AI Research

Latest AI research — summarized and explained.

Point to Navigate: AI Solves 3D With 2D

Imagine teaching a robot to navigate a room not by forcing it to calculate complex 3D vectors, but by simply asking it to point at a spot on a screen. This coun...

Read more

Fixing Machine Translation's Gender Blind Spot

Machine translation has a blind spot when it comes to gender. When converting text from English, a language with no grammatical gender, into Romanian, a highly...

Read more

New AI Architecture Fixes Shallow Reasoning

Imagine if your brain could remember the complex, unspoken thoughts it was processing before it finally spoke them out loud. Current AI models, known as autoreg...

Read more

AI Breaks Memory Bottleneck With Smart Prefetching

Imagine your computer’s fastest memory is so full it chokes on its own data. As large language models tackle longer, more complex reasoning tasks, they drown in...

Read more

AI Fixes Its Biggest Credit Assignment Flaw

Imagine teaching an AI to navigate a complex maze by only telling it if it reached the exit, leaving it to figure out which specific turns were the right ones....

Read more

AI Generates 37000 Tasks for 5 Cents

Training AI agents to handle complex, long-term tasks is notoriously expensive and difficult. Human experts can’t scale, and standard AI tools often break the d...

Read more

AI Agents Fail at Long Tasks

Imagine asking an AI to plan a complex project. At first, it seems brilliant. But as the task grows, the AI starts hallucinating, forgetting steps, and making d...

Read more

AI That Navigates Your Phone While You Sleep

Imagine an AI that doesn't just chat, but actually navigates your phone, clicks through apps, and finishes complex tasks while you sleep. This is the promise of...

Read more

AskChem Stops AI Hallucinations in Chemistry

Imagine asking an AI to summarize the latest breakthroughs in battery technology, only to have it hallucinate a connection that never existed. For scientists, t...

Read more

This 2.8T Model Changes AI Forever

The open-source AI race just got a serious injection of adrenaline. Researchers have unveiled Kimi K3, a massive 2.8 trillion parameter model that challenges th...

Read more

Ascend NPUs Outperform GPUs for Trillion-Parameter AI

Imagine training a model with trillions of parameters without your hardware melting down. That is the challenge researchers tackled by optimizing the DeepSeek-V...

Read more

AI Agent Audits Its Own Mistakes

Imagine an AI researcher that does not just search for answers, but actively audits its own work, spots its own mistakes, and digs deeper until it gets it right...

Read more

New AI Model Masters 3D Robot Interaction

Robots are getting smarter, but making them truly useful in the real world remains a stubborn challenge. A new research breakthrough introduces RynnBrain 1.1, a...

Read more

Train AI Agents on 2M Tokens for Free

AI agents are drowning in data. As these systems accumulate observations, tool outputs, and prior decisions over long trajectories, the context windows required...

Read more

Open Source AI Rivals Top Proprietary Image Generators

What if you could build a top-tier AI image generator without spending millions? Researchers have unveiled Boogu-Image-0.1, an open-source model that challenges...

Read more

Robots Get Lifelong Memory to Stop Amnesia

Robots are getting smarter, but they still suffer from a terrible case of amnesia. While modern vision-language models can see and react in the moment, they str...

Read more

AI Robot Solves Navigation Drift With Two-Brain Design

Imagine a robot that doesn’t just blindly follow commands but actually thinks about where it is going. Current AI navigation systems often drift off course or f...

Read more

Train Powerful AI Using Lessons from Weaker Models

Imagine teaching a genius by studying the study habits of a struggling student. It sounds counterintuitive, but researchers have discovered a way to boost power...

Read more

AI Finally Explains Molecular Behavior

What if artificial intelligence could finally explain the why behind molecular behavior, rather than just predicting the what? For decades, AI in science has be...

Read more

New Benchmark Exposes AI Data-Science Blind Spot

Large language models are rapidly evolving into autonomous data science agents, yet we still lack a reliable way to test if they truly understand cause and effe...

Read more

Voice Control Real-Time Video Generation

Imagine directing a video in real time, simply by speaking. Researchers have unveiled Vidu S1, a groundbreaking model that lets users control digital characters...

Read more

Music On A Raspberry Pi? Yes, Really

Imagine generating complex, high-quality music directly on a Raspberry Pi, with zero reliance on heavy cloud servers or Python frameworks. This is no longer a d...

Read more

Pharo LLMs Beat Giant Models

Most developers rely on large language models to autocomplete their code, but this technology largely ignores programming languages with smaller communities. If...

Read more

One AI Layer Matches Full Training

We have spent years building massive language models, assuming that every single layer in the neural network is equally vital to learning new skills. But what i...

