Large Language Models Pass the Turing Test https://arxiv.org/pdf/2503.23674
AI-powered therapy shows shocking results in mental health study https://interestingengineering.com/health/groundbreaking-ai-therapy-shows-positive-results
“Dartmouth researchers developed an AI therapy chatbot and found it matches actual mental health professionals! The bot achieved a 51% and 31% reduction in depression and anxiety, taking less time than human therapists Many also formed bonds with it! https://x.com/rowancheung/status/1907304583065456983
Segment Any Motion in Videos https://motion-seg.github.io/
“🚨This week’s top AI/ML research papers: – GPT-4o System Card: Native Image Generation – Anthropic’s On the Biology of a LLM – Gemma 3 Technical Report – Qwen2.5-Omni Technical Report – Reasoning to Learn from Latent Thoughts – Defeating Prompt Injections by Design – Scaling https://x.com/TheAITimeline/status/1906470808563626322
“New Anthropic research: Do reasoning models accurately verbalize their reasoning? Our new paper shows they don’t. This casts doubt on whether monitoring chains-of-thought (CoT) will be enough to reliably catch safety issues. https://x.com/AnthropicAI/status/1907833407649755298
“We found Chains-of-Thought largely aren’t “faithful”: the rate of mentioning the hint (when they used it) was on average 25% for Claude 3.7 Sonnet and 39% for DeepSeek R1. https://x.com/AnthropicAI/status/1907833416373895348
“Introducing TTSTSTT, a breakthrough new AI model architecture. Unlike prior LLM architectures trained entirely on text tokens, Text To Speech To Speech To Text (TTSTSTT) models are trained to perform reasoning entirely within the auditory domain, but also have conversions to text https://x.com/juberti/status/1907180245687669160
“Fine-tuning LLMs for specific tasks often reduces their safety alignment. This paper presents a theoretical framework to understand this safety-capability trade-off in two common safety-aware fine-tuning strategies. 📌 The paper mathematically shows safety degrades less if https://x.com/rohanpaul_ai/status/1906599340694499334
“LVLMs visual understanding behaviors are underexplored. This paper introduces a heatmap visualization method for Large Vision-Language Models to reveal relevant image regions for open-ended question Answering. 📌 Token selection pinpoints image-relevant words in free-form https://x.com/rohanpaul_ai/status/1905829266715213908
“Kevin Frans and colleagues at @UCBerkeley introduced a new way to speed up image generation with diffusion models. Their “shortcut” method trains models to take larger noise-removal steps—the equivalent of multiple smaller ones—without losing output quality. Unlike established https://x.com/DeepLearningAI/status/1906768474816295165
“Creating detailed prompts needed by Text-to-Image models from simple user input is difficult and inefficient. TIPO (Text to Image with Text Presampling for Prompt Optimization) uses a light-weight model to refine simple prompts into detailed ones. Usually, if you just type a https://x.com/rohanpaul_ai/status/1905679791929655415
“UniCombine, a Diffusion Transformer framework for unified multi-conditional image generation. 📌 It uses Conditional Multi-Modal Diffusion Transformer Attention. This allows for unified handling of diverse conditions. 📌 Pre-trained Condition Low-Rank Adaptation modules enable https://x.com/rohanpaul_ai/status/1905797557638365472
[2502.01385v1] Detecting Backdoor Samples in Contrastive Language Image Pretraining https://arxiv.org/abs/2502.01385v1
PaperBench: Evaluating AI’s Ability to Replicate AI Research https://cdn.openai.com/papers/22265bac-3191-44e5-b057-7aaacd8e90cd/paperbench.pdf
Evals – PydanticAI https://ai.pydantic.dev/evals/
“Did we just drop personalized AI evaluation?! This tool auto-generates custom benchmarks on your docs to test which models are the best.” / X https://x.com/fdaudens/status/1907526027435454475
“🚨Multi-Token Attention🚨 📝: https://x.com/jaseweston/status/1907260086017237207
