Image created with Flux Pro Ultra. Image prompt: A Minecraft screenshot displaying an advanced redstone city with automated systems, computers made from blocks, and technological contraptions, with “TECH” written in pixelated Minecraft font across the top
“@AgentOpsAI 16/ Accelerate Super Long-Context LLM Inference Hosted by: Boyuan Huang, Product Director of Big Data and AI platform, Alibaba Cloud Intelligence Group As the context length of LLM serving continues to increase, the inference time escalates dramatically. To strike this issue,” / X https://x.com/AtomSilverman/status/1907898671754326348
“Let’s take AI predictions from blog posts, podcasts and tweets and move them to betting markets, our state of the art in truth. My struggle has been coming up with good, concrete, resolvable predicates. Ideally, predicates related to industry metrics and macroeconomics. Eg” / X https://x.com/karpathy/status/1908109168952676855
“In this post I dive into MLIR, a vibrant and thriving compiler technology that is part of the LLVM Foundation. I explore its origin story, impact and also why there is so much confusion around MLIR as a compiler-y thing and MLIR as an AI thing. 👇” / X https://x.com/clattner_llvm/status/1910151124407222534
“If you use AI models a lot it is not hard to tell which are optimized for the benchmarks and which are actually big advances.” / X https://x.com/emollick/status/1909078752614658114
“⚖️🧱 Custom OpenEvals output schemas 🧱⚖️ By popular demand, you can now customize output schemas for OpenEvals LLM-as-judge evaluators! While the defaults cover many cases, this gives you total flexibility over model responses. Available in Python and JS – docs in thread 👇 https://x.com/Hacubu/status/1909636114278965468
“Nice tool for accessing llms.txt! :D” / X https://x.com/jeremyphoward/status/1909383500131950673
“PyTorch just released an awesome tool to visualize matrices and what’s happening inside them. Matrix multiplications (matmuls) are the building blocks of today’s models. It can even run in browser. https://x.com/LiorOnAI/status/1908233269998403980
“Introducing the Evals API. You can now programmatically define tests, automate evaluation runs, and quickly iterate on prompts. Evals are still available in the dashboard—and now through the API, so you can integrate them anywhere in your workflow. https://x.com/OpenAIDevs/status/1909721613853139353
“AI has a jargon problem https://x.com/rasbt/status/1910153499716759954
The 2025 AI Index Report | Stanford HAI https://hai.stanford.edu/ai-index/2025-ai-index-report
“Memory is the next scaling laws paradigm shift” / X https://x.com/EdwardSun0909/status/1910384097786290497
bespokelabs/Bespoke-Stratos-17k · Datasets at Hugging Face https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k
World-first: Quantum computer tunes billion-parameter AI model https://interestingengineering.com/culture/china-ai-model-tuned-quantum-computer?group=test_a
“We’ve added an estimated ~10 million more full-text papers to Elicit over the last month! This makes Elicit Reports more comprehensive/rigorous, since Reports will extract more info per paper. https://x.com/elicitorg/status/1908157705912775093
[2501.18795] Rope to Nope and Back Again: A New Hybrid Attention Strategy https://arxiv.org/abs/2501.18795
[2504.00660] Learning to Normalize on the SPD Manifold under Bures-Wasserstein Geometry https://arxiv.org/abs/2504.00660
“Github 👨🔧: Synthetic data curation for post-training and structured data extraction → Python library simplifying synthetic data generation and curation pipelines for LLM post-training and structured data extraction. → Interactive viewer helps monitor data generation in https://x.com/rohanpaul_ai/status/1909076354009546960
“@rubyrangerr I think you are misunderstanding what this tech demo actually is, but I will engage with what I think your gripe is — AI tooling trivializing the skillsets of programmers, artists, and designers. My first games involved hand assembling machine code and turning graph paper” / X https://x.com/ID_AA_Carmack/status/1909311174845329874