Read more

How To Train AI On Problems It Cant Solve

Standard AI training methods often hit a wall when faced with difficult problems. If a model cannot already imagine the right path to a solution, it simply cann...

Read more

AI Speeds Up 85% Without Losing Accuracy

Imagine if your AI assistant could think faster without sacrificing accuracy. Researchers have introduced DSpark, a new framework that dramatically speeds up ho...

Read more

New AI Model Thinks in Three Modes

What if your AI could think in three different ways at once? Researchers have unveiled Nemotron-Labs-Diffusion, a groundbreaking language model that unifies aut...

Read more

High-Res AI With Zero Lag

Imagine a digital human that looks sharp enough to read, yet responds as fast as a real conversation. Researchers have just unveiled Wan-Streamer v0.2, a signif...

Read more

New Trick Makes AI Video Models Lightning Fast

Imagine generating high-quality images and videos with a fraction of the computational power. Diffusion transformers are currently the gold standard for AI crea...

Read more

AI Translations Are Fine But Humans Win

You are reading a translated novel. The prose flows, the plot makes sense, and the characters feel real. So why does a nagging sense of something missing linger...

Read more

Tiny AI Rivals Supercomputers With New Memory Trick

Imagine running a supercomputer-level AI brain on your smartphone, solving complex puzzles faster than the cloud. For years, this has been the holy grail of edg...

Read more

Why AI Search Agents Fail Without Clarification

Imagine asking a search agent for the best restaurant in town, only to receive a list of places in a different city because you forgot to mention the state. Thi...

Read more

AI Finally Solves Video Continuity Problem

Imagine a video where a character walks behind a wall and reappears on the other side, still holding the exact same coffee cup, wearing the same clothes, and re...

Read more

New Benchmark Exposes AI Data Agents

Data science is drowning in complexity. While large language models promise to automate the tedious grind of cleaning, analyzing, and interpreting raw data, we...

Read more

AI Router Cuts Latency by Ignoring Load

Imagine a high-speed highway where traffic lights ignore congestion, causing gridlock even when lanes are technically empty. This is the hidden bottleneck in se...

Read more

Stop Guessing Block Sizes in AI

Imagine speeding up your AI’s response time without sacrificing a single bit of accuracy. That is the promise of speculative decoding, a technique that lets a f...

Read more

Small AI Runs Locally Without Cloud

Everyone is obsessed with massive AI models, but what about the tiny ones? New research suggests that small language models are not just viable alternatives but...

Read more

Teach AI to Summarize Text Into Tiny Tokens

Imagine reading a thousand-page novel and instantly recalling only the sentences that matter to your question. That is the promise of Simplified Sparse Attentio...

Read more

Tatoxa Beats LLMs in Tatar Safety

Online harassment thrives in the shadows of low-resource languages, leaving communities like Tatar speakers vulnerable to unchecked abuse. Researchers have fina...

Read more

AI Hallucinations Are Predictable And Preventable

We have reached an era where AI can generate stunningly realistic video simulations of the future, yet these models are fundamentally broken. They produce visua...

Read more

New AI Method Cuts Memory Costs

Large language models are getting smarter at reasoning, but this new capability comes with a heavy price: bloated memory usage. As these models tackle complex p...

Read more

JetSpec Shatters LLM Speed Limits

Large language models are powerful, but they are painfully slow. They generate text one token at a time, creating a bottleneck that limits how quickly they can...

Read more

AI Agents Fail 80% of Complex Tasks

We often assume today’s AI agents are ready to handle complex, real-world jobs. But a new study suggests we are dangerously overestimating their abilities. Whil...

Read more

This AI Listens and Watches in Real Time

Imagine a digital companion that doesn’t just listen and speak, but watches and reacts in real time, creating a conversation that feels genuinely alive. This is...

Read more

Why Most Agentic AI Projects Fail

Forget the hype cycle. Building truly autonomous AI isn't just about prompting a chatbot; it is about engineering a complex, multi-layered system that thinks, r...

Read more

Chain-of-Thought Fails at Search-Based Reasoning

We often assume that if a computer can solve a problem with a short program, we can simply teach an AI to do the same by showing it the step-by-step logic. This...

Read more

AI Wastes Half Its Compute

Imagine a brain that only wakes up for hard problems. That is the promise of Grouped Query Experts, a new approach designed to fix one of the biggest bottleneck...

Read more

AI Fails Physics Research 33% of Time

Imagine an AI that doesn’t just answer questions but actively conducts scientific experiments, navigating complex physics and chemistry problems with the rigor...

Read more

Tiny Model Matches 10B Image Inpainting Quality

Imagine filling in missing parts of an image with stunning realism, but doing it on your laptop instead of a server farm. For years, high-quality image inpainti...

