Search the archive

VOICECRAFT CLONING AND TTS

Zero shot voice cloning and speech generation capabilities in a powerful 700m parameter model.

INTERRUPTING COW

In natural conversation, people sometimes interrupt or talk over one another. This can be key to quickly coming to a consensus. This AI assistant predicts tokens while the person is talking and if it predicts enough in a row it will interrupt.

TESTING THE CODING ABILITIES OF LLMS WITH A NEW CODE BENCHMARK

EvoEval is a new benchmark suite that tests the coding abilities of Large Language Models more rigorously than ever before.

QWEN MOE

Qwen MoE is equivalent in performance to a strong 7B model with 1⁄3 of the activated parameters.

HOW AUTONOMOUS RACING IS PUSHING SELF-DRIVING CARS FORWARD

Autonomous racing is advancing AI and machine learning in high-stress conditions, with competitions like the Indy Autonomous Challenge accelerating innovation in vehicle safety. Researchers and students use platforms like F1tenth for algorithm development, pushing the limits of autonomous vehicle capabilities on actual racetracks. These high-speed challenges contribute to a better understanding of machine perception, decision-making, and control systems for real-world traffic applications.

“THE KING IS DEAD” — CLAUDE 3 SURPASSES GPT-4 ON CHATBOT ARENA FOR THE FIRST TIME

Anthropic's Claude 3 Opus has surpassed OpenAI's GPT-4 for the first time on Chatbot Arena. Chatbot Arena is a leaderboard run by the Large Model Systems Organization, a research organization dedicated to open models. Its site allows visitors to rate outputs from various models, enabling it to calculate the best models in aggregate. While Claude's rise is notable, GPT-4 is now over a year old.

DOES AI NEED A “BODY” TO BECOME TRULY INTELLIGENT?

The embodiment hypothesis argues true intelligence necessitates physical interaction, prompting advancements in AI through simulations and real-world testing, despite challenges like the "sim-to-real gap," leading to the cautious deployment of AI robots in industries.

MICROSOFT COPILOT AI WILL SOON RUN LOCALLY ON PCS

Microsoft's Copilot AI will soon run locally on PCs, requiring future AI PCs to have built-in neural processing units capable of over 40 TOPS.

AIRTABLE AI (PRODUCT)

Harness the power of AI and incorporate it directly into your workflows.

AI21 RELEASES JAMBA LANGUAGE MODEL

Mamba is a style of model that is supposed to beat Transformers in efficiency while matching performance. Jamba is a novel variant that includes MoE layers. It can run at 1.6k tokens per second with a context length of 128k tokens. It achieves 67% on the MMLU benchmark. The weights are available.

HUME INTRODUCES EMPATHIC VOICE INTERFACE

Hume has released the Empathic Voice Interface (EVI), the first conversational AI with emotional intelligence.

GOOGLE STARTS TESTING AI OVERVIEWS FROM SGE IN MAIN GOOGLE SEARCH INTERFACE

Google is piloting AI-generated overviews in search results for complex queries with a subset of U.S. users without requiring opt-in to its Search Generative Experience. This feedback-driven test aims to enhance user experience and may impact site traffic dynamics due to changes in ad placements and organic search visibility.

1 BIT, POST TRAINING, QUANTIZATION

1 Bit language models are exciting. This work shows how to quantize the Linear layers of a language model without sacrificing performance. This can result in a 70B model running on consumer GPUs.

LONG FORM FACTUALITY IN LARGE LANGUAGE MODELS

You can use language models to generate factual information. Google has released a dataset and benchmarks to show how well each model performs. This report gives insight into how to improve a model’s factuality and shows language models winning against human annotators in a majority of cases.

IMPROVING MODEL ADAPTATION

CoDA is a new approach to Unsupervised Domain Adaptation (UDA). It helps AI models better adapt to unlabeled, challenging environments by learning from differences at both the scene and image levels.

real-time data available for retrieval. Watch the demo

USING AI TO IDENTIFY SOURCES OF CONFUSION IN YOUR CONTENT

This approach pinpoints the exact content that triggers user questions online, from lectures to news articles. This method, known as backtracing, aims to help content creators refine their work by identifying and understanding the causes of confusion, curiosity, or emotional reactions.

HIGH-RESOLUTION LARGE LANGUAGE-VISION ASSISTANT

Researchers have developed a new method, Mixture-of-Resolution Adaptation (MRA), which improves AI's ability to recognize details in images.

3D MEDICAL IMAGING DATASET

CT-RATE is a dataset that marries 3D medical imaging with textual reports. CT-CLIP is a versatile AI framework optimized for these images.

EXPONENTIAL GROWTH AND AI ESCAPE VELOCITY

AI Futurist Ray Kurzweil invented the law of accelerating returns, which states that the rate of progress is increasing at an exponential rate over time. In a recent discussion, Kurzweil explored various topics including prospects that are only getting better, the direction in which the AI economy is headed, human relationships with AIs, longevity escape velocity, and much more.

PLENTIFUL, HIGH-PAYING JOBS IN THE AGE OF AI

AI experts are exploring automating human tasks, leading to concerns about job loss and wage declines. However, the view that AI advancement will inevitably result in human job obsolescence may not be correct. Humans may retain employment in an AI-dominated future due to constraints like computing power and opportunity costs, although this scenario is not guaranteed.