Scene-Centric Unsupervised Panoptic Segmentation https://visinf.github.io/cups/
[2504.01382v1] An Illusion of Progress? Assessing the Current State of Web Agents https://arxiv.org/abs/2504.01382v1
OmniCaptioner https://alpha-innovator.github.io/OmniCaptioner-project-page/
[2504.07095] Neural Motion Simulator: Pushing the Limit of World Models in Reinforcement Learning https://arxiv.org/abs/2504.07095
GitHub – babycommando/neuralgraffiti: Live-bending a foundation model’s output at neural network level. https://github.com/babycommando/neuralgraffiti
[2504.02495] Inference-Time Scaling for Generalist Reward Modeling https://arxiv.org/abs/2504.02495
“Researchers from UC Berkeley and Stanford tapped Test Time Training to generate one-minute videos They added TTT layers to a pre-trained Transformer and fine-tuned it to generate cartoons with strong temporal consistency Here’s a single-shot example: https://x.com/rowancheung/status/1909496813503013066
“A little but important follow-up from a paper that just came out earlier this week: “A Sober Look at Progress in Language Model Reasoning”, https://x.com/rasbt/status/1910707770518560810
FantasyTalking: Realistic Talking Portrait Generation via Coherent Motion Synthesis https://fantasy-amap.github.io/fantasy-talking/
[2504.02324] Dynamic Assortment Selection and Pricing with Censored Preference Feedback https://arxiv.org/abs/2504.02324
[2503.23674] Large Language Models Pass the Turing Test https://arxiv.org/abs/2503.23674
Scientists unveil starfish-inspired wearable tech for heart monitoring // Show Me Mizzou // University of Missouri https://showme.missouri.edu/2025/scientists-unveil-starfish-inspired-wearable-tech-for-heart-monitoring/
OmniSVG: A Unified Scalable Vector Graphics Generation Model https://omnisvg.github.io/
“Key takeaways from NYT article on 1X ⦿ 1X plans to deploy NEO in more than 100 homes this year ⦿ They will build “call centers” for remote operations ⦿ Data collected won’t be used for AI training until 24 hours later, giving customers time to delete unwanted footage https://x.com/TheHumanoidHub/status/1908274713547465001
“Why use mathematics to test AI reasoning? @ElliotGlazer explains how our FrontierMath benchmark challenges AI to perform long-form reasoning and develop a coherent worldview—crucial steps for broader reasoning capabilities and scientific thinking. (00:13) – Why has Epoch AI put https://x.com/EpochAIResearch/status/1908199401773813915
“Universal text embeddings lack negation awareness, often interpreting negated sentences as semantically similar to originals. This paper introduces a data-efficient embedding re-weighting method that enhances negation understanding without modifying the original embedding https://x.com/rohanpaul_ai/status/1909757651572969829
“Why do LLMs attend to the first token? This new paper explains why LLMs obsessively focus attention on the first token — a phenomenon known as an attention sink. Their theory: it’s a useful trick to prevent representational collapse in deep Transformers. • Sinks = over-mixing https://x.com/omarsar0/status/1908187563422261411
[2503.23758] Exact Solution of the Frustrated Potts Model with Next-Nearest-Neighbor Interactions in One Dimension: An AI-Aided Discovery https://arxiv.org/abs/2503.23758
[2504.02398v1] Scaling Analysis of Interleaved Speech-Text Language Models https://arxiv.org/abs/2504.02398v1
Goku https://saiyan-world.github.io/goku/
“Open source model weights here: https://x.com/reach_vb/status/1909706444028670311
SeedLM: Compressing LLM Weights into Seeds of Pseudo-Random Generators – Apple Machine Learning Research https://machinelearning.apple.com/research/seedlm-compressing
“From @SumitGulwani @ruhmamehek (AI @ Microsoft): a new approach to program synthesis recovers from LLM failures not by retrying, but by decomposing. LLMs break down programming-by-example (PBE) tasks into subtasks — a step toward compositional reasoning. https://x.com/ndea/status/1910059834084651025
AI Index 2025: State of AI in 10 Charts | Stanford HAI https://hai.stanford.edu/news/ai-index-2025-state-of-ai-in-10-charts
