iiinova — where your digital self keeps evolving

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 04/09/2025

Has anyone successfully done RL post-training of GPT-oss with meaningful performance gains? What libraries even support it? I guess technically TRL/axolotl, maybe Unsloth... but there are no good examples of doing it...

291

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 01/04/2025

I have EXCITING news: I've started a company! Introducing Sophont We’re building open multimodal foundation models for the future of healthcare. We need a DeepSeek for medical AI, and @sophontai.bsky.social will be that company! Check out our website & blog post for more info (link below)

1302

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 17/03/2025

NEW BLOG POST: LLMs in medicine: evaluations, advances, and the future www.tanishq.ai/blog/posts/l... A short blog post discussing how LLMs are evaluated for medical capabilities and what's the future for LLMs in medicine (spoiler: it's reasoning!)

1212

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 19/02/2025

I restarted my blog a few weeks ago. The 1st post was: Debunking DeepSeek Delusions I discussed 5 main myths that I saw spreading online back during the DeepSeek hype. It may be a little less relevant now, but hopefully still interesting to folks. Check it out → www.tanishq.ai/blog/posts/d...

4211

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 20/01/2025

Okay so this is so far the most important paper in AI of the year

2231

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 11/01/2025

Anthropic, please add a higher tier plan for unlimited messages 😭🙏

4160

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 10/01/2025

Decentralized Diffusion Models UC Berkeley and Luma AI introduce Decentralized Diffusion Models, a way to train diffusion models on decentralized compute with no communication between nodes. abs: arxiv.org/abs/2501.05450 project page: decentralizeddiffusion.github.io

0202

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 10/01/2025

The GAN is dead; long live the GAN! A Modern Baseline GAN This is a very interesting paper, exploring making GANs simpler and more performant. abs: arxiv.org/abs/2501.05441 code: github.com/brownvc/R3GAN

0132

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 10/01/2025

Happy birthday to my incredible and awesome Mamma! 🥳🎉🎂 To many more years of health and happiness. Tiara (my sister) and I love you very much ❤️❤️❤️

190

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 28/12/2024

Happy 19th birthday to my amazing sister Tiara Abraham! 🥳🎉 🎂 Proud of you graduating with your Master's degree at 18 and starting your doctorate in music degree this past year! Excited to see what this final teen year holds for you!

0150

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 10/12/2024

Inventors of flow matching have released a comprehensive guide going over the math & code of flow matching! Also covers variants like non-Euclidean & discrete flow matching. A PyTorch library is also released with this guide! This looks like a very good read! 🔥 arxiv: arxiv.org/abs/2412.06264

110927

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 10/12/2024

Normalizing Flows are Capable Generative Models Apple introduces TarFlow, a new Transformer-based variant of Masked Autoregressive Flows. SOTA on likelihood estimation for images, quality and diversity comparable to diffusion models. arxiv.org/abs/2412.06329

arxiv.org

Normalizing Flows are Capable Generative Models

Normalizing Flows (NFs) are likelihood-based models for continuous inputs. They have demonstrated promising results on both density estimation and generative modeling tasks, but have received relati...

1549

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 10/12/2024

Refusal Tokens: A Simple Way to Calibrate Refusals in Large Language Models "We introduce a simple strategy that makes refusal behavior controllable at test-time without retraining: the refusal token." arxiv.org/abs/2412.06748

061

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 10/12/2024

Can foundation models actively gather information in interactive environments to test hypotheses? "Our experiments with Gemini 1.5 reveal significant exploratory capabilities" arxiv.org/abs/2412.06438

0101

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 10/12/2024

Training Large Language Models to Reason in a Continuous Latent Space Introduces a new paradigm for LLM reasoning called Chain of Continuous Thought (COCONUT) Directly feed the last hidden state (a continuous thought) as the input embedding for the next token. arxiv.org/abs/2412.06769

2528

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 10/12/2024

[MASK] is All You Need New paper from CompVis group, introduces a new method called Discrete Interpolants that builds on top of discrete flow matching. Achieves SOTA performance on MS-COCO, competitive results on ImageNet 256. arxiv.org/abs/2412.06787

1326

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 09/12/2024

A new tutorial on RL by Kevin Patrick Murphy, a Research Scientist at Google DeepMind who also wrote several comprehensive, well-regarded textbooks on ML/DL. This ought to be a good read 👀 arxiv.org/abs/2412.05265

1385

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 09/12/2024

Birth and Death of a Rose abs: arxiv.org/abs/2412.05278 Generating temporal object intrinsics - temporally evolving sequences of object geometry, reflectance, and texture, such as blooming of a rose - from pre-trained 2D foundation models.

0130

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 09/12/2024

Frontier Models are Capable of In-context Scheming abs: arxiv.org/abs/2412.04984 "Our results show that o1, Claude 3.5 Sonnet, Claude 3 Opus, Gemini 1.5 Pro, and Llama 3.1 405B all demonstrate in-context scheming capabilities"

0111

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 09/12/2024

BigDocs: An Open and Permissively-Licensed Dataset for Training Multimodal Models on Document and Code Tasks abs: arxiv.org/abs/2412.04626 project page: bigdocs.github.io BigDocs-7.5M is a high-quality, open-access dataset comprising 7.5 million multimodal documents across 30 tasks.

