social & @emilymbender.bsky.social for the 1st Mystery AI Hype Theater 3k livestream this year. ⏰12pm pacific. ➡️https://twitch.tv/dair_institute We'll start 2025 by ridiculing ARC, o3 & all things OpenAI. Until then, catch up on prior episodes @ dair-institute.org/maiht3k/.

submitted 48 days ago • 1 comment

Really happy to share our preprint now on @elife.bsky.social on quantifying patterns in a large behavioral dataset from >100 socially-housed marmosets elifesciences.org/reviewed-pre.... We hope this data be useful for identifying atypical patterns of behavior in disease models in our lab and others.

submitted 69 days ago • 3 comments

This simple pytorch trick will cut in half your GPU memory use / double your batch size (for real). Instead of adding losses and then computing backward, it's better to compute the backward on each loss (which frees the computational graph). Results will be exactly identical

submitted 70 days ago • 4 comments

The glass amphipod Cystisoma is transparent because in the open ocean there is nowhere to hide, and its round head is completely covered in giant eyes. So when you're swimming far from shore just remember: you can't see them, but they can see you. 📽️ Dr. Alejandro Damian-Serrano 🧪🌊🌿🦑🐙

submitted 71 days ago • 39 comments

Fantastic #neurips keynote by Arnaud Doucet! Really like this slide tracing back many of the modern flow-matching / stochastic interpolants ideas to a 1986 result by probabilist Istvan Gyongy describing how to "Markovianize" a diffusion process (eg. having coefficients depending on all the past)

submitted 75 days ago • 1 comment

Learning style from a single image is difficult, but what if you had access to an image pair instead? I’m excited to share our #SIGGRAPHASIA2024 work PairCustomization, on customizing text-to-image models with a single image pair!! project page: paircustomization.github.io 1/3

submitted 84 days ago • 1 comment

New paper shows AI art models don't need art training data to make recognizable artistic work. Train on regular photos, let an artist add 10-15 examples their own art (or some other artistic inspiration), and get results similar to models trained on millions of people’s artworks

submitted 85 days ago • 6 comments

SuperGaussians: Enhancing Gaussian Splatting Using Primitives with Spatially Varying Colors Rui Xu, Wenyue Chen, Jiepeng Wang, Yuan Liu, Peng Wang, Lin Gao, Shiqing Xin, Taku Komura, Xin Li, Wenping Wang arxiv.org/abs/2411.18966

submitted 87 days ago • 1 comment

My mailbox is literally flooded with discussions of whether “Deep learning is hitting a wall”, almost none of which cite my much-ridiculed-at-the-time 2022 paper called “Deep learning is hitting a wall”. nautil.us/deep-learnin...

submitted 89 days ago • 11 comments

Graphics has long studied how to make: (1) realistic images 📷 (2) non-photorealistic images ✍️ (3) realistic sounds 🎤 What about (4) "non-phono-realistic" sounds? What could that even mean? Next week at SIGGRAPH Asia, MIT undergrad Matt Caren will present our proposal… 🧵 arxiv.org/abs/2409.13507

submitted 88 days ago • 2 comments

⚠️경고! 유언비어 주의!⚠️ "여성을 위한 크리스마스 선물로 화장품이나 악세서리를 추천" 한다는 유언비어가 돌고 있습니다만, 절대로 믿지 마세요! 여성이 정말 가지고 싶은 것은 Geforce RTX 4090

submitted 89 days ago • 1 comment

How does the script work? - It connects to BluesSky API with your username/pwd - You can pass a handle and it will retrieve the las 72hours replies - It will iterate the replies and run OpenAI moderation APi on each one (you can replace OpenAI by the moderation filter of your liking)

submitted 89 days ago • 1 comment

Yeah, many people miss this: Gumbel softmax is not applicable when you want to branch! You can hack your way around it, but bias compounds very rapidly

submitted 90 days ago • 5 comments

just because everything is public doesn’t mean u have to be a piece of shit about it

submitted 91 days ago • 22 comments

A lot of RAG content online is either (1) too academic and prescriptive or (2) is a basic toy example, so reading something like this is a nice reminder that "the best way to do something is the one that actually works", as obvious as it sounds.

