Your browser does not support JavaScript!

The Current State and Key Developments in AI Technology and Applications

GOOVER DAILY REPORT July 9, 2024
goover

TABLE OF CONTENTS

  1. Summary
  2. Advancements and Delays in AI Technologies
  3. Startup Competition and Market Dynamics
  4. Security Breaches and Risks
  5. Economic and Market Impact
  6. Integrations and Regulations
  7. Advancements in Google AI
  8. Apple's AI Innovations
  9. Emerging AI Tools and Techniques
  10. Conclusion

1. Summary

  • The report titled 'The Current State and Key Developments in AI Technology and Applications' offers a detailed overview of the present landscape of AI technology. It delves into various aspects including advancements and delays in AI developments such as OpenAI's ChatGPT Voice Mode, Meta AI's Llama 3, and Amazon's Metis AI. Further, the report examines market dynamics, highlighting the competition between startups like Perplexity and giants like Google, and the financial and technical hurdles these companies face. Security breaches and vulnerabilities, especially in OpenAI and Microsoft's AI models, underscore the importance of cybersecurity in AI development. Additionally, the economic impacts are discussed, particularly in the fluctuation of AI crypto tokens and the integration of AI in cryptocurrency mining. The document also covers the integration of AI in platforms like Apple's Siri and Google's search functions under Gemini, alongside regulatory changes affecting cryptocurrency reporting. Lastly, it explores emerging AI tools and techniques relevant to enterprises, emphasizing the challenges of integrating generative AI technologies and advancements in AI model training techniques such as Google's JEST.

2. Advancements and Delays in AI Technologies

  • 2-1. Delays in ChatGPT Voice Mode

  • OpenAI has announced a delay in the rollout of the new advanced Voice Mode for ChatGPT, which was initially planned for June. The postponement is due to the need for further improvements in content detection and user experience. The full launch for all ChatGPT Plus customers might be pushed to fall, subject to safety and reliability checks. Meanwhile, new video and screen-sharing capabilities, unrelated to Voice Mode, have been rolled out for ChatGPT on various platforms. Additionally, OpenAI has started offering the ChatGPT app for macOS, which can be activated using a keyboard shortcut (Option + Space) and supports text, speech, and video.

  • 2-2. Meta AI's Offerings

  • Meta AI, powered by the Llama 3 model, is now available as a standalone website. This service is free to use but requires a Facebook account for access. The capabilities of Meta AI include generating text and images, creating animated clips from AI-generated images, drafting emails and essays, and answering various questions by providing up-to-date information from the web. Unlike other AI services, Meta AI offers all features for free without any limitations.

  • 2-3. Amazon's Metis AI Service

  • Amazon is working on a new AI service named Metis, designed to compete with ChatGPT. Metis utilizes retrieval-augmented generation to provide up-to-date information. The intended capabilities of Metis include delivering text and image responses, suggesting follow-up queries, turning on lights, and booking flights. This development indicates Amazon's efforts to join the competitive AI market and catch up with other established players.

3. Startup Competition and Market Dynamics

  • 3-1. Perplexity vs Google

  • Google, for years, has been a dominant force in the search engine market, backed by its mission to 'organize the world’s information and make it universally accessible and useful.' However, the advent of AI technology has introduced new challengers, one of which is Perplexity. Perplexity, a startup described as an 'answer engine,' has rapidly risen in prominence with its innovative approach to search queries. Unlike traditional search engines, Perplexity combines features from Google and ChatGPT, providing immediate, single-answer responses derived from large language models, while still linking back to the source sites. This user-friendly interface has attracted millions of users and high-profile tech endorsements, pushing Perplexity's valuation past $1 billion with $20 million in annual recurring revenue (ARR) achieved by fewer than 50 employees. Despite its small size, Perplexity's fast and accurate results, compared to giants like Google and ChatGPT, are attributed to hardcore engineering and strategic use of multiple AI APIs.