COFRAME LIVING IMAGES (PRODUCT LAUNCH)

Coframe’s generative A/B testing tool allows you to optimize images on your website in minutes with generative AI.

META IS ADDING AI TO ITS RAY-BAN SMART GLASSES

Meta is introducing AI features to its Ray-Ban smart glasses, including translation, and the identification of objects, animals, and monuments, with activation through voice commands.

GOOGLE BRINGING GEMINI NANO TO PIXEL 8

The Pixel 8 will receive Gemini Nano in a developer preview to enhance features like Summarize in Recorder and Gboard Smart Reply.

Access to Hosted Anyscale Private Preview

DBRX MOE

Databrix and Mosaic have trained a 132B parameter MoE model with impressive performance. They trained the model on 3,000 H100s and have released the weights. The model is also available on the Databricks API.

AMAZON INVESTS ANOTHER $2.75B IN ANTHROPIC

Amazon finalized a $4 billion investment in Anthropic, its largest venture investment yet.

HERE’S WHY AI SEARCH ENGINES REALLY CAN’T KILL GOOGLE

Emerging AI-driven search tools challenge Google by offering direct and explorative answers, but struggle to match its speed, diverse functionalities, and efficient data presentation, underscoring the complexity of replacing traditional search with AI.

BINARY SEARCH VECTORS

Searching over embedding vectors is a key to RAG pipelines. If you replace the fp32 numbers with a single 0 or 1, then use a KNN clusterer and reranker, you can maintain performance while shrinking memory requirements 30x.

DEEPFAKE TECHNOLOGY AND DETECTION METHODS

This comprehensive survey delves into the advancements and challenges of deepfake technology and its detection, highlighting the arms race between deepfake creators and those developing technologies to spot them.

AUDIO-DRIVEN ANIMATION

AniPortrait is a framework designed to create lifelike animated portraits from a single reference image and audio input. By translating audio into 3D representations and then mapping these onto 2D facial landmarks, this method produces animations that excel in natural facial expressions, varied poses, and high visual quality.

FinOps for data. Register for the 2024 virtual event (free!)

BENCHMARK LLMS BY PLAYING STREET FIGHTER

LLMs are useful in as much as they are fast, accurate, and follow directions. This combination makes a street fighter emulator with text input an excellent way to figure out which models are good at these three criteria.

IMAGE GENERATION WITH TEXT AND POSE CONDITIONS

AID and its variant PAID are two techniques designed to improve image interpolation by incorporating conditions like text and poses. These methods ensure the production of images with enhanced consistency, smoothness, and fidelity without requiring additional training.

BOOSTING EFFICIENCY IN MODELS WITHOUT EXTRA TRAINING

The OPTIN framework introduces a novel way to enhance the efficiency of transformer-based AI models across various domains without the need for re-training. By using a technique called intermediate feature distillation, OPTIN can compress networks under specific constraints while barely affecting accuracy.

INSIDE THE SHADOWY GLOBAL BATTLE TO CONTROL AI

The world is grappling with the challenge of regulating AI. A series of high-profile meetings and conferences involving global leaders, tech executives, and policymakers revealed divisions and a lack of consensus on how to control this transformative technology.

BUILD EVALUATION PIPELINES ON YOUR OWN DATA

New models are regularly released that claim to be the state of the art on standard benchmarks. It is important to measure these models on your own tasks and data. Superpipe is a tool that helps build these evaluation pipelines on your data.

HACKERS CAN READ PRIVATE AI-ASSISTANT CHATS EVEN THOUGH THEY’RE ENCRYPTED

Researchers discovered a side-channel attack that can decipher encrypted AI assistant chats with high accuracy on specific topics by exploiting token transmission within the encryption. The attack utilizes large language models to reconstruct token sequences into readable text, potentially exposing sensitive user conversations. Major AI assistants, except for Google Gemini, are vulnerable to this method, prompting providers to seek mitigation strategies.

development. Get Early Bird tickets (25% off)

OPENAI IS STARTING TO TEST GPT EARNING SHARING

OpenAI is partnering with a small group of US builders to test usage-based GPT earnings.

NVIDIA TOPS MLPERF’S INFERENCING TESTS

Nvidia’s GPUs, particularly the H200, led the MLPerf’s inferencing benchmarks.

CREATIE (PRODUCT)

A one-stop product design tool amplified by AI.

ELON MUSK SAYS ALL PREMIUM SUBSCRIBERS WILL GAIN ACCESS TO AI CHATBOT GROK THIS WEEK

Elon Musk announced that xAI’s Grok chatbot, previously exclusive to Premium+ subscribers, will soon be available to all Premium subscribers on X.

OPENAI’S CHATBOT STORE IS FILLING UP WITH SPAM

OpenAI's GPT Store has become a mix of innovative tools and questionable content, including potentially copyright-infringing GPTs and bots promoting academic dishonesty.

APPLE’S BIG WWDC 2024 ANNOUNCEMENT MAY BE AN AI APP STORE

Apple is planning to enhance its AI offerings at WWDC 2024, potentially introducing an AI App Store where consumers can access AI apps from various vendors.