Read more

Why Your Voice Assistant Is Slower Than It Needs To Be

Your voice assistant is likely slower than it needs to be, not because of weak hardware, but because of a stubborn scheduling habit. Most automatic speech recog...

Read more

AI Can’t Build Complex Games Despite Success

AI has mastered generating individual game assets and writing simple scripts, but it still struggles to build entire games from scratch. This gap exists because...

Read more

Teach AI New Code Languages Without Breaking It

Imagine asking an AI to write code in a language it has never seen before. For most developers, this sounds impossible. Yet, in the real world of proprietary so...

Read more

AI Learns from Mistakes to Boost Reasoning

What if the secret to smarter AI isn’t just getting the right answer, but mastering the art of fixing a wrong one? Researchers have unveiled a new training meth...

Read more

How EfficientRollout Cuts LLM Training Lag

Reinforcement learning (RL) has become a representative post-training paradigm for LLMs, enabling strong reasoning and agentic capabilities. However, rollout ge...

Read more

Why AI Agents Keep Forgetting What They Can't See

We trust AI to remember everything, but what happens when it cannot see what it needs to act? As multimodal models move from passive chatbots to active agents i...

Read more

OneRank Fixes the Flaw in AI Recommendation Engines

Recommender systems are the invisible engines driving what you watch, buy, and read, yet they are built on a flawed foundation. Most modern systems treat the co...

Read more

Skip Fine-Tuning: New Method Boosts AI Reasoning

Chain-of-Thought prompting gives AI models the ability to reason step-by-step, but it comes with a steep price: high latency and massive inference costs. The us...

Read more

New AI Tool Cuts Coding Tokens by 60%

Imagine asking a coding assistant to fix a bug, only to watch it waste hours sifting through thousands of irrelevant files before it even starts writing code. T...

Read more

Orchestra-o1 Conducts AI's Multimodal Symphony

Imagine an AI that doesn't just read your text but watches a video, listens to audio, and analyzes images simultaneously to solve complex problems. This is the...

Read more

AI Agents Fail 60% in Dynamic Worlds

Most large language model agents are tested in static, unchanging environments. This creates a dangerous illusion of competence. In the real world, conditions s...

Read more

AI Speeds Up 14x With New Attention Trick

Imagine an AI that can read an entire codebase or remember every detail of a year-long conversation without slowing to a crawl. Current large language models hi...

Read more

AI Solves Multilingual Medical Diagnosis Gap

Imagine asking a doctor in rural India about a mysterious rash, but the AI assistant only understands English and cannot interpret your photo. This is the reali...

Read more

Slim Submodels Speed Up AI Inference

Imagine if your computer could read your mind just enough to guess what you want to say next, then double-check its hunches without breaking a sweat. That is th...

Read more

How AI Teams Beat Visual Hallucinations

Imagine asking an AI to analyze a complex image, only for it to confidently hallucinate details that aren’t there. This happens because traditional models tend...

Read more

xLSTM Beats Mamba-2 and DeltaNet in Efficiency Battle

Transformers dominate modern sequence modeling, but their quadratic attention incurs substantial computational cost. Subquadratic architectures offer a scalable...

Read more

AI Passes Tests But Fails Real Jobs

AI systems are crushing standardized tests, yet they remain frustratingly useless in actual professional environments. This disconnect suggests the problem isn’...

Read more

AI Teaches Itself by Playing Both Sides

Imagine an AI that teaches itself by playing both sides of the game. Researchers have introduced Role-Agent, a framework that breaks the traditional barrier bet...

Read more

AI Cuts Memory Needs by Ignoring Most Data

Imagine an AI that reads a thousand-page manual but only remembers the five sentences that matter. That is the promise of FlashMemory-DeepSeek-V4, a new system...

Read more

Most AI Agents Fail When Tools Break

We assume AI agents are robust, but they are actually fragile. When a tool fails, most models do not recover; they just keep failing in loops. This disconnect b...

Read more

Silent Reasoning: AI Thinks Without Speaking

Large language models are notorious for their chatty nature, often wasting time and compute on verbose step-by-step explanations. But what if the model could th...

Read more

AI Rewrites Its Own Instructions to Get Smarter

What if your AI assistant could rewrite its own instruction manual to get smarter? Researchers have introduced SePO, a self-evolving prompt agent that stops tre...

Read more

AI Agents Fail When Rules Change

We assume AI agents can handle complex real-world tasks, but a new benchmark reveals they struggle significantly when rules change mid-game. Researchers have in...

Read more

AI That Learns From Its Own Mistakes

Imagine an AI that doesn’t just write code but actually learns from its own mistakes to invent better algorithms. This is the promise of MLEvolve, a new framewo...