  • 3-2. Financial and Technical Challenges

  • Perplexity faces significant financial and technical hurdles as it aims to sustain its position against major competitors. The startup relies heavily on APIs from industry giants like OpenAI and Anthropic, posing a critical dependency risk. If these companies decide to revoke API access or introduce competing products, Perplexity’s core capabilities could be severely undermined. Additionally, the high cost of AI search queries, which are approximately ten times more expensive than traditional ones, creates considerable financial strain. Despite raising an impressive $165 million, Perplexity’s resources are dwarfed by the billions available to its competitors. In response to these challenges, Perplexity is developing its own search index and large language models based on open-source technologies like Meta’s LLaMa 3. On the monetization front, Perplexity has introduced a subscription model and plans to incorporate advertisements, though it remains to be seen how effective these will be in a single-answer search environment compared to Google’s established advertising system. The strategic moves to build self-reliance and diversify revenue streams are critical as Perplexity navigates the complexities of competing with entrenched industry leaders.

4. Security Breaches and Risks

  • 4-1. OpenAI Security Breach

  • In early 2023, a hacker infiltrated OpenAI's internal messaging systems, obtaining sensitive information regarding the design processes of the company's AI products. However, it was confirmed that the main AI repository systems and customer data were not compromised. Employees expressed concerns about potential national security risks, fearing that rivals such as China might exploit this breach. OpenAI's internal communications about this event were only disclosed to the company's board and employees, and the incident was not reported to federal law enforcement agencies. OpenAI has since heightened its security measures, forming a Safety and Security Committee which includes members such as former NSA head Paul Nakasone.

  • 4-2. Employee Concerns and Company Response

  • Employees of OpenAI have voiced significant concerns post-breach, particularly regarding the potential for geopolitical rivals, like China, enhancing their own AI models using the stolen information. Leopold Aschenbrenner, a technical manager, warned of the company's inadequacies in preventing foreign actors from breaching the systems. This led to his termination, which he claims was politically motivated. The company confirmed these internal security issues in May 2023 and has since faced criticism for not responding adequately to safeguard against such breaches in the first place.

  • 4-3. Microsoft's Skeleton Key Jailbreak

  • Microsoft unveiled a vulnerability known as the Skeleton Key Jailbreak, which can disable model alignment and prompt AI models to produce harmful and dangerous outputs. This exploit affects various models from OpenAI, Google, Meta, and others, allowing users to direct the AI to generate results involving violent, unethical, or illegal content. Microsoft's testing between April and May 2024 confirmed the exploit’s impact on major models, excluding some resistance shown by GPT-4. To mitigate these risks, Microsoft has provided specific guardrails to protect against such exploits. The high vulnerability of AI models has sparked significant concerns among tech leaders, such as Elon Musk, about the potential dangers surpassing even those posed by nuclear weapons.

5. Economic and Market Impact

  • 5-1. Fluctuations in AI Crypto Tokens

  • AI crypto tokens have emerged as a trending topic within the cryptocurrency ecosystem, capturing the attention of enthusiasts and investors alike. These tokens experienced a significant uptrend in the first few months of 2024, reaching a peak in February with soaring trading volumes and market valuations across various exchanges. According to data from Bitget, the AI token zone saw a 400% increase in trading volume throughout February, achieving a total market capitalization exceeding $39 billion. This surge was driven by investor interest in AI-focused projects such as Worldcoin (WLD), Livepeer (LPT), and Arkham (ARKM), whose prices skyrocketed in recent months. However, over the past week, tokens like SingularityNET (AGIX), Fetch AI (FET), and Ocean Protocol (OCEAN) experienced declines of over 20%, with FET down 25.8%, AGIX by 24.0%, and OCEAN by 23.5%. Furthermore, according to CoinGecko, the market capitalization of AI crypto tokens plummeted by 6% in the past 24 hours, reaching $33.286 billion as of Tuesday, and dropped further to $31.6 billion, a 7.8% change. The decline occurred despite the consolidation of AI-themed cryptocurrencies FET, AGIX, and OCEAN into a new ASI token being scheduled for 15 July, aiming to establish ASI as a leading open-source, decentralized network. The delay in the planned token merger, initially scheduled for 13 June, was attributed to technical constraints and regulatory considerations. Despite these fluctuations, the commitment and substantial progress toward creating a decentralized superintelligence network remain strong.