0121

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 09/12/2024

Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling abs: arxiv.org/abs/2412.05271 model: huggingface.co/OpenGVLab/In... Introduces new InternVL-2.5 model, the first open-source MLLMs to surpass 70% on the MMMU benchmark

0101

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 06/12/2024

NVILA: Efficient Frontier Visual Language Models abs: arxiv.org/abs/2412.04468 NVIDIA introduces NVILA, a family of open VLMs designed to optimize both efficiency and accuracy.

0121

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 06/12/2024

Infinity: Scaling Bitwise AutoRegressive Modeling for High-Resolution Image Synthesis abs: arxiv.org/abs/2412.04431 New visual autoregression framework that performs bitwise token prediction w/ an infinite-vocabulary tokenizer & classifier, a new record for autoregressive text-to-image models.

082

Reposted by Tanishq Mathew Abraham

Nick Stracke @rmsnorm.bsky.social · 04/12/2024

🤔 Why do we extract diffusion features from noisy images? Isn’t that destroying information? Yes, it is - but we found a way to do better. 🚀 Here’s how we unlock better features, no noise, no hassle. 📝 Project Page: compvis.github.io/cleandift 💻 Code: github.com/CompVis/clea... 🧵👇

24210

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 04/12/2024

Leading computer vision researchers Lucas Beyer (@giffmana.ai), Alexander Kolesnikov (@kolesnikov.ch), Xiaohua Zhai have left Google DeepMind to join OpenAI! They were behind recent SOTA vision approaches and open-source models like ViT, SigLIP, PaliGemma

1190

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 02/12/2024

the restrictions on post and video length is gonna make it harder to paper-post here ngl

3101

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 02/12/2024

Reverse Thinking Makes LLMs Stronger Reasoners abs: arxiv.org/abs/2411.19865 Train an LLM to be able to generate forward reasoning from question, backward question, and backward reaoning from backward question Shows an average 13.53% improvement over the student model’s zero-shot performance

05410

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 02/12/2024

GaussianSpeech: Audio-Driven Gaussian Avatars abs: arxiv.org/abs/2411.18675 project page: shivangi-aneja.github.io/projects/gau...

080

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 02/12/2024

some people managed to find some AoC-solving code from qianxyz in a github repo that has now been deleted seems like an automated pipeline using gpt-4o-mini with a pretty basic prompt

1121

Reposted by Tanishq Mathew Abraham

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 01/12/2024

how does someone solve Advent of Code problem in 9 seconds??!!

4121

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 01/12/2024

how does someone solve Advent of Code problem in 9 seconds??!!

4121

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 28/11/2024

At #XPANSE in Abu Dhabi last week: - Met @anilseth.bsky.social backstage between our talks, discussed studying the nature of consciousness w/ neuroimaging. Appreciated him gifting me a signed copy of his book! - Met @seanmcarroll.bsky.social who gave a great talk about entropy vs. complexity

1181

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 27/11/2024

Many SOTA image generation models use an adversarial loss (VAE for latent diffusion for example), which counts I would say...

170

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 25/11/2024

My Bluesky follower count (1.6k followers) has now surpassed my Threads follower count (1.1k). I still see a few AI folks on Threads but it seems so much more dead compared to BlueSky.

4351

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 25/11/2024

Every time conference reviews and rebuttals come in we hear complaints about how bad the process is. Which ML conference has the best review process and what's stopping other conferences from improving their processes?

1100

Reposted by Tanishq Mathew Abraham

Yoshitomo Matsubara @yoshitomo-matsubara.net · 21/11/2024

Here is a list of ML OSS & Open Source / Science enthusiasts I found on Bluesky 🦋 go.bsky.app/8MFcfXd Let me know if you find such people here! I'm still new here and probably the list misses many must-add people, so let's built it together💪

4011249

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 21/11/2024

Why does he follow me if no one wants me there? 🤔

2100

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 19/11/2024

What should I post here? Anything different from Twitter?

4140

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 18/11/2024

Excited to talk at XPANSE WORLD in Abu Dhabi, looking forward to meeting other esteemed speakers!

020

Reposted by Tanishq Mathew Abraham

Gabriele Corso @gcorso.bsky.social · 17/11/2024

Thrilled to announce Boltz-1, the first open-source and commercially available model to achieve AlphaFold3-level accuracy on biomolecular structure prediction! An exciting collaboration with Jeremy, Saro, and an amazing team at MIT and Genesis Therapeutics. A thread!

18610204

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 30/06/2023

A reminder: if you want it done right do it yourself

000

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 07/06/2023

Posting on LinkedIn or Instagram can be extremely infuriating and I am grateful that Twitter managed to get it correct and luckily remains fairly functional over this past year. Other social media platforms would do well to imitate some of Twitter's awesome posting features.

000

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 05/06/2023

Twitter and YouTube heavily pushing Across the SpiderVerse spoilers 😩 Wasn't even this bad when NWH was released...

000

Tanishq Mathew Abraham @iscienceluvr.bsky.social · 01/06/2023

I'm really excited to share MedARC's first paper since our public launch 🥳 🧠👁️ MindEye! Our state-of-the-art fMRI-to-image approach that retrieves and reconstructs images from brain activity! Project page: medarc-ai.github.io/mindeye arXiv: arxiv.org/abs/2305.18274

040