🍏 Apple's New Pocket-Sized AI

Hey TEDchats Fam!

Happy Friday! Imagine walking into a room and your phone automatically knows to switch on the light, adjust the temperature, and even play your favorite song — all before you even say a word. Welcome to a future where your smartphone isn’t just smart; it's practically clairvoyant. Well, we’re not there yet but we’re certainly closer thanks to Apple's latest leap in AI technology!

If you're enjoying TEDchats, could you do us a quick favor? Please reply to this email with a 'Happy Friday' to help our email metrics 🙏

👀 SNEAK PEEK

🚨FEATURE STORY

Apple's ‘Efficient Language Model’ May End Up in Our Pockets

What if the power of AI was not confined to the cloud, but lived right in your pocket? Apple's latest move hints at a future where your iPhone might carry a powerful AI, without the need of a server!

A Giant Challenging THE GIANTS

Traditionally, the field of generative AI has been dominated by behemoths like Google, Microsoft, and Meta, which rely on massive AI models that demand substantial computing power, typically via cloud servers. Apple, on the other hand, has taken a different approach. The launch of OpenELM, a set of four ultra-compact AI models, marks a pivotal shift in Apple’s strategy, potentially bringing AI directly to the devices we use every day.

Our Future of Pocket-Sized Supercomputers

While the tech world has largely invested in more powerful yet bulkier AI models capable of handling complex tasks like advanced reasoning and data analysis, these Large Language Models (LLMs) come with a cost. This cost includes increased demand for bandwidth, the need for continuous internet connectivity, and the requirement for private data to be stored off-premises. This poses a significant challenge for users who don't have the resources, are in low-bandwidth environments, or are concerned with privacy and data security.

The release of OpenELM represents a significant turn in the narrative. These models, ranging from 270 million to 3 billion parameters, are not just smaller and cheaper to operate, but are designed to be efficient and powerful enough to operate directly on iPhones and iPads. This move could democratize AI, making it accessible wherever a user goes, irrespective of internet connectivity.

In The Meantime…

While the transition from cloud to pocket won't happen overnight, the potential of these Small Language Models (SLMs) is underscored by Microsoft's recent release of Phi-3, a 3.8B language model, and Google's latest release of Gemma 2B. In the meantime, Apple introduces another dimension to its OpenELM — making it openly accessible.

In contrast to past practices, Apple is offering a complete framework for training and evaluating the OpenELM models on publicly available datasets, along with training logs, checkpoints, and pre-training configurations. Additionally, they will provide code to convert its models to the MLX library for use on Apple devices. According to the company, this initiative aims to "empower and strengthen the open research community" (see OpenELM on Hugging Face and GitHub).

🧑‍💻 DAILY PROMPT

🍕 QUICK BITES

Synthesia's New Expressive AI Avatars

In a notable advancement in generative AI, Synthesia introduces its latest version of AI video avatars, now equipped to display nuanced emotional expressions and improved lip tracking. This update aims to provide more natural and human-like interactions in video content, specifically tailored for business applications like training and marketing. These avatars are designed to mimic human speech and reactions more closely, making enterprise videos more engaging and relatable (see Synthesia’s announcement).

Generative AI's Slow Steady Path to Profitability at Meta

Alongside positive profitability insights for the first quarter of 2024, CEO Mark Zuckerberg has laid the groundwork for a long-term investment strategy in generative AI for Meta, emphasizing the extensive and expensive process of turning these technologies into profitable ventures for investors. Despite integrating AI features across major platforms like Instagram, Facebook, and WhatsApp, Zuckerberg highlights the substantial investments and time required before these innovations begin yielding financial returns (get the full details here).

Under the Microscope: AI Mergers or Masquerades?

The UK's Competition and Markets Authority (CMA) has initiated inquiries into whether recent collaborations and talent acquisitions between tech behemoths like Microsoft and Amazon and various AI startups, such as Mistral AI and Anthropic, might actually be circumventing traditional merger regulations. These partnerships are under scrutiny to determine if they unfairly restrict competition within the UK's rapidly growing AI market (see the CMA’s official announcement).

🧰 AI TOOLBOX 

🤯 MIND SPARK

FlexAI Launches with $30M Seed Funding: Emerging from stealth, the Paris-based startup FlexAI is set to revolutionize AI compute with less complexity for developers, backed by a whopping $30 million seed investment (see more).

Augment Debuts with $252M Backing: The AI-powered coding platform Augment, which empowers software teams with AI, has emerged from stealth with an impressive $252 million in funds and a valuation of $977 million (see more).

Parloa Raises $66M for Its Powerful Services: German AI startup Parloa has secured $66 million in funding to enhance and expand its AI-driven customer service platform, bringing its total capital raised to $98 million. Notably, the startup has already signed contracts with several Fortune 200 companies in the U.S. within just six months (see more).

🎨 AI ART