  • 5-2. Integration of AI in Cryptocurrency Mining

  • Bitcoin miners are finding new opportunities by integrating AI into their operations. With access to substantial energy resources, many mining firms are now offering data-center services to AI companies. The demand for electricity in AI applications, such as ChatGPT queries requiring 10 times the power of a Google search, provides a lucrative market for these miners. A JPMorgan report indicates that 14 publicly traded U.S. miners have about 3.6 gigawatts of spare power out of the 5 gigawatts they control, with an additional 4.5 gigawatts in planned energy contracts. This shift towards providing high-performance computing (HPC) services could see significant investments from AI firms, which are less constrained by the volatile price and mining difficulty of Bitcoin. Companies like Core Scientific (CORZ), Iris Energy (IREN), and Applied Digital (APLD) are already moving towards HPC, while others continue their focus on pure-play mining. This transition is expected to cater to the immediate energy demands of AI companies, contrasting the 3-5 years it typically takes to build dedicated HPC data centers.

6. Integrations and Regulations

  • 6-1. OpenAI's Products in China

  • According to a report from The Information, despite OpenAI's attempts to restrict access to its platform in China, users can still access OpenAI products through Microsoft's Azure cloud computing platform. This continues OpenAI's partnership with Microsoft, which has provided Microsoft unrivaled access to its products. The partnership aims to expand enterprise computing using AI features. Notably, OpenAI announced it would restrict its API access to Chinese customers starting July 9th, but Microsoft's Azure OpenAI service allows these customers to use OpenAI's conversational AI models. This service is available to enterprise customers in China using Azure's cloud platform. The report confirmed this by speaking to three Chinese businesses that confirmed their access to OpenAI models through Azure. Despite these restrictions, OpenAI's chatbot service, ChatGPT, remains banned in China, with Chinese authorities controlling access to information. Chinese companies face challenges in developing local AI models due to extensive U.S. chip restrictions, specifically targeting AI chips necessary for training models like ChatGPT.

  • 6-2. IRS Reporting Requirements for Cryptocurrency Brokers

  • The U.S. Treasury has implemented new reporting requirements for cryptocurrency brokers to curb tax evasion, beginning in 2026 for transactions executed the previous year. Cryptocurrency brokers must disclose details about customer trades to ensure digital assets are not used to hide taxable income. This new requirement aims to simplify tax reporting for crypto investors, placing it on par with traditional securities. The IRS has introduced a new form, 1099-DA, to facilitate this reporting. Although decentralized exchanges and certain transaction types, such as wrapping and unwrapping transactions, liquidity provider transactions, staking transactions, short sales of digital assets, and notional principal contracts (swaps), are exempt from reporting for now, future guidelines may include them. These exemptions were detailed in Notice 2024-57. This move is part of a broader effort to increase transparency and accountability in the cryptocurrency market.

7. Advancements in Google AI

  • 7-1. AI Overviews in Google Search

  • Google Search is experiencing a significant shift with the increasing implementation of AI Overviews, primarily driven by Google's Gemini project. Officially launched in May 2024, these AI-generated summaries, also known as Search Generative Experiences (SGE), are becoming more prominent in search results, notably in New Zealand. AI Overviews enhance user interactions by integrating multimedia content, providing a richer search experience. Powered by a blend of sophisticated algorithms and Google's extensive web index, AI Overviews analyze large volumes of data to generate summaries. This process involves natural language processing (NLP) to understand the context and highlight key information. AI Overviews are specifically triggered by queries that demand comprehensive understanding or factual information. Conversely, hyper-specific, transactional, or local searches do not typically trigger AI Overviews. While AI Overviews increase content visibility and engagement, they may challenge traditional search engine result visibility, necessitating optimized content for both AI and traditional SEO practices. Notably, AI Overviews achieve a 4% higher click-through rate (CTR) compared to traditional listings. However, they may result in more non-click search sessions, potentially reducing organic traffic to lower-ranked websites. Tools like Google Search Console can help monitor and adapt to these trends. In New Zealand, AI Overviews present new opportunities and challenges for advertisers, requiring strategic content alignment with potential AI Overview triggers. Unlike competitors' AI systems, Google's AI Overviews seamlessly integrate into search results, enhancing the user experience. Despite these advancements, maintaining accuracy and reliability remains crucial, with ongoing algorithm improvements and user feedback mechanisms playing vital roles in enhancing content relevance and accuracy. Importantly, AI Overviews are designed to coexist with traditional search results, highlighting the need for adaptable SEO strategies.

