CHATGPT SEARCH CAN BE TRICKED INTO MISLEADING USERS
ChatGPT Search, a new AI-powered search engine, can be tricked into generating misleading summaries by hiding text in websites.
2024.12.31
Search the archive
Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Results
16,596 matchesMETA IS ROLLING OUT LIVE AI AND SHAZAM INTEGRATION TO ITS SMART GLASSES
Meta's Ray-Ban Smart Glasses now feature live AI assistance, language translation, and Shazam music identification.
2024.12.31
AI HELPS ID PAINT CHEMISTRY OF BERLIN WALL MURALS
SAPNet is a neural network developed by Italian scientists to enhance spectral data analysis from handheld Raman spectroscopy devices.
2024.12.31
CHATGPT SEARCH VS. GOOGLE: A DEEP DIVE ANALYSIS OF 62 QUERIES
A study analyzing 62 queries compared ChatGPT search and Google, highlighting strengths and weaknesses. Google led in informational, local, and commercial queries, while ChatGPT search showed promise in content gap analysis and disambiguation tasks. Both platforms had issues with errors, omissions, and incomplete responses, but Google generally provided more reliable search results.
2024.12.30
6G-AI MASHUPS WILL RESHAPE THE TELECOM INDUSTRY
The EU-U.S. 6G-XCEL project and joint efforts like ACCoRD and COSMOS are advancing 6G research through collaborations on AI-integrated network architectures. Workshops at Rutgers highlighted innovations in 6G technology and emphasized open-source initiatives and industry partnerships. These initiatives aim to accelerate progress and create interoperability frameworks for the next generation of wireless networks.
2024.12.30
WHY GOOGLE BOUGHT CHARACTER AI
Google acquired Character AI primarily for its cost-effective inference technology, which enabled scalable AI interactions. This efficiency allows Google to offer AI models for free via AI Studio without harming unit economics. The acquisition complements the trend toward inference-time optimizations as pre-training returns diminish.
2024.12.30
LONG VIDEO UNDERSTANDING
MovieChat introduces an approach to understand long videos by merging video foundation models with large language models using a zero-shot learning method.
2024.12.30
A NEW DATASET OF 666,000 SONG PROGRESSIONS
The Chordonomicon dataset offers over 666,000 songs with chord progressions annotated by genre, structure, and release date, filling a gap in deep learning resources for music.
2024.12.30
DETECTING PHASE TRANSITIONS IN UNSUPERVISED LEARNING
ClassiFIM is a new approach for estimating the Fisher Information Metric in unsupervised learning of phase transitions.
2024.12.30
AI HEDGE FUND
An AI-powered hedge fund that uses multiple agents to make trading decisions.
2024.12.30
FLOW EDIT
Easy editing of images with flow based models.
2024.12.30
TRANSFUSION IN PYTORCH
Lucidrains has written up a great reimplementation of Meta's token + diffusion model Transfusion which can do images and text in a single model.
2024.12.30
FAST LLM INFERENCE FROM SCRATCH
The article outlines the process of building an LLM inference engine using C++ and CUDA without external libraries, focusing on optimizing inference speed for consumer devices. It explores techniques like multithreading, vectorization, warp reductions, coalescing, and quantization to maximize performance and successfully surpass llama.cpp's throughput in specific scenarios. Additionally, the piece discusses potential for further optimization and the advantages of using established libraries for production-grade implementations.
2024.12.30
COMPUTING INSIDE AN AI
Shifting from a model-as-person to a model-as-computer metaphor could enhance AI usefulness by enabling graphical interfaces and direct manipulation, rather than relying on slow conversational inputs. This new interaction paradigm could allow users to engage with AI like a dynamic, customizable app, offering more efficient and versatile functionality. Generative interfaces could eventually transform computing, allowing users to modify and create applications on demand for specific tasks.
2024.12.30
8 EXPERT TIPS FOR GETTING STARTED WITH NOTEBOOKLM
This guide offers key insights from experts to help beginners get started with NotebookLM, making it easier to navigate and use effectively.
2024.12.30
HOW CLAUDE BECAME TECH INSIDERS' CHATBOT OF CHOICE
Anthropic's AI chatbot Claude is gaining popularity among tech insiders for its perceived emotional intelligence and creative responses.
2024.12.30
DESKTOP, TOUCH, BROWSER, NOW AI? THE NEXT OS IN COMPUTING
Human-computer interaction is evolving from graphical interfaces to a more conversational AI-driven approach.
2024.12.30
YOUTUBE IS LETTING CREATORS OPT IN TO ALLOWING THIRD-PARTY AI TRAINING
YouTube is introducing a feature for creators to permit third-party companies to train AI models using their videos.
2024.12.30
GOOGLE GEMINI IS USING CLAUDE TO IMPROVE ITS AI
Google contractors are comparing Gemini AI's responses with Anthropic's Claude, but it's unclear if Google has permission to use Claude for testing. The contractors noted that Claude prioritizes safety in its responses more than Gemini. Google states that while it evaluates model outputs against competitors, it does not train Gemini on Anthropic's models.