Read more

AI Role-Playing Agents Fail to Evolve

Imagine a character who refuses to grow. They start as a timid hero but remain timid even after surviving a war. This is the current reality for many role-playi...

Read more

AI Memory Failures Revealed by New Benchmark

We watch hours of video daily, yet current AI models are surprisingly bad at remembering what they just saw. As multi-modal systems tackle longer, more complex...

Read more

Domino Speeds Up AI Fivefold Without Sacrificing Accuracy

Imagine speeding up your AI’s responses by five times without sacrificing accuracy. That is the promise of a new framework called Domino, which tackles one of t...

Read more

AI Learns Robot Motion Like Human Language

Imagine a robot that can instantly adapt to any movement it has never seen before, without needing a single hour of retraining. This is no longer science fictio...

Read more

VideoMLA Cuts Memory 92% for Minute-Scale Video

Video generation is hitting a wall. As models attempt to create longer, more coherent clips, the memory demands of tracking context explode, forcing a trade-off...

Read more

Search Agents Now Dive Straight Into Raw Text

Imagine a search agent that doesn't just skim the surface of indexed documents but dives straight into the raw text to find answers. This is the core promise of...

Read more

Your Video Generator Is Stuck on Frame One

Video generation models have a fatal flaw: they are too obsessed with the first frame. This initial image acts as a rigid anchor, locking the camera angle and s...

Read more

AI Diagnoses Its Own Mistakes to Fix Prompts

Prompt engineering is usually a grind. You tweak words, pray for better results, and repeat. It is tedious, sensitive to tiny changes, and often feels like gues...

Read more

Bias in AI Search Is Learned Not Built In

We have all felt the frustration of a search engine ignoring the perfect answer just because it was buried at the bottom of a page. It turns out, this isn't jus...

Read more

AI Agents Fail to Track Real-World Changes

We treat memory like a hard drive, but for AI agents, it is more like a living diary. As multimodal models take on longer, more complex tasks, the way they reme...

Read more

AI Agents Waste Time Waiting

We assume AI agents are efficient, but they are actually waiting around for nothing. New research reveals that current large language models struggle significan...

Read more

AI Agents Fail Because of This Hidden Gap

Why do AI agents keep failing at complex tasks? The answer lies in a hidden flaw in how they learn: a disconnect between thinking and doing. Researchers have id...

Read more

Train AI in Chaos to Stop It Crumbling

LLM agents look brilliant on paper but crumble in the real world. Why? Because they are trained in sterile labs, not the chaotic mess of actual user interaction...

Read more

Why AI Fails at Minute-Long Videos

Current audio-visual AI is stuck in a short-term memory problem. While models can generate impressive clips, they consistently fail to maintain quality over lon...

Read more

AI Cheating Exposed by Deleting Reasoning Steps

Large language models are boasting incredible reasoning skills, but a growing shadow looms over these achievements: data contamination. While we celebrate AI so...

Read more

One AI Model Replaces Dedicated Speech Tools

What if one AI model could transcribe your voice, speak with human-like emotion, and hold a real-time conversation, all without switching systems? Researchers h...

Read more

Transform Full Attention LLMs to Sparse in 100 Steps

Long-context AI is hitting a wall. The quadratic cost of processing massive amounts of text is slowing down models to a crawl, forcing a painful choice between...

Read more

AI Fails at Video Because It Reads

Why do AI models keep missing the subtle cues in a video? Current multimodal systems force continuous audio and visual signals into rigid text tokens to reason,...

Read more

AI Personality Checks Are Mostly Biased

We often assume that when an AI judges your personality, it is carefully analyzing your behavior. But new research suggests these models might just be guessing...

Read more

Why Self-Distillation Fails at Math Reasoning

Self-improving AI models have long promised to learn from their own mistakes, but a new technique reveals why that process often fails silently. Researchers hav...

Read more

AstraFlow Makes Agentic AI Training 2.7x Faster

Reinforcement learning is increasingly used to improve the reasoning, coding, and tool-use capabilities of large language models, but agentic RL remains prohibi...

Read more

Modern LLMs Have Silent Activation Spikes

If you are building AI apps, your model’s internal numbers might be breaking your code. New research reveals that activation spikes in modern large language mod...

Read more

AI Fails at Real Dexterity Without This Benchmark

Robotic hands are finally getting dexterous, but how do we know if they are actually good at it? Most existing tests treat complex hands like simple clamps, mis...

Read more

AI Generates Real-Time Video in 1 Step

Imagine generating video in real-time, frame by frame, with the speed of a heartbeat. This is the holy grail for interactive AI, yet current methods struggle wi...

