Research News & Updates
Your central hub for AI news and updates on Research. We're tracking the latest articles, discussions, tools, and videos from the last 7 days.
‘Sycophantic’ AI chatbots tell users what they want to hear, study shows
Scientists warn of ‘insidious risks’ of increasingly popular technology that affirms even harmful behaviourTurning to AI chatbots for personal advice ...
Why it matters:
The study highlights the need to develop more critically literate users and responsible AI chatbot developers that prioritize users' well-being over short-term engagement metrics.
Google CEO drops bombshell quantum computing breakthrough - TheStreet
Google CEO drops bombshell quantum computing breakthrough TheStreetGoogle’s Quantum Computer Makes a Big Technical Leap The New York TimesGoogle Just ...
Why it matters:
This breakthrough marks a significant step toward the first real-world application of quantum computing and could unlock a massive long-term revenue engine for Google.
Historical images made with AI recycle colonial stereotypes and bias – new research
Generative AI is known to mirror sexist and racist stereotypes, but it also carries a colonial bias that is reinforcing outdated ideas about the past....
Why it matters:
This research underscores the need for AI developers and users to critically evaluate the biases and limitations of generative AI tools, especially in contexts where they can perpetuate social injustices.
This Open Source Robot Brain Thinks in 3D
Open source language models are crucial to AI innovation. Can open robotics models do the same for physical machines?...
Why it matters:
The release of SPEAR-1 marks significant progress in making industrial robots more dexterous and flexible, potentially paving the way for more widespread AI adoption in industries such as manufacturing and logistics.
Could ChatGPT help smokers kick the habit? GW University researchers study new chatbot - WJLA
Could ChatGPT help smokers kick the habit? GW University researchers study new chatbot WJLA...
Why it matters:
The study on BeFreeBot highlights the potential of AI chatbots in smoking cessation, but further research is needed to determine their effectiveness.
The brain power behind sustainable AI
PhD student Miranda Schwacke explores how computing inspired by the human brain can fuel energy-efficient artificial intelligence....
Why it matters:
Miranda Schwacke's research into energy-efficient AI devices is an important step towards making AI more sustainable and environmentally friendly, and her approach to science communication and community engagement is commendable.
Researchers Find Severe Vulnerabilities in AI Browser
Go figure that letting an AI control your access to the internet can be a dangerous thing. The post Researchers Find Severe Vulnerabilities in AI Brow...
Why it matters:
This study highlights the significant security concerns associated with AI browsers, which may deter organizations from adopting this emerging technology.
AI Models Get Brain Rot, Too
A new study shows that feeding large language models low-quality, high-engagement content from social media lowers their cognitive abilities....
Why it matters:
This study highlights the risks of training AI models on low-quality data and underscores the need for more careful curation of training datasets.
AI chatbots fail at accurate news, major study reveals - DW
AI chatbots fail at accurate news, major study reveals DWAI assistants make widespread errors about the news, new research shows ReutersAI models misr...
Why it matters:
The study highlights the need for more accurate and trustworthy AI assistants in the media, and calls for governments and AI companies to take action to address these issues.
Creating AI that matters
How the MIT-IBM Watson AI Lab is shaping AI-sociotechnical systems for the future....
When Models Manipulate Manifolds: The Geometry of a Counting Task (new interpretability paper from Anthropic)
Why it matters:
This research provides crucial insights into task-specific mechanisms, enhancing the general understanding of deep learning processes and AI model behavior.
AI turns x-rays into time machines for arthritis care
Researchers at the University of Surrey developed an AI that predicts what a person’s knee X-ray will look like in a year, helping track osteoarthriti...
Why it matters:
This breakthrough AI system could revolutionize osteoarthritic knee care, enabling doctors and patients to better understand and manage the condition.
UC Santa Cruz Maps Coastal Flooding With NVIDIA Accelerated Computing
Coastal communities in the U.S. have a 26% chance of flooding within a 30-year period. This percentage is expected to increase due to climate-change-d...
Trending AI Repos & Tools
minimind
30637🚀🚀 「大模型」2小时完全从0训练26M的小参数GPT!🌏 Train a 26M-parameter GPT from scratch in just 2h!...
No videos found
Check back soon for video content
Community talk
[Experiment] Qwen3-VL-8B VS Qwen2.5-VL-7B test results
"Discovering state-of-the-art reinforcement learning algorithms"
Méta introduces Continuous Learning via Sparse Memory Finetuning: A new method that uses Sparse Attention to Finetune only knowledge specific Parameters pertaining to the input, leading to much less memory loss than standard Finetuning, with all it's knowledge storing capability
Google breakthrough in using Quantum computing for drug discovery and material science
[R] We figured out how to predict 32B model reasoning performance with a 1B model. 100x cheaper. Paper inside.
"DeepSeek-OCR: Contexts Optical Compression"
I open-sourced Stanford's "Agentic Context Engineering" framework - agents that learn from their own execution feedback
The Innovations in DeepSeek OCR
[P] Open-Source Implementation of "Agentic Context Engineering" Paper - Agents that improve by learning from their own execution feedback
Hierarchical Agentic RAG: What are your thoughts?
Un-LOCC (Universal Lossy Optical Context Compression), Achieve Up To 3× context compression with 93.65% Accuracy.
(Meta) "Encode,Think,Decode (ETD): Scaling reasoning through recursive latent thoughts." ¦¦ Improving the reasoning of base models by training them to iterate over a subset of reasoning-critical NN layers during mid-training. ¦¦ Modest improvements on Math Benchmarks (+36% on Math with OLMo 2 1B)
(Meta) The Free Transformer: An improvement to Transformers, adding a Latent Random Variable to the decoder, allowing the model to decide in a hidden state how it guides its output before it predicts the next token. ¦¦ +3% Compute overhead, +30% GSM8K, +35% MBPP and +40% HumanEval+ on a 1.5B Model.
UCLA researcher uses GPT-5 pro to solve an open problem in convex optimization
AlphaXiv,Compare the Deepseek-OCR and Mistral-OCR OCR models
Confirmed: Junk social media data makes LLMs dumber
[By GLM Team] Glyph: Scaling Context Windows via Visual-Text Compression
[P] Built a searchable gallery of ML paper plots with copy-paste replication code
I open-sourced Stanford's "Agentic Context Engineering" implementation - agents that learn from execution
Turn any dataset into a reasoning dataset easily and cheaply
Reverse Engineering and Tracing internal thoughts of LLM
"Self-Adapting Language Models"
Medical model: Bio-Medical-ContactDoctorVLLM
AI has passed the Music Turing Test
Is Chain of Thought Still An Emergent Behavior?
Alpha Arena is the first benchmark designed to measure AI's investing abilities. Each model is given $10,000 of real money, in real markets, with identical prompts and input data. AI
Scientists just proved that large language models can literally rot their own brains
LLMs can get brain rot from scrolling junk content online, just like humans
VISTA: A Test-Time Self-Improving Video Generation Agent (Google)
Brainjacking
ChatGPT 5 is on the 10th place on LMArena. Confirmed to be significantly worse than both GPT 4o and Gemini-2.5
"A Neural Symbolic Model for Space Physics"
"Paralysed man can feel objects through another person's hand"