  • 7-2. Google Gemini's Enhanced Capabilities

  • Google Gemini represents a critical evolution in Google's AI development, aimed at transforming user interactions with Google Search. Announced as 'Bard' in February 2023 and widely available by May 2024, Gemini integrates advanced AI technology to generate detailed, contextual responses to natural language prompts. Unlike traditional search results, Gemini offers comprehensive summaries and enables interactive, conversation-based follow-up questions. The AI-driven system fits neatly into Google’s productivity tools like Docs, Drive, and Assistant, enhancing these applications with features such as text summarization and image creation. Gemini's ability to understand and integrate multiple data forms, including text, images, audio, and video, enables more nuanced and personalized search experiences. However, security and privacy concerns have initially limited its context retention capabilities. Significant user data from platforms like Chrome, Gmail, and Workspace necessitate careful handling to ensure privacy and security. Early user feedback has highlighted areas for improvement, particularly in accuracy of AI-generated answers and multilingual performance. Despite mixed reactions, Google emphasizes continuous upgrades to enhance Gemini. For marketers, Gemini's introduction marks a shift from traditional SEO to strategies that align with AI-driven search. This involves creating content that meets user intent and preferences, optimizing for both AI-generated and traditional results. While Gemini's AI results may impact organic traffic, especially for 'how-to' content, marketers can adapt by providing diverse, in-depth content. The integration of Gemini into Google’s ecosystem suggests potential shifts in user behavior and advertiser strategies, requiring businesses to balance AI and traditional search optimizations. Overall, Google Gemini is poised to significantly impact user search interactions, but ongoing improvements are crucial for its widespread acceptance and effectiveness.

8. Apple's AI Innovations

  • 8-1. Siri Integration with ChatGPT

  • Apple is set to integrate ChatGPT into Siri with the release of iOS 18, intending to vastly improve Siri's capabilities in handling complex queries and creative tasks. This integration enables Siri to utilize what Apple calls 'world knowledge', effectively allowing it to answer complex questions that were previously beyond its ability. Users can leverage this feature without the need for additional apps or an OpenAI account. Apple emphasizes privacy in this integration by ensuring there is no logging of requests and requiring user permission each time ChatGPT is used. Optionally, users can link their OpenAI account for enhanced features, and Siri as a hub is expected to incorporate other AI assistants in the future. This significant enhancement makes Siri more versatile across all Apple devices, including iPhones, iPads, and Macs.

  • 8-2. AI Capabilities in Vision Pro

  • Apple is expanding its AI capabilities to the Vision Pro headset, incorporating Apple Intelligence to enhance the device's functionality. The Vision Pro, with its 16GB of memory, will support advanced AI features, and rumors suggest the development of AirPods with built-in cameras to complement spatial audio experiences. The Apple Intelligence+ subscription is anticipated, following a model similar to iCloud's premium tiers. This integration reflects Apple's strategic shift toward software and AI to fuel growth, especially as hardware innovations decelerate. This AI enhancement in mixed reality could fundamentally redefine user interaction with digital content and their environment, signifying a broader application of AI across Apple's product ecosystem.

  • 8-3. Apple Intelligence Announcements

  • At the WWDC 2024, Apple introduced 'Apple Intelligence,' a sophisticated AI system designed to enhance personalization on iPhones, iPads, and Macs. This system includes functionalities such as managing notifications, drafting email responses, and summarizing text. Siri received notable upgrades, including better control of apps and integration with ChatGPT for improved contextual understanding and task performance. Other AI-powered features showcased include custom emoji generation and photo background cleanup. Apple ensures user privacy by utilizing on-device processing and a private cloud computing system. These announcements, along with updates to iOS 18 and visionOS, highlight Apple's commitment to integrating AI across its consumer products to stay competitive in the AI space.

