Today:

• Jesse Lyu Demos New Rabbit R1

• Nvidia Acquires AI Leader Run:ai 

• Apple Debuts OpenELM

• Perplexity Eyes $250M Raise, $3B Valuation

• Snowflake Unveils Arctic

• UK Probes Microsoft, Amazon AI Deals

• Tesla's Optimus Robot Set for Useful Tasks

Rabbit R1 - Hands On Live Unboxing Demo by Jesse Lyu [SUPERCUT]

The Rabbit R1 reveal showcased a plethora of features, particularly highlighting its impressive UI, advanced settings accessible through a simple shake, and its unique color. The gadget supports both 4G LTE and Wi-Fi, boasting settings that enhance user interaction with tactile feedback. Demonstrations included real-time weather updates, internet connectivity, and even a brief dive into language translation without the need to alternate turns, enhancing fluid conversation.

The R1's robust capabilities extend to AI-assisted tasks like note-taking and voice recording, with the future promise of a community-requested keyboard feature and ongoing updates to enhance functionality and user experience. 

WATCH THE VIDEO ON YOUTUBE

Nvidia Agrees to Acquire Israeli AI Software Provider Run:ai

Nvidia Corp. has agreed to acquire Israeli startup Run:ai, known for AI computing software. The Santa Clara-based chipmaker didn’t disclose the deal's terms, but sources estimate it at $700 million. Run:ai, a close Nvidia collaborator since 2020, specializes in managing AI computing resources. 

This move strengthens Nvidia's foothold in AI technology, aligning with its strategic vision. The acquisition underscores Nvidia's commitment to advancing AI capabilities, a pivotal aspect of modern technology. This deal reflects Nvidia's proactive stance in evolving tech landscapes, positioning itself as a dominant force in AI innovation.

BLOOMBERG

Apple releases OpenELM: small, open source AI models designed to run on-device

Apple introduces OpenELM, a series of open source AI models aimed for on-device use, competing with Google, Samsung, and Microsoft. These models, released on Hugging Face, vary in size from 270 million to 3 billion parameters, balancing efficiency and capability. Unlike traditional pre-training, OpenELM focuses on instruction tuning for specific user requests. 

Despite Apple's secretive nature, these models target on-device applications and aim to empower the research community. Performance-wise, they are respectable but not cutting-edge, with potential for improvement. Overall, OpenELM signifies Apple's foray into open source AI, challenging industry norms.

VENTUREBEAT

Perplexity is raising $250M+ at a $2.5B-$3B valuation for its AI search platform, sources say

Perplexity, an AI search startup, is raising over $250 million, valuing it between $2.5 to $3 billion. The company, founded in 2022, has already seen rapid valuation growth, with recent rounds totaling $137 million. This new fundraising aims to capitalize on market interest. Perplexity's product, a generative AI search engine, sets it apart, with plans to expand into enterprise solutions. Its unique approach targets a specific niche, akin to Snap's pre-IPO fundraising strategy. 

Despite competition, Perplexity's focus on search distinguishes it in the AI landscape. The fundraising frenzy reflects the growing importance of AI-driven solutions in tech.

TECHCRUNCH

Snowflake launches Arctic, an open ‘mixture-of-experts’ LLM to take on DBRX, Llama 3

Snowflake introduces Arctic, an open "mixture-of-experts" large language model (LLM) tailored for enterprise tasks like SQL and code generation. Using a unique architecture, Arctic achieves high performance while remaining efficient. It competes with models from Databricks, Meta, and Mistral, targeting enterprise AI applications like chatbots and data assistants. Arctic's Dense MoE hybrid architecture optimizes parameter usage, resulting in targeted performance with minimal compute consumption. 

Snowflake emphasizes Arctic's efficiency, achieved with a modest training budget compared to competitors. The model is available under the Apache 2.0 license, fostering open use and development across various platforms. Snowflake also provides resources like data recipes and research cookbooks to facilitate model customization and understanding.

VENTUREBEAT

UK regulator examines Microsoft and Amazon’s AI dealmaking

British regulators are examining Microsoft and Amazon’s partnerships with small AI companies. The Competition and Markets Authority is soliciting opinions on Microsoft's deal with French firm Mistral and Amazon's investment in U.S. startup Anthropic. They're also scrutinizing Microsoft's hiring of former Inflection AI staff. 

This inquiry precedes a formal review by British regulators. Microsoft invested €15 million in Mistral, while Amazon put $4 billion into Anthropic. Both deny these actions constitute mergers. This scrutiny reflects a tougher stance by British regulators on major U.S. tech companies, following delays in Microsoft's acquisition of Activision Blizzard.

FINANCIAL TIMES

Tesla’s Optimus robot will start performing “useful tasks” by the end of this year.

Elon Musk announced during Tesla's earnings call that the Optimus robot will begin undertaking "useful tasks" by year-end, with potential external sales by next year. However, he cautioned that these timelines are speculative.

THE VERGE

🧠RESEARCH

OpenELM: An Efficient Language Model Family with Open-source Training and Inference Framework

OpenELM is a new open-source language model that's more transparent and easy to replicate. It uses a smart method to manage its parts for better accuracy, outperforming older models while using fewer resources. It's fully available online with all you need to train it yourself.

Multi-Head Mixture-of-Experts

Multi-Head Mixture-of-Experts (MH-MoE) is a model that improves on Sparse Mixtures of Experts by splitting tokens into sub-tokens handled by multiple experts. This enhances understanding and activation of experts, reducing overfitting and improving performance across various language tasks. MH-MoE is also easy to integrate with existing models.