Read more

This Simple Trick Beats Gold-Medal Math

Gold-medal performance in the world’s toughest math and physics competitions used to require massive, specialized systems. Now, researchers have shown that a su...

Read more

AI Fails at Simple Shapes

Look at a simple drawing of nested circles. To you, it is obvious which shape contains which. To the most advanced AI models, it is a nightmare. Researchers hav...

Read more

Open Source Agent Framework Beats Proprietary Giants

Open-source AI is finally catching up to the giants, but only if we stop reinventing the wheel every time. Researchers have unveiled Orchard, a new framework th...

Read more

Train Millions of AI Models Without Rebuilding Them

Imagine managing millions of specialized AI personalities without duplicating the massive computing power required to build them. Researchers have unveiled MinT...

Read more

AI That Fixes Its Own Mistakes

Imagine an AI that doesn’t just follow your prompt but actively thinks about what you really mean, then critiques its own work to fix mistakes before showing yo...

Read more

Robots That Predict the Future Are Here

Robots are no longer just following scripts; they are learning to predict the future. This shift is driven by world models, a technology that allows machines to...

Read more

New AI Memory Breakthrough Slashes Costs

Imagine a language model that remembers everything you say, yet refuses to slow down as the conversation grows longer. For years, the trade-off has been brutal:...

Read more

Parallel Search Cuts AI Tool Calls by 5x

Imagine a search agent that doesn't just dig deeper, but looks wider. Current multimodal AI tools are painfully slow, processing one piece of information at a t...

Read more

LLMs Write GPU Code That Is Wrong And Slow

LLMs are touted as the future of high-performance computing, promising to write faster GPU code than humans. But a new benchmark reveals a harsh reality: while...

Read more

AI Finally Masters the Art of Reading Tables

We have mastered text, but the world’s data lives in tables. For years, AI has struggled to read spreadsheets with the same fluency it reads novels. This gap is...

Read more

AI’s Next Big Leap: Hearing and Seeing Together

Imagine a world where AI doesn't just see or hear, but truly experiences both simultaneously. This is the promise of Audio-Visual Intelligence, a frontier that...

Read more

New Method Fixes Blurry Fast AI Images

Imagine generating high-quality images in just a few steps, without the blurry artifacts that usually plague fast AI models. Researchers have cracked a key bott...

Read more

AI Understands More By Remembering Less

Long-context AI systems are drowning in data, struggling to find the needle in the haystack. Researchers have found a way to mimic human cognition by compressin...

Read more

AI Agents Think Deeply to Save Energy

We often assume that complex AI agents succeed because of their intricate orchestration layers, but new research suggests the real magic happens inside the mode...

Read more

Forget Billions: Simple Data Builds Elite Search AI

Frontier AI agents are usually the exclusive domain of tech giants with bottomless budgets. But a new open-source model proves that you don't need billions in c...

Read more

AI Agents Fail at Basic Cross-App Tasks

We rely on computers to handle complex professional workflows, yet our most advanced AI agents still struggle to navigate between different programs. New resear...

Read more

End-to-End Training Shatters Image Generation Limits

Image generation has long been stuck in a two-step dance, where one system compresses a picture and another tries to recreate it. This disconnect often leads to...

Read more

Teach AI to See by Making It Speak

What if you could teach a computer to see by making it speak? Researchers have introduced a new way to train AI to understand images, turning visual data into l...

Read more

Why AI Ignores Your Orders

We tell AI models exactly how to think, but they often ignore us. New research reveals that large language models have a stubborn preference for sensibility ove...

Read more

AI Now Writes Authentic Arabic Poetry

Arabic poetry is more than just words; it is a vital thread in the cultural fabric of millions. Yet, for years, artificial intelligence has largely ignored the...

Read more

Speculative Decoding Speeds Up AI Training by 2.5x

Training frontier AI models is hitting a wall. As language models get smarter, the process of refining them through reinforcement learning becomes painfully slo...

Read more

NVIDIA Just Made Multimodal AI Tiny and Fast

Imagine an AI that doesn't just read your screen but hears your voice, watches your video calls, and understands complex documents in real time. NVIDIA has just...

Read more

Stop Guessing: AI Now Verifies Image Edits

Current AI image editors are stuck in a loop of guesswork. They rely on vague scoring systems that fail to understand the nuance of your specific instructions,...

Read more

Tiny AI Model Beats Giants With New Trick

Large language models are getting bigger, but they don’t have to stay that way. Researchers have unveiled a method to shrink massive diffusion-based AI models i...

Read more

Debug Your LLM Like Software Code

What if you could debug a large language model’s ignorance with the same precision as fixing a bug in software code? Researchers have turned the chaotic process...

Read more

Why Perfect Audio AI Feels Robotic

We have spent years teaching AI to think in words, but what happens when we force it to reason through sound? A new breakthrough challenges the dominant method...

