Huh. Looks like Plato was right. A new paper shows all language models converge on the same "universal geometry" of meaning. Researchers can translate between ANY model's embeddings without seeing the original text. Implications for philosophy and vector databases alike.

LaViDa: A Large Diffusion Language Model for Multimodal Understanding "We introduce LaViDa, a family of VLMs built on DMs. We build LaViDa by equipping DMs with a vision encoder and jointly fine-tune the combined parts for multimodal instruction following. " "LaViDa achieves

🚀 Excited to share the most inspiring work I’ve been part of this year: "Learning to Reason without External Rewards" TL;DR: We show that LLMs can learn complex reasoning without access to ground-truth answers, simply by optimizing their own internal sense of confidence. 1/n

frontier ai today

AshutoshShrivastava x.com/slow_developer…

adding layers till you get new SOTA

> be apple > richest company in the world, every advantage imaginable > go all in on AI, make countless promises > get immediately lapped by everyone > 2 years into the race, nothing to show for it > give up, write a paper about how it's all fake and gay and doesn't matter anyway

AI can't reason and this seminal paper proves it:

41% of YC AI startups are solving tasks workers don't need automated New Stanford study shows workers actually DO want AI, but for repetitive work that frees them up for higher value tasks Startups are chasing full automation where partnership would work better

🤹 New blog post! I write about our recent work on using hierarchical trees to enable sparse attention over irregular data (point clouds, meshes) - Erwin Transformer. blog: maxxxzdn.github.io/blog/erwin/ paper: arxiv.org/abs/2502.17019 Compressed version in the thread below:

This will probs be one of the most influential papers of the year

please don’t hallucinate bro they got my family 😭

Excited to share Energy-Based Transformers (EBTs), which allows you to implement system 2 thinking in any modality! EBTs formulate reasoning as an energy optimization problem, allowing models to internally think without complexities like CoT or multiple recurrent latents.

In-context learning is just gradient descent without explicit training! This paper "Learning without training: The implicit dynamics of in-context learning" shows that ICL can be mathematically interpreted as an implicit low-rank weight update during inference.

Excited to share Flow Matching Policy Gradients: expressive RL policies trained from rewards using flow matching. It’s an easy, drop-in replacement for Gaussian PPO on control tasks.

I'm noticing that due to (I think?) a lot of benchmarkmaxxing on long horizon tasks, LLMs are becoming a little too agentic by default, a little beyond my average use case. For example in coding, the models now tend to reason for a fairly long time, they have an inclination to

𝑳𝑳𝑴𝒔 can really 𝑺𝒆𝒍𝒇-𝑬𝒗𝒐𝒍𝒗𝒆, 𝒘𝒊𝒕𝒉𝒐𝒖𝒕 𝑯𝒖𝒎𝒂𝒏 𝑫𝒂𝒕𝒂! -- One LLM, two roles: Challenger creates tasks, Solver answers them. -- No data, no labels, just a base model that learns and improves itself! We name it 𝑹-𝒛𝒆𝒓𝒐: arxiv.org/abs/2508.05004

Beautiful Paper. An LLM teaches itself from a single topic prompt, no human-written questions, no labels. An LLM plays both teacher and student, creates its own questions, and learns with reinforcement learning. By just splitting into a proposer that writes problems and a

RLVR/RLHF libraries: • verl - ByteDance • TRL - HuggingFace • slime - Zhipu AI • prime-rl - Prime Intellect • ROLL - Alibaba • Nemo-RL - NVIDIA • AReaL - Ant Research • SkyRL - UC Berkeley • open-instruct - Allen AI • torchtune - PyTorch Any I am missing? Which do you

Here's an overview of what the app does: - First search the docs with user query - Evaluate if the retrieved context is relevant using LLM - Only keep the relevant context - Do a web search if needed - Aggregate the context & generate response Now let's jump into code!