Pegasus-v1 Technical Report

Pegasus-1 is a multimodal language model tailored for video content. It tackles the challenge of understanding and interacting with video data by interpreting time and space details. The report details its architecture, training methods, and how it performs in tasks like video conversation, question answering, and summarization, offering a well-rounded view of its strengths and areas for improvement.

SnapKV: LLM Knows What You are Looking for Before Generation

SnapKV is a new method enhancing Large Language Models (LLMs) by efficiently managing the Key-Value (KV) cache size without sacrificing performance. SnapKV focuses each attention head on crucial parts of input prompts, significantly speeding up processing and reducing memory use, especially with long inputs. It performs comparably to traditional models on various datasets and can handle extremely long texts with minimal accuracy loss.

Transformers Can Represent n-gram Language Models

The paper explores how transformer models can embody n-gram language models (LMs), historically simple yet essential types of LMs. It shows that transformers, particularly those using specific types of attention mechanisms, can perfectly replicate any n-gram LM, establishing a clear baseline for their capability to model probability distributions over text strings. This research marks a significant step in understanding the probabilistic capabilities of transformer architectures.

🛠️TOP TOOLS

Wondera - Sing One Song to Create Your AI Voice

Meals.chat - Offers a Telegram bot for diet tracking by analyzing pictures or descriptions of your meals.

Dart - AI project management tool

Verb - A new space for writers

Beehiiv - Newsletter platform designed to boost growth, offering intuitive tools for creating, publishing, and monetizing content.

📲SOCIAL MEDIA

🗞️MORE NEWS

HubSpot unveils Zendesk-like updates to its Service Hub and other AI tools for SMBs

HubSpot boosts its platform with AI features to aid small businesses in competing with larger rivals. Updates to its Service Hub focus on managing customer interactions and driving retention. The Content Hub, now powered by AI, offers tools for generating diverse content. HubSpot aims to empower SMBs with cost-effective solutions. VENTUREBEAT

Coca-Cola and Microsoft Announce $1.1 Billion USD AI and Cloud Partnership

Coca-Cola and Microsoft join forces in a $1.1 billion USD deal to propel cloud and AI technologies across Coca-Cola's operations. The partnership aims to enhance business functions like marketing and manufacturing while boosting workplace productivity with tools like Microsoft 365 Copilot. This strategic move reflects Coca-Cola's digital-first approach to business. HYPEBEAST

SK Hynix Plans $15 Billion Chip Expansion to Meet AI Demand

SK Hynix Inc. is investing around $14.6 billion to construct a new memory chip complex in South Korea, aiming to meet the rising demand for semiconductors used in AI development. SK Hynix leads in high-bandwidth memory chips for AI, competing with Samsung Electronics Co. in this sector. The company also plans expansions in the US and other domestic investments in South Korea. YAHOO!

Microsoft is blocking employee access to Perplexity AI, one of its largest Azure OpenAI customers

Microsoft is blocking its employees from accessing Perplexity AI, a major user of its Azure OpenAI service. Perplexity's AI chatbot search engine relies on Azure OpenAI. Microsoft's move extends to blocking other AI tools like Google's Gemini chatbot. The extent of the restriction across Microsoft remains unclear. BUSINESS INSIDER

Nooks raises $22M to empower sales reps with its AI-powered call platform

Nooks, an AI startup, secures $22M in Series A funding to enhance sales rep productivity. Using AI, it automates tasks like call analysis, providing insights to boost sales performance. Originally focused on remote work collaboration, Nooks pivoted to sales feedback loops, now serving 200+ clients. It aims to revolutionize sales processes. VENTUREBEAT

Parloa, a conversational AI platform for customer service, raises $66M

Parloa, a German conversational AI platform for customer service, secures $66 million in Series B funding, led by Altimeter Capital. With a focus on the U.S. market, it aims to enhance automation using advanced AI models, particularly emphasizing voice interactions. Parloa has already gained traction with Fortune 200 companies like Swiss Life and Decathlon. TECHCRUNCH

Google-Backed AI Firm Is Hong Kong’s Latest Stock Debut Flop

Mobvoi Inc., a Chinese AI developer supported by Google, saw its Hong Kong stock debut fall 3.2%, the city's third recent flop. Despite an IPO priced near the lower end, it raised HK$321 million ($41 million), falling short of previous targets. This contrasts with regional market gains. BLOOMBERG

China’s SenseTime Surges 36% After Unveiling a New AI Model

SenseTime Group Inc.'s stock surged over 36% following the unveiling of its latest SenseNova 5.0 generative AI model. The platform, likened to ChatGPT, showcases significant enhancements in linguistic and creative capabilities. The sharp rise underscores China's ambitious endeavors in advancing artificial intelligence technology. BLOOMBERG

Smartwatch AI predicts atrial fibrillation 30 minutes before it arrives

An AI model trained on heart rate data can predict episodes of atrial fibrillation (AF) up to 30 minutes in advance, offering potential for early intervention and improved patient outcomes. The model, called WARN (Warning of Atrial fibRillatioN), utilizes deep learning and R-R interval data from electrocardiography (ECG) to forecast AF onset. It achieved an accuracy of 83% and 73% on test and validation sets, respectively, with an average lead time of 31 to 33 minutes. The researchers envision integrating this technology into smartphones for real-time monitoring via smartwatches, enabling personalized early warning systems for individuals at risk of AF. NEW ATLAS

Move over, Black Mirror; the BBC is showing AI can be a force for good in media

The BBC is demonstrating AI's positive potential in media, diverging from dystopian portrayals. They're exploring AI applications like converting audio to text for live text pages, enhancing audience value and efficiency. While embracing experimentation, they prioritize transparency and responsibility, aiming to lead best practices in broadcasting. FORTUNE