submitted 89 days ago • 1 comment

[SATURDAY THREAD] ☕️ 🧑‍🎓 In case you spent the week reading GDPR legislation and missed everything. It’s all about vision language models and image preference datasets. >> 🧵 Here are the models and datasets you can use in your projects.

submitted 89 days ago • 5 comments

Some recent discussions made me write up a short read on how I think about doing computer vision research when there's clear potential for abuse. Alternative title: why I decided to stop working on tracking. Curious about other's thoughts on this. lb.eyer.be/s/cv-ethics....

submitted 89 days ago • 19 comments

Okay, last week was reminding everyone who thinks all lolicon art is illegal that they're wrong; this week it's apparently time to remind everyone who thinks all lolicon art is legal that they're wrong, too. Content note: discusses depictions of minors engaged in sexual activity.

submitted 93 days ago • 41 comments

🚨 [AI BOOK CLUB] "AI Snake Oil: What AI Can Do, What It Can’t, and How to Tell the Difference" by @randomwalker.bsky.social & @sayash.bsky.social is a MUST-READ for everyone interested in AI, and it's our 🎉 15th selected book: 📖 About the book:

submitted 90 days ago • 2 comments

🤔 Can you turn your vision-language model from a great zero-shot model into a great-at-any-shot generalist? Turns out you can, and here is how: arxiv.org/abs/2411.15099 Really excited to this work on multimodal pretraining for my first bluesky entry! 🧵 A short and hopefully informative thread:

submitted 90 days ago • 2 comments

Textured Gaussians for Enhanced 3D Scene Appearance Modeling Brian Chao, Hung-Yu Tseng, Lorenzo Porzi, Chen Gao, Tuotuo Li, Qinbo Li, Ayush Saraf, @jbhuang0604.bsky.social, Johannes Kopf, Gordon Wetzstein, Changil Kim tl;dr: augment Gaussians with RGBA texture maps arxiv.org/abs/2411.18625

submitted 91 days ago • 0 comments

It's pretty sad to see the negative sentiment towards Hugging Face on this platform due to a dataset put by one of the employees. I want to write a small piece. 🧵 Hugging Face empowers everyone to use AI to create value and is against monopolization of AI it's a hosting platform above all.

submitted 91 days ago • 29 comments

Fuck it! Structured Generation w/ SmolLM2 running in browser & WebGPU 🔥 Powered by MLC Web-LLM & XGrammar ⚡ Define a JSON schema, Input free text, get structured data right in your browser - profit!!

submitted 90 days ago • 4 comments

Just FYI because it seems relevant and I've seen it misstated a few times, LAION retrained their dataset and provided a diff to migrate over to the filtered Re-LAION-5B dataset HuggingFace, like most platforms that handle user content, do checks for CSAM too. laion.ai/blog/relaion...

submitted 91 days ago • 1 comment

I really fucking wish the AI people who don't understand consent, privacy, risk mitigation, or compassionately answering user concerns about any of the former would stop talking to the public because it's already goddamn hard enough to get people to understand not all machine learning is this shit

submitted 91 days ago • 27 comments

my take on the dataset drama is that i looked at the scraping script. it's 123 lines of very simple code, bc atproto is inherently distributed/open—bsky can't prevent scraping the 🤗 guy is a faceposter so can be bullied into taking the dataset down, but i bet an anon will post a replica any day now

submitted 92 days ago • 9 comments

Check out CAT4D: our new paper that turns (text, sparse images, videos) => (dynamic 3D scenes)! I can't get over how cool the interactive demo is. Try it out for yourself on the project page: cat-4d.github.io

submitted 91 days ago • 1 comment

Asking the following earnestly: what is the strongest case for GANs standing the "test of time"? Are they important 10 years later in modern ML research? How have they influenced the way we think about generative models today?

submitted 91 days ago • 19 comments

NeurIPS Test of Time Awards: Generative Adversarial Nets Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, Yoshua Bengio Sequence to Sequence Learning with Neural Networks Ilya Sutskever, Oriol Vinyals, Quoc V. Le

submitted 91 days ago • 6 comments

Test of Time Paper Awards are out! 2014 was a wonderful year with lots of amazing papers. That's why, we decided to highlight two papers: GANs (@ian-goodfellow.bsky.social et al.) and Seq2Seq (Sutskever et al.). Both papers will be presented in person 😍 Link: blog.neurips.cc/2024/11/27/a...

submitted 91 days ago • 1 comment