Hire a world class AI team for 80% less

Trusted by leading startups and Fortune 500 companies.

Building an AI product is hard. Engineers who understand AI are expensive and hard to find. And there's no way of telling who's legit and who's not.

That's why companies around the world trust AE Studio. We help you craft and implement the optimal AI solution for your business with our team of world class AI experts from Harvard, Princeton, and Stanford.

Our development, design, and data science teams work closely with founders and executives to create custom software and AI solutions that get the job done for a fraction of the cost.

P.S. Wondering how OpenAI DevDay impacts your business? Let’s talk!

Today:

• Google's Notebook LM Personal Data Management

• Google Considers AI Search Fees

• OpenAI's New Fine-Tuning Tools and Custom Models Program

• Hugging Face's Significant Security Breach

• Cohere's Command R+ Surpasses GPT-4 Turbo

• Apple Eyes AI Robotics

Google's STUNNING Notebook LM | Personalized AI to Build Your "Second Brain" | Notebook LM Tutorial

In 2001, David Allen's book "Getting Things Done" proposed a method to manage life's overwhelming tasks. Despite various systems emerging, such as "Build a Second Brain" by Thiago Forte, none fully solved information overload.

Recently, products like OpenAI's Context Connector and Google's Notebook LM aim to streamline organization. Notebook LM, endorsed by Forte, promises AI assistance in managing personal data and documents. Despite skepticism due to past Google service discontinuations, Notebook LM impresses with its efficient retrieval of specific information from uploaded documents, even offering creative outputs like poems. While experimental, it shows promise in tackling information overload.

WATCH THE VIDEO ON YOUTUBE

Google considers charging for AI-powered search in big change to business model

Google is considering charging for AI-enhanced search, a major shift in its revenue model. This move follows the high costs of providing the service, with experts suggesting all major players may adopt subscription models. The proposed plan would offer the new search feature exclusively to premium subscription users. AI search, more costly to compute than traditional search, aims to recover these expenses. 

While training AI models is expensive, the majority of costs lie in running them. Competitors also offer subscription plans, with some providing advanced features at a monthly rate. 

FINANCIAL TIMES

OpenAI releases new AI fine-tuning tools: ‘vast majority of organizations will develop customized models’

OpenAI has unveiled significant upgrades to its fine-tuning API and custom models program, aiming for more tailored artificial intelligence solutions. The enhanced API grants developers greater control over model refinement, facilitating personalized models for specific industries or tasks. Notable features include epoch-based checkpoints and a new comparative UI for evaluation. 

Additionally, the expanded Custom Models Program offers assisted fine-tuning and fully custom-trained models, catering to organizations with specialized needs. OpenAI anticipates widespread adoption of customized AI models, empowering businesses to harness AI's full potential. 

VENTUREBEAT

Hugging Face Confronts a New Security Flaw; Creators Share Their AI Skepticism

A significant security flaw discovered in Hugging Face, a widely used repository of machine learning models, highlights cybersecurity concerns over AI. Cloud security firm Wiz found the flaw, allowing them to execute commands on Hugging Face's servers, accessing and modifying private AI models of organizations like Meta Platforms and Microsoft. 

This vulnerability could lead to data breaches or manipulation of AI applications. The incident underscores the importance of prioritizing basic cybersecurity measures in the AI landscape, eclipsing concerns about futuristic AI scenarios like Skynet.

THE INFORMATION

Cohere launches Command R+, a powerful enterprise LLM that beats GPT-4 Turbo

Cohere has launched Command R+, an advanced enterprise-scale large language model (LLM) surpassing GPT-4 Turbo in performance. Designed for real-world business applications, Command R+ boasts enhanced capabilities like multilingual support and advanced retrieval augmented generation (RAG). It outperforms competitors in key enterprise benchmarks like ToolTalk and Berkeley Function Calling. 

Moreover, Cohere prioritizes data privacy and security, remaining cloud-agnostic and offering deployment options on Microsoft Azure and on-premises. With a track record of success in production environments and impressive revenue growth, Cohere is poised to lead the enterprise AI market with Command R+ and its suite of AI solutions.

VENTUREBEAT

Apple reportedly exploring AI-powered home robots, including these two products

Apple is reportedly exploring the development of AI-powered home robots, including a mobile assistant and a tabletop device with a moving display. Led by John Giannandrea, Apple's senior VP of machine learning, the project is in early stages with no guarantee of release. This initiative aligns with Apple's recent focus on AI, evidenced by its research paper introducing Reference Resolution As Language Modeling (ReALM) to enhance Siri's conversational abilities. 

While Amazon and Samsung have already entered the home robot market, Apple's potential entry signals its ambitions in AI and robotics. The company's WWDC event in June may offer insights into its AI advancements.

ZDNET

🧠RESEARCH

Mixture-of-Depths: Dynamically allocating compute in transformer-based language models

"Mixture-of-Depths" is a technique for transformer-based language models to flexibly distribute computation resources across input sequences. By dynamically allocating FLOPs to specific positions, models optimize performance at different layers. This approach maintains a set compute budget while allowing fluidity in token selection, resulting in efficient compute allocation.

Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction

"Visual AutoRegressive modeling (VAR)" redefines image autoregressive learning by predicting next-scale or next-resolution instead of next-token, improving AR transformers' image generation beyond diffusion transformers. VAR boosts ImageNet 256x256 performance, enhancing Frechet inception distance (FID) from 18.65 to 1.80, inception score (IS) from 80.4 to 356.4, with 20x faster inference speed. Empirical evidence shows VAR outperforms Diffusion Transformer (DiT) in image quality, speed, efficiency, and scalability, with power-law scaling akin to LLMs and zero-shot generalization in tasks like in-painting and editing. 

Language Models as Compilers: Simulating Pseudocode Execution Improves Algorithmic Reasoning in Language Models

"Think-and-Execute" is a framework enhancing large language models' (LLMs) algorithmic reasoning by decomposing the reasoning process into Think and Execute steps. It utilizes pseudocode to express task-level logic shared across instances, improving reasoning performance across seven algorithmic tasks compared to instance-specific approaches. Pseudocode facilitates reasoning in LLMs compared to natural language instructions.

On the Scalability of Diffusion-based Text-to-Image Generation

The paper investigates scaling properties of diffusion-based Text-to-Image (T2I) models, exploring model and data size effects. Empirical studies range from 0.4B to 4B parameters on datasets up to 600M images, focusing on UNet and Transformer variants. Findings highlight the importance of cross attention in UNet designs and data quality in improving text-image alignment. Efficient UNet variant identified, with scaling functions predicting alignment performance.

InstantStyle: Free Lunch towards Style-Preserving in Text-to-Image Generation

The paper presents "InstantStyle," addressing challenges in style-consistent text-to-image generation. It tackles issues like underdetermined style concepts and style degradation. InstantStyle decouples style and content from reference images in the feature space and injects reference image features into style-specific blocks. This framework achieves superior stylization results without cumbersome weight tuning, balancing style intensity and text controllability. Code available on GitHub.

🛠️TOP TOOLS

SciPhi - open source platform that makes it easy for developers to build the best RAG system.

Claros - Find the best products to buy using LLMs and Reddit.

Jack AI - AI-powered marketing tool that enhances copywriting and improves audience engagement for marketers.

Faune - AI-powered platform that brings advanced models like Mistral and GPT-4 to your devices, prioritizing privacy.

Text Blaze - Eliminate repetitive typing and mistakes.

📲SOCIAL MEDIA

🗞️MORE NEWS

Elon Musk says he’s increasing salaries for Tesla engineers because Sam Altman’s OpenAI keeps trying to recruit them

Elon Musk boosts Tesla engineers' pay, citing OpenAI's talent poaching. Musk, in feud with OpenAI's Altman, faces criticism for moving staff between companies. Concerns rise over Musk's management methods amid stagnant Tesla growth. Musk aims to rebrand Tesla as an AI leader, deflecting attention from his own talent war. FORTUNE

SiMa.ai secures $70M funding to introduce a multimodal GenAI chip

SiMa.ai, a Silicon Valley startup, secures $70M funding to launch its GenAI chip, designed for embedded AI processing. With a focus on edge computing, it targets various sectors. The chip promises versatility, high performance, and energy efficiency, positioning SiMa.ai as a key player in the evolving AI landscape. TECHCRUNCH

Ravel announces Orchestrate GenAI simple interface software

Ravel, Inc. introduces Ravel Orchestrate GenAI, a user-friendly interface for managing and deploying generative AI. It now supports AMD GPUs, simplifying setup across organizations. The product aims to streamline AI integration for IT and DevOps professionals, offering control, automation, and compatibility with various genAI tools. VENTUREBEAT

DataStax acquires the startup behind low-code AI builder Langflow

DataStax, known for its Apache Cassandra NoSQL database, is now focused on building a comprehensive GenAI stack. With the acquisition of Logspace, the creator of Langflow, a low-code tool for building Retrieval-Augmented Generation (RAG) applications, DataStax aims to enhance its offering. This acquisition complements DataStax's existing stack, enabling developers to accelerate the development of generative AI applications. Langflow will operate independently under DataStax, continuing its mission to democratize and streamline generative AI development. TECHCRUNCH

Nvidia, Indosat plan $200 mln AI centre investment in Indonesia, government says

Nvidia and Indonesia's telecom company PT Indosat Ooredoo Hutchison have announced plans to establish a $200 million artificial intelligence center in Central Java in 2024. The center, expected to be located in Surakarta city, could comprise telecommunication infrastructure or a human resource center. The choice of Surakarta is attributed to its readiness, quality human resources, and existing 5G infrastructure. Indosat's CEO shared the investment plan with Surakarta's mayor Gibran Rakabuming Raka, who is also Indonesia's vice president-elect and the son of President Joko Widodo.  REUTERS

The world’s top tech companies are launching a consortium to address AI’s impact on the workforce

Tech giants like Cisco, Google, and Microsoft are teaming up to tackle the impact of AI on jobs. With billions of lives affected, they form a consortium to recommend skilling opportunities, realizing they can't address the challenge alone. The focus is on evaluating AI's effect on various job roles and providing actionable insights for workers. FORTUNE