Read more

AI's 3D Intelligence Scores Are a Lie

Current tests for how well AI models understand 3D space are fundamentally broken. They rely on outdated data that ignores how modern vision-language models act...

Read more

Why Text Search Fails AI Agents

Finding the right AI agent for a specific task feels like searching for a needle in a haystack, but not because the needles are hidden. They are there, yet stan...

Read more

AI Produces Science Without Actually Thinking

AI systems are increasingly tasked with conducting autonomous scientific research, but a new study reveals a disturbing gap: these agents can produce results wi...

Read more

Harmless Audio Silently Breaks AI Safety

You might think training an AI on harmless audio is safe. It isn’t. Researchers have discovered that even completely innocent data can silently break the safety...

Read more

AI That Knows When to Shut Up

Large language models are getting better at reasoning, but they have a dangerous habit: when faced with impossible questions, they often lie to please you. This...

Read more

AI Finally Learns to Think in Time

Large Language Models are great at words, but they struggle to make sense of numbers over time. Current benchmarks are messy and ambiguous, leaving AI systems g...

Read more

Creating 3D Images Makes AI Smarter

We often assume that if an AI can describe a scene perfectly, it truly understands space. But describing is not the same as building. A new study challenges thi...

Read more

One Model to See and Paint with Discrete Diffusion

Imagine an AI that doesn't just describe what it sees, but can also paint it from scratch, all within a single brain. This is no longer science fiction. Researc...

Read more

LLMs Write GUI Code That Fails When You Try It

Large language models are getting better at writing code, but they still struggle to build functional graphical interfaces. While these AI systems can churn out...

Read more

AI Video Speed Doubled Without Losing Quality

Speeding up AI video generation has always been a balancing act between quality and speed. Researchers have now cracked the code that makes streaming high-quali...

Read more

OneVL Makes Autonomous Cars Think Instantly

Autonomous cars are stuck in a paradox: to drive safely, they must think deeply, but thinking takes time. Current systems rely on step-by-step reasoning that is...

Read more

Your AI Assistant Finally Remembers Who You Are

Your AI assistant knows your favorite coffee order but forgets your birthday. It is smart, yet fundamentally shallow. Current models treat every conversation as...

Read more

New Test Exposes Deep Research Agents' Fatal Flaws

Imagine asking an AI to conduct a deep research project, only for it to confidently invent sources or miss crucial details because the internet is too chaotic t...

Read more

AI Now Builds Navigable 3D Worlds from One Photo

Imagine stepping into a digital realm that instantly materializes from a single photo or a simple text description, complete with realistic lighting and charact...

Read more

Why Your AI Images Are Mediocre and How to Fix It

Why do most AI image generators keep producing mediocre results? Because their evaluators are essentially guessing, reducing complex human preferences to a sing...

Read more

Self-Revision Turns Binary Rewards Into Dense Supervision

Imagine teaching a student to ace a test not by showing them the right answers, but by letting them critique their own wrong ones and then learning from that se...

Read more

Smarter AI Cripples Smaller Models Unless You Fix This

Using a smarter AI to write training data for a smaller one sounds like a perfect shortcut, yet it often backfires spectacularly. New research reveals that simp...

Read more

New Tree Structure Doubles Speculative Decoding Speed

Large language models are powerful but painfully slow, often chugging along one word at a time. Researchers have found a clever workaround: using a lightweight...

Read more

AI Solves Hard Problems With Minimal Hints

Large language models often stumble on complex reasoning tasks because they lack the specific hints needed to find the right path. While current methods try to...

Read more

Unified Agents Fail 55% of Real-World Tasks

Imagine an AI assistant that can look at a screen, search the web, and write code all at once to solve complex problems. While these unified digital agents are...

Read more

Synthetic Speed Tests Overestimate Real AI Gains

Large language models are getting faster thanks to speculative decoding, but how do we truly know if they are fast enough? Current testing methods often paint a...

Read more

Tiny AI Beats Giant Models At Real Robot Control

Imagine a robot that doesn't just see the world but truly understands it, predicting how objects move and planning complex actions in real time. New foundation...

Read more

AI Automates Weeks of Manual Data Labeling

Answering complex research questions often requires digging through massive document collections to find structured evidence. Traditionally, this means manually...

Read more

AI Agents Fail at Real-World Web Tasks

AI agents promise to automate your inbox and handle routine life tasks, but can they truly navigate the messy reality of everyday online interactions? A new eva...

Read more

Why Reasoning AI Fails Before It Improves

The idea that reinforcement learning makes AI smarter while supervised fine-tuning just helps it memorize is a popular belief, but new findings suggest this sim...