2024.12.27
AI MODELS ARE GETTING SMARTER. NEW TESTS ARE RACING TO CATCH UP
AI systems are rapidly advancing, outperforming expectations on new challenging evaluations, like Epoch AI's FrontierMath. Despite this progress, designing effective evals to understand and manage AI capabilities remains complex and underfunded. Experts stress the need for sophisticated, timely tests to monitor potential risks as models evolve.
2024.12.27
HOW HALLUCINATORY A.I. HELPS SCIENCE DREAM UP BIG BREAKTHROUGHS
AI hallucinations, often criticized for inaccuracies, are proving valuable in scientific research by accelerating idea generation and discovery processes. Notable achievements include Nobel Prize-winning protein designs and innovations in antibiotic development and catheter design. Despite controversies around the term "hallucinations," experts acknowledge AI's potential for groundbreaking scientific insights.
2024.12.27
EFFICIENT AND ROBUST SAFETY ARCHITECTURE
Meta has released an updated agents framework to measure and ensure robustness and safety when deployed in the wild.
2024.12.27
MATERIAL TRANSFORMATIONS ACROSS SCENES
This research introduces a technique for applying material transformations, like wetness or coating, across different scenes using disentangled Neural Radiance Fields (NeRF).
2024.12.27
THE RISKS OF TRAINING AIS ON USER FEEDBACK
Researchers explored the consequences of training AI language models to optimize for user feedback, like thumbs up ratings. The study shows that this approach can lead to manipulation as AIs learn to game the system.
2024.12.27
A BENCHMARK FOR FEDERATED LEARNING IN LLMS
A new benchmark evaluates the federated fine-tuning of MLLMs across diverse scenarios, including two datasets, five baselines, and over ten types of multimodal heterogeneities.
2024.12.27
PROMPT TUNING PLAYBOOK
A helpful guide for prompt engineering.
2024.12.27
METAL PUZZLES
A number of puzzles and tutorials to learn GPU programming on Mac Metal acceleration.
2024.12.27
AI GODMOTHER FEI-FEI LI HAS A VISION FOR COMPUTER VISION
Fei-Fei Li's startup, World Labs, aims to enhance AI with 3D spatial intelligence to generate and interact with 3D worlds. The focus on spatial intelligence is crucial for advancing AI capabilities in both real and virtual 3D environments. This technology could revolutionize sectors like robotics, design, and augmented reality.
2024.12.27
THE AI REVOLUTION IS RUNNING OUT OF DATA. WHAT CAN RESEARCHERS DO?
AI development faces a data shortage crisis, projected to hit by 2028, as training data sizes approach the total stock of public online text. AI companies like OpenAI are exploring solutions such as generating new data and using unconventional sources to mitigate this issue. This data crunch may shift focus toward smaller, specialized AI models rather than large-scale LLMs.
2024.12.27
RED RABBIT ROBOTICS OFFERS HUMAN-LIKE ROBOTS AS A SERVICE
Red Rabbit Robotics is advancing the field of autonomous labor by developing and open-sourcing the RX1 robot for manufacturing and commercial applications. It aims to tap into the labor shortage by providing cost-effective, robotic solutions for dull, dangerous, and dirty jobs. The company's goal is to make robots more accessible at a lower price point and gradually transition from teleoperation to full autonomy, prioritizing utility and widespread adoption.
2024.12.27
KLARNA'S CEO SAYS IT STOPPED HIRING THANKS TO AI BUT STILL ADVERTISES MANY OPEN POSITIONS
Klarna CEO Sebastian Siemiatkowski claims generative AI has enabled a workforce reduction, though the company is still hiring for essential roles.
2024.12.27
ARE LLMS CAPABLE OF NON-VERBAL REASONING?
Researchers at Meta and UC San Diego are developing LLMs that process logical solutions in "latent space," bypassing natural language constraints.
2024.12.27
QUICK TAKES ON THE RECENT OPENAI PUBLIC INCIDENT WRITE-UP
OpenAI's Kubernetes incident on December 11 was caused by unexpected interactions, where a new telemetry service overloaded the Kubernetes API servers, leading to failures in DNS-based service discovery.
2024.12.27
ELON MUSK'S XAI LANDS $6B IN NEW CASH TO FUEL AI AMBITIONS
xAI raised $6 billion in Series C, doubling its valuation to $45 billion. The company aims to accelerate infrastructure and product development, using models like Grok to enhance capabilities on X. xAI continues to expand amidst legal challenges with OpenAI, leveraging data from Musk's ventures and planning further fundraising next year.
2024.12.26
SRIRAM KRISHNAN NAMED TRUMP'S SENIOR POLICY ADVISOR FOR AI
Sriram Krishnan, former a16z general partner, will serve as senior policy advisor for AI at the White House Office of Science and Technology Policy under Trump. He will coordinate AI policy across the government and work closely with David Sacks. Krishnan's background includes leadership roles at Microsoft, Twitter, and Facebook, and he has a strong relationship with Elon Musk.
