🤖 Practical AI - Innovations and Concerns in AI
Hugging Face introduces Chat Assistants, Apple making progress with on-device LLMs and the issue of identity verification in the digital age.
Hi! I’m Mike and welcome to this free edition of the 'Practical AI’ newsletter. Each week, I’ll cover what’s happened, what’s happening and how that applies to you.
This week's newsletter dives into recent developments in the AI sector, including the introduction of open-source chat assistants by Hugging Face, Apple's strides towards on-device Large Language Models (LLMs), and the pressing issue of identity verification in the digital age.
Google prepares to launch Gemini Advanced
Google is preparing to launch "Gemini Advanced" on February 7, setting the stage for a head-to-head competition with OpenAI's GPT-4. This enhanced version of Google’s Bard chatbot, powered by the Gemini Ultra 1.0 model, is poised to redefine expectations for coding, reasoning, and collaborative capabilities in AI-powered platforms. Unlike its predecessors, Gemini Advanced will be a paid service.
This move by Google underscores the intensifying competition in the AI space, where major players are rapidly evolving their offerings to meet the growing demand for sophisticated, AI-driven solutions. Gemini Advanced’s promise of enhanced capabilities marks a pivotal moment in AI development, highlighting the ongoing race to develop AI models that not only match but exceed human-like proficiency in complex tasks.
Hugging Face Introduces Chat Assistants
Hugging Face has launched Chat Assistants, offering an open-source alternative to OpenAI's GPT Store. This initiative allows users to create custom chat interfaces with a variety of open-source LLMs, including Llama2 and Mistral, without any subscription fees. Despite lacking features like Retrieval-augmented generation and web search, these are planned for future updates. The platform has seen rapid adoption, with over 4,000 assistants created shortly after launch. Hugging Face also plans to introduce additional open-source stores, expanding user options further​​.
Is Apple Preparing for LLMs?
Rumours suggest Apple is deepening its commitment to privacy and efficiency with a pioneering approach to running LLMs directly on devices. A recent paper, ‘LLM a flash: Efficient Large Language Model Inference with Limited Memory’, showcases a technological breakthrough that might enable such capability, signalling a major shift towards integrating AI into consumer technology with a strong emphasis on local processing and privacy.
In a separate but equally groundbreaking initiative, Apple is working on enhancing Siri's capabilities through the development of an on-device LLM. This ambitious project aims to significantly improve Siri, enabling the voice assistant to understand more complex queries and offer a more personalised user experience. While Apple utilises an advanced internal chatbot informally referred to as "AppleGPT" for internal purposes, this project is distinct from the efforts to augment Siri. The term "AppleGPT" is unofficial and has been used to describe Apple's internal explorations in AI, which remain largely under wraps.
The internal use of AppleGPT and the initiative to empower Siri with on-device LLM technology are part of Apple's broader strategy to innovate in AI while prioritising user privacy. By leveraging LLMs directly on devices, Apple aims to set new standards for privacy-conscious, efficient AI integration in consumer technology source.
As revealed by Tim Cook, we can expect more on this topic over the year. Read more
Navigating the AI-Generated Fake ID Challenge
A recent investigation into a site producing AI-generated fake IDs, including driver's licenses and a British Passport, has highlighted the issue of digital identity security. With these counterfeit IDs successfully bypassing established verification services such as Jumio, the incident has not only caused concern amongst some lawmakers but also emphasised the critical need for fortified identity protections in our increasingly digital world.
The role of AI in crafting sophisticated fake IDs that can fool verification processes has underscored the dual-edged nature of technological advancements. While AI offers immense potential for innovation and efficiency, it also presents new vulnerabilities that can be exploited for fraudulent activities. The cryptocurrency platform OKX, utilising Jumio's verification services, has acknowledged the severity of this challenge, stating, "It is apparent that the abusive use of AI to conduct fraudulent activity is an evolving and industry-wide challenge which OKX is comprehensively addressing."
This situation highlights whether our current identity verification processes and legislative frameworks are equipped to contend with the realities of today's technology landscape. The widespread reliance on digital document scans for identity verification across various sectors, especially in financial services, is a testament to the urgency of addressing these vulnerabilities.
As we move forward, the incident acts as a catalyst, urging a reevaluation of both our technological approaches to identity verification and the legal measures that underpin them. The need for secure, authenticated IDs has never been more critical. This scenario signals a pivotal moment for stakeholders across industries to collaborate in fortifying defences against AI-enabled fraud, ensuring the integrity of digital identities in an era of sophisticated technological capabilities.
Our view? Things are going to get worse before they get better and we’ll see an increase in fraud linked to AI generated identity documents.
AI News Roundup
Europe within reach of landmark AI rules after nod from EU countries: Europe is on the verge of adopting groundbreaking AI regulations after EU countries, including France, endorsed a political agreement, setting a global precedent for the governance of AI technology across various industries
Mistral's Miqu Model Makes Waves: A leak regarding Mistral's new model, Miqu, suggests it boasts performance levels near that of GPT-4, creating a buzz in the AI community.
Google's MobileDiffusion: Google announced MobileDiffusion, a text-to-image model optimised for mobile devices, capable of generating high-quality images in half a second, highlighting the push towards more efficient and accessible AI-powered applications.
Twin Labs' Raises $3 million for AI-driven Automation: Twin Labs, a Paris-based startup, is leveraging GPT-4V to automate repetitive tasks, the startup raised $3 million in pre-seed funding from investors like Betaworks and plans to offer pre-trained tasks before opening its platform for custom tasks.
AI Chatbots choose violence: In wargame simulations involving the US military, AI chatbots frequently opt for aggressive strategies, including nuclear strikes, raising concerns about their integration into military planning and the potential for escalating conflicts​
If you found yourself here, thank you. Please remember to subscribe for more practical information about AI and if you have any questions, feel free to reach out via X or LinkedIn, or simply just reply to this email.
Thank you!