Read more

SkillClaw Lets AI Agents Evolve From Everyone's Mistakes

Imagine a digital workforce where every mistake made by one employee instantly teaches everyone else how to do better. Currently, AI agents rely on static skill...

Read more

Unlock Hidden AI Skills With One Linear Trick

Imagine giving a smaller AI brain the genius-level logic of a massive one without lifting a finger to retrain it. New research suggests this isn't magic, but a...

Read more

Faithful GRPO Cuts AI Hallucinations by 93%

Multimodal AI models are getting smarter at solving visual puzzles, but they often cheat to get there. While these systems produce correct final answers, their...

Read more

AI Simulated Humans Are Too Perfect And Boring

Large language models are poised to become powerful user simulators, yet they currently fail to mimic the messy reality of human life. Existing tests trap AI in...

Read more

AI Overthinking Slashed By 42 Percent

Large language models are getting smarter at solving complex problems, but they often get stuck in a loop of unnecessary thinking. Instead of finding an answer...

Read more

Your AI Assistant Now Speaks 1.7x Faster

Autoregressive language models typically generate text one token at a time, even when the next token is obvious. This hesitation slows down conversations and wa...

Read more

AI Team Finds Papers in Seconds

The explosion of scientific literature has left researchers drowning in data, making it nearly impossible to find, evaluate, and synthesize relevant work quickl...

Read more

AI Thinks Silently 30x Faster Than Before

Imagine an AI that reasons like a human but thinks at lightning speed. Current methods force models to write out every step of their thought process in text bef...

Read more

Small AI Teams Beat Big Ones With Smart Memory

Large language model teams face a critical choice: grow by adding more members or learn from past experiences? New research suggests that simply hiring more age...

Read more

New Framework Fixes AI Blind Spot in Rare Scientific Images

Scientists often struggle to identify rare anomalies in images because standard AI models are trained on vast amounts of common examples, leaving them blind to...

Read more

Rare Inputs Break AI; Common Text Fixes It

Imagine asking a smart AI to solve a complex problem, only for it to stumble because the question itself is too rare in its training data. New research reveals...

Read more

AI Driver Finally Understands Real 3D Depth

Imagine an AI driver that doesn't just guess what comes next but actually understands the three-dimensional geometry of the road ahead. New research introduces...

Read more

Polite AI Agents Still Execute Deadly Harmful Steps

Computer-use agents are evolving beyond simple chatbots to become persistent workers that can manipulate files and run code. However, this new capability create...

Read more

New AI Filter Cuts Review Costs By 82%

As AI agents handle more complex tasks, their decision trails grow too vast to review manually. Researchers have developed a lightweight system that acts like a...

Read more

AI Sees Better But Forgets How To Talk

Turning a standard language model into a system that sees images often breaks its ability to write or reason with words. This happens because the new visual tra...

Read more

New AI Agents Evolve Without Human Rules

Imagine a team of digital explorers that never sleep, constantly refining their own skills to solve problems no human has cracked yet. Current AI methods often...

Read more

AI Decides Before It Thinks a Single Word

When you ask an AI to solve a problem, does it think first and then act, or does it decide the answer before it even starts writing? New research reveals that l...

Read more

AI Cuts Coding Wait Time By 55%

Current AI coding tools waste massive amounts of time by waiting until a program is fully written before running it. This old method forces the system to sit id...

Read more

Why AI Now Thinks in Hidden Space

Modern AI is quietly moving beyond simple word-by-word generation to operate in a hidden realm of continuous data known as latent space. This shift occurs becau...

Read more

AI Learns Better Code From Its Own Mistakes

Can an artificial intelligence learn to write better code just by studying its own mistakes? New research suggests the answer is yes, and it does so without nee...

Read more

New Hybrid System Cracks AI Depth Bottleneck

Large language models are getting smarter by running longer and harder, but standard AI engines choke on the extra computing power required. Researchers have cr...

Read more

New Method Lets Code AI Reason Exactly When Needed

Modern coding assistants are hitting a wall: they often expend excessive mental effort on simple tasks while failing to address complex problems that require de...

Read more

AI Learns To Commit Like A Real Developer

When large language models produce code that functions perfectly yet gets rejected by maintainers, the issue is rarely bugs. The true culprit is a lack of organ...

Read more

Recover Lost Vector Files Instantly with AI Magic

Imagine transforming a blurry, uneditable photograph of a complex engineering diagram into a pristine, scalable vector file with a single command. This is no lo...

Read more

Stop Hand-Designing Heuristics: Let Agents Evolve Code Autonomously

Traditional software development relies on human experts to manually tune complex algorithms such as multi-head attention kernels. However, a new method called...