“Here is the new stealth model on my vibe check. It is now the best non-thinking model (at least it has no thinking tokens…). The outputs are super short, it loves Certainly! and listicles. Super interested to see who is behind that one 👀 https://x.com/TheXeophon/status/1907880330985390215
“A Survey of Efficient Reasoning for LLMs This survey focuses on reasoning economy in LLMs, analyzing how to balance deep reasoning performance with computational cost. It reviews inefficiencies, behavioral patterns, and potential solutions at both post-training and inference https://x.com/omarsar0/status/1907072213142151488
Snowplow: Effective Kernel Fuzzing with a Learned White-box Test Mutator https://sishuaigong.github.io/pdf/asplos25-snowplow.pdf
[2503.21088v1] ZJUKLAB at SemEval-2025 Task 4: Unlearning via Model Merging https://arxiv.org/abs/2503.21088v1
GeometryCrafter https://geometrycrafter.github.io/
[2503.21777] Test-Time Visual In-Context Tuning https://arxiv.org/abs/2503.21777
“Code-generating LLs struggle to find the shortest program for a data sequence, which requires advanced reasoning. This paper introduces the KOLMOGOROV-TEST (KT), a benchmark where models compress data by generating the shortest program that produces a given sequence, thus https://x.com/rohanpaul_ai/status/1905780909531607262
“Introducing ArithmeticBench: an extremely challenging benchmark designed to test AIs on the frontiers of arithmetic. We worked with top mathematicians to come up with the biggest numbers ever before revealed on a benchmark. Tasks include numbers exceeding 100 digits in length!🧵 https://x.com/EpochAIResearch/status/1907199415678578804
“This paper proposes that the human brain, unlike a digital computer, cannot be explained by classical computation due to consciousness’s history dependence. 📌 The paper’s core idea is that human conscious experience depends on its entire past history in a way classical digital https://x.com/rohanpaul_ai/status/1905891174558425261
[2504.00460] MetaLoRA: Tensor-Enhanced Adaptive Low-Rank Fine-tuning https://arxiv.org/abs/2504.00460
“🚀Run interactive evals in the LangSmith Playground📊 You can now create datasets inline and add examples to existing datasets without leaving the Playground. This update makes building a dataset to evaluate LLM calls much easier, especially for non-developers. Try it out https://x.com/LangChainAI/status/1907497539492073498
DreamActor-M1: Holistic, Expressive and Robust Human Image Animation
with Hybrid Guidance https://arxiv.org/pdf/2504.01724
“🎉 We are happy to share that a new paper from SkyPilot has been accepted to EuroSys 2025. 🎉 “SkyServe: Serving AI Models across Regions and Clouds with Spot Instances”! 💡 SkyServe intelligently provisions and spreads spot and on-demand instances across regions and clouds, https://x.com/skypilot_org/status/1906685409309974548
“Deep Neural Networks (DNNs) excel in accuracy but often lack robustness, fairness, and other critical qualities. This paper systematically evaluates 326 models across nine quality dimensions beyond accuracy, analyzing training and architecture impacts. It introduces the QUBA https://x.com/rohanpaul_ai/status/1906532903086948702
“Incredible.. This GitHub repo (with 20.8K+ stars ) has a collection of 300+ MCP servers. https://x.com/rohanpaul_ai/status/1906484638937485660
MoCha: Towards Movie-Grade Talking Character Synthesis https://arxiv.org/pdf/2503.23307v1
[2502.03382] High-Fidelity Simultaneous Speech-To-Speech Translation https://arxiv.org/abs/2502.03382
[2501.18100v1] Panacea: Mitigating Harmful Fine-tuning for Large Language Models via Post-fine-tuning Perturbation https://arxiv.org/abs/2501.18100v1
“NEW MODEL: GemmaCoder3-12b is a code reasoning model that improves performance on the LiveCodeBench benchmark 11 points over the base model. This makes for a useful code model because: – At 8 bit it runs nicely on 32gb of RAM – Gemma3’s 128k context length is great for large https://x.com/ben_burtenshaw/status/1907074356997423322
A Unified Image-Dense Annotation Generation Model for Underwater Scenes
https://hongklin.github.io/TIDE/
[2503.21755] VBench-2.0: Advancing Video Generation Benchmark Suite for Intrinsic Faithfulness https://arxiv.org/abs/2503.21755