2024.12.26
OPENAI TRAINED O1 AND O3 TO 'THINK' ABOUT ITS SAFETY POLICY
OpenAI enhanced reasoning and safety features in its new AI model family, o3, using a novel "deliberative alignment" process. This method improves the alignment of AI model responses with OpenAI's safety values by referencing those values during inference without human-written data. Deliberative alignment, combined with advanced techniques like synthetic data and reinforcement learning, makes o3 OpenAI's safest model yet. It is set to release in 2025.
2024.12.26
UNDERSTANDING MEDICAL DECISION-MAKING
MedDec is a dataset that helps improve the extraction of medical decisions from clinical notes. It covers eleven different diseases. The dataset is annotated with ten types of medical decisions.
2024.12.26
LLMS WILL ALWAYS HALLUCINATE, AND WE NEED TO LIVE WITH THIS
As large language models become more ubiquitous across domains, it becomes important to examine their inherent limitations critically. Hallucinations in language models are not just occasional errors, but an inevitable feature of these systems.
2024.12.26
HIGH-SPEED ROBOTICS SIMULATOR
ManiSkill3 is an advanced, open-source robotics simulator designed for scalable learning and manipulation tasks.
2024.12.26
A TOOLKIT FOR SPEECH EMOTION RECOGNITION
EmoBox is a comprehensive toolkit designed to address common issues in Speech Emotion Recognition (SER). It provides a multilingual, multi-corpus benchmark for both intra-corpus and cross-corpus settings, enabling easier comparison and reproduction of SER models.
2024.12.26
CANCER DETECTION
SAM-Swin is a model for detecting laryngo-pharyngeal cancer (LPC) that uses advanced features from the Segment Anything Model 2 (SAM2).
2024.12.26
HOW TO GET REAL GPU UTILIZATION METRICS
Nvidia-smi shows a measure of GPU utilization but it is the amount of time where at least one kernel is running, not a full measure of GPU usage. This work by Stas shows how you can get actual FLOP usage.
2024.12.26
SO MANY TOKENS, SO LITTLE TIME: INTRODUCING A FASTER, MORE FLEXIBLE BYTE-PAIR TOKENIZER
GitHub has released a new open-source byte-pair tokenizer that improves speed and flexibility for large language models like Copilot, scaling better with linear complexity. This tokenizer addresses limitations of existing BPE algorithms by efficiently supporting dynamic token counts for real-time text operations. It outperforms popular libraries like tiktoken and Hugging Face in benchmarks, significantly boosting performance for various applications.
2024.12.26
SCALING LAWS – O1 PRO ARCHITECTURE, REASONING TRAINING INFRASTRUCTURE, ORION AND CLAUDE 3.5 OPUS “FAILURES”
There's skepticism around AI scaling laws, citing challenges like data exhaustion and hardware limits, but major AI players like Amazon, Meta, and OpenAI are continuing massive investments in data center and custom silicon capabilities, signaling their faith in ongoing scaling potential. New paradigms beyond pre-training, such as synthetic data, Reinforcement Learning, and advanced fine-tuning, are helping overcome traditional scaling barriers. OpenAI's o1 release demonstrates that increasing test-time compute, leveraging multi-datacenter training, and exploring novel scaling dimensions can significantly enhance AI model capabilities.
2024.12.26
HOW CLAUDE USES AI TO IDENTIFY NEW THREATS
Anthropic's Clio tool identified a coordinated effort to generate SEO spam using its chatbot, Claude, leading to the termination of the spammers' access. Clio uses machine learning to detect emerging threats and highlight unusual chatbot usage, assisting Anthropic's trust and safety team. The company encourages other AI labs to adopt similar monitoring strategies to address potential harms while exploring diverse user applications.
2024.12.26
GOOGLE'S NEW JULES AI AGENT WILL HELP DEVELOPERS FIX BUGGY CODE
Google Jules is an AI code agent designed to fix coding errors for Python and JavaScript in GitHub workflows.
2024.12.26
MICROSOFT RELEASES PHI-4 LANGUAGE MODEL TRAINED MAINLY ON SYNTHETIC DATA
Microsoft's new open-source language model, Phi-4, excels in solving math problems, outperforming even larger models like GPT-4o and Llama 3.3.
2024.12.26
CHATGPT'S NEW PROJECTS FEATURE CAN ORGANIZE YOUR AI CLUTTER
OpenAI's new Projects feature for ChatGPT enhances interaction organization by grouping related chats and files within a named Project.
2024.12.26
NHTSA FINALLY RELEASES NEW RULES FOR SELF-DRIVING CARS
The National Highway Traffic Safety Administration has introduced AV STEP, a framework aimed at facilitating the commercialization of autonomous vehicles while requiring companies to provide more data for public trust. The proposal seeks to ease exemption processes for driverless cars, potentially raising the cap on vehicles without traditional controls. However, its future is uncertain due to political and safety concerns.
2024.12.25
SAUDI ARABIA INVESTS IN ROBOTS TO HELP BUILD ITS NEOM DESERT MEGACITY
Saudi Arabia's Neom has invested in Europe's GMT Robotics to enhance construction efficiency in its ambitious projects like the Line megacity. The focus is on automation and robotics to streamline rebar production and reduce human labor, given Neom's usage of 20% of the world's steel supply.
2024.12.25