9. Emerging AI Tools and Techniques

  • 9-1. Building AI Assistants for Enterprises

  • For enterprises seeking to leverage AI assistants, integrating generative AI technologies like ChatGPT presents unique challenges, particularly when handling proprietary data, security permissions, and scaling infrastructure. Large Language Models (LLMs) such as GPT-4 and PALM are excellent at general reasoning but struggle with enterprise-specific contexts. Direct fine-tuning of these models on private data can lead to increased hallucinations due to unfamiliarity with specific company knowledge. Alternatively, the Retrieval Augmented Generation (RAG) paradigm is suggested, where knowledge is retrieved through a search system and then used by the LLM. This approach ensures data remains current and secure, as the system only accesses information the user is permitted to see. Key challenges include developing scalable, permission-aware indexing systems that can handle large volumes of enterprise data and building a unified knowledge graph to provide accurate, context-rich search results. Glean’s approach includes centralized indexing, hybrid search techniques, and creating extensive knowledge graphs to personalize search results for individual users, enhancing the overall effectiveness of enterprise AI assistants.

  • 9-2. Advanced Techniques in AI Model Training

  • OpenAI faced a significant breach in early 2023 when a hacker accessed internal forums discussing the latest AI advances. This incident highlighted vulnerabilities in data security but did not compromise core AI systems or customer information. To address these risks, OpenAI established a Safety and Security Committee that includes former NSA head Paul Nakasone. Advanced techniques in AI model training include Google's DeepMind's Joint Example Selection Technique (JEST). JEST accelerates AI training by up to 13 times and reduces computing requirements tenfold by intelligently selecting data batches most beneficial for training. This method has achieved top-tier performance using only 10% of the data required by previous models. These innovations indicate a shift towards more energy-efficient and faster AI training methods, crucial as the industry faces increasing scrutiny over energy consumption.

10. Conclusion

  • The AI landscape is characterized by rapid innovation and intensifying competition, with entities like OpenAI, Google, and Apple leading significant technological progress. Advancements in AI models and integrations, exemplified by Apple Siri's integration with ChatGPT and Google Gemini’s enhanced search capabilities, indicate the strides being made towards more sophisticated AI applications. However, the industry faces crucial security concerns, as highlighted by incidents such as the OpenAI security breach and Microsoft Skeleton Key exploit, stressing the need for stronger safeguards. Financially, market fluctuations in AI crypto tokens and new regulatory measures for cryptocurrency brokers reflect the economic volatility and regulatory impact on the sector. Despite these challenges, the development of new AI tools and techniques, particularly those aimed at enhancing enterprise operations, showcases the transformative potential of AI. However, realizing this potential demands a balanced approach that addresses security, regulatory compliance, and technological scalability. Moving forward, continuing innovation coupled with vigilant security measures and adaptive regulatory frameworks will be essential to harness the full capabilities of AI technology in diverse fields.

11. Glossary

  • 11-1. OpenAI [Company]

  • OpenAI is a leading AI research organization known for its advanced language models like GPT-4. The company faces challenges in security and market competition but continues to innovate with integrations like ChatGPT in various platforms.

  • 11-2. Google Gemini [Technology]

  • An AI ecosystem by Google designed to enhance search capabilities with advanced features including text, image, audio, and video integration. It aims to provide more personalized and contextual search results.

  • 11-3. Perplexity [Startup]

  • An AI-powered search engine that aims to provide faster and more direct answers compared to traditional search engines like Google. It faces challenges related to dependencies on APIs and the need for monetization.

  • 11-4. Microsoft Skeleton Key [Security Exploit]

  • A jailbreak technique discovered by Microsoft's CTO, Mark Russinovich, which can disable alignment constraints on major AI models, potentially leading to unethical outcomes. Highlighting the need for robust security measures in AI development.

  • 11-5. Apple Siri [Product]

  • Apple's virtual assistant that has been enhanced with AI capabilities such as integration with ChatGPT, aimed at improving user interaction and functionality on iOS devices.

12. Source Documents