Read more

LLMs Master Complex Financial Tasks with New Benchmark

While artificial intelligence promises to revolutionize finance, can Large Language Models truly handle the gritty reality of banking tools? A new benchmark cal...

Read more

Skip Training To 2x Faster Diffusion LLM Decoding

Imagine waiting an eternity for your AI to generate a story, only to realize it can do so twice as fast without lifting a finger. A new method called S2D2 promi...

Read more

Stop Predicting Just the Top Answer

Modern large language models have long relied on a strategy known as mode collapse, where they are trained to predict only the single most probable answer for a...

Read more

Code Flow Training Unlocks Dynamic Software Logic Evolution

Traditional code large language models rely on static representations that often miss the nuanced evolution of software logic. A new approach called Code-Flow m...

Read more

Unlock Precise Face Editing Without Losing Identity

Facial filters have long struggled with a fundamental flaw: tweaking an expression often ruins a person's identity. Researchers at Stockholm Tech and AI Konsult...

Read more

First One Trillion Parameter Scientific Model Released

In a move that redefines the boundaries of open-source artificial intelligence, researchers have unveiled Intern-S1-Pro, marking a historic milestone with its o...

Read more

New Dataset Reveals Secret To Building General-Purpose AI Agents

Computer-use agents stand on the brink of revolutionizing desktop automation, but a critical shortage of continuous human demonstration videos has stalled their...

Read more

One Parameter Revs Up Diffusion Transformers

Hidden efficiencies within diffusion transformers have long gone unnoticed, but a new method called Calibri is forcing them to reveal their true potential. By t...

Read more

Nano Banana Pro: Restores Real Images Without Data Limits

Current image restoration tools struggle with real-world dirt and scratches because their training data is too limited or mismatched to actual conditions. While...

Read more

AI Agents Now Master Video Understanding Without Manual Workflows

Multimodal large language models have struggled for years to interpret long videos, often getting lost in redundant frames and endless sequences of data. Previo...

Read more

Self-Distillation Silences Uncertainty and Kills Math Reasoning

Artificial intelligence models that practice self-distillation are getting smarter faster, but a new study reveals a dangerous side effect: they are learning to...

Read more

UI-Voyager Turns Every Failed Click into an Evolutionary Leap

In the vast landscape of mobile automation, failure has long been viewed as a dead end rather than a stepping stone. A new approach flips this narrative by teac...

Read more

Discover hidden agent vulnerabilities with our new red-teaming breakthrough

While previous safety tests focused on detecting harmful text from large language models, a new gap has emerged regarding multi-step actions using tools like th...

Read more

New Dataset Solves Multi-Image Generation Bottleneck

For years, generating images from multiple visual references has stalled because models fail to connect disparate concepts. Current AI struggles with density an...

Read more

New TTS Model Learns Voices From Just 3 Seconds Of Audio

Imagine cloning a voice instantly with just three seconds of recording, defying the industry standard that typically demands hours of sample data. This bold lea...

Read more

Break the 1 Million Token Limit With This New Memory Model

Long-term memory remains elusive for artificial intelligence. While human brains effortlessly retain lifetime experiences, current large language models are typ...

Read more

Your Code Passes Tests But Will It Survive?

Modern coding AI agents face a troubling reality: they can pass tests today, but their code crumbles under pressure tomorrow. New research reveals that these to...

Read more

Why Only 9% Of LLM Agents Can Self-Improve

LLMs promise to build themselves, but reality is fragile. Despite years of research, fewer than one in ten agents actually utilize automated optimization. The i...

Read more

First-Person POV Video Benchmark Exposes Agentic Reasoning Gaps

Imagine playing a complex video game while wearing a mask that blindfolds you, forcing the AI to guess your moves solely from shaky camera footage. That is esse...

Read more

Models Train Themselves Better Without Any Human Help

Multimodal AI models have mastered complex reasoning tasks, but their progress has long depended on expensive human-annotated data or reliance on teacher models...

Read more

Generative Video Models Are Missing the Real World's Motion Pulse

New research reveals that even the most realistic AI-generated videos often fail to capture the true speed of real-world motion. While current models excel at v...

Read more

Train Any Control Modality Fast With AVControl Framework

Generating realistic video and audio content usually demands complex setups where every new control requirement, such as adjusting depth or camera movement, for...

Read more

3DGS Tracking Fixed by Spectral Moments

While 3D Gaussian Splatting promises real-time, photorealistic video tracking, a hidden flaw has long plagued its practical application in the wild. Standard me...

Read more

Stop Overfitting: Build Perfect 360° Dynamic Objects From One Video

Existing dynamic object reconstruction tools often hit a wall when attempting to capture full 360-degree views, struggling to maintain consistent geometry becau...

Read more