The ChatGPT Observer Edition 30

If you're short on time, here's your one-minute summary of this week's key AI news and advancements:

?? Top Story: OpenAI expands Advanced Voice Mode access to most Plus and all Team subscribers, offering nine lifelike voices and enhanced interaction capabilities.

?? This Week's Spotlight: Meta Connect 2024 unveils Llama 3.2 models, updates to Meta AI assistant, and new AI tools for creators.

?? ChatGPT & OpenAI News: Sam Altman predicts superintelligence in "a few thousand days"; OpenAI launches OpenAI Academy; Jony Ive confirms AI hardware project with OpenAI.

?? Corporate AI Developments: Google enhances NotebookLM with audio and YouTube analysis; Visa acquires AI-based fraud prevention firm Featurespace; IBM and NASA team up for open-source AI climate model.

?? AI Innovations: AI2 introduces Molmo, an open-source AI model outperforming GPT-4 and Claude on certain benchmarks.

?? Applications of AI: T-Mobile partners with OpenAI for AI-powered customer experience platform; Duolingo introduces interactive AI chat and games for language learning.

? Responsible AI: LinkedIn halts AI training using UK user data amid privacy concerns; FTC targets deceptive AI claims in crackdown on five companies.

?? AI Learning Hub: Google Cloud's Gemini drives real-world AI solutions across industries; Understanding synthetic data in AI & machine learning.

?? For more details, read the full newsletter!

?? Top Story: OpenAI Introduces Advanced Voice Mode with New Voices

As of September 27, 2024, OpenAI has significantly broadened access to its Advanced Voice Mode (AVM) for ChatGPT, marking a major leap forward in AI-powered voice interactions. This enhanced feature is now available to all Team and most Plus subscribers, with some geographic limitations.

Get more natural, real-time conversations with advanced voice. Senses and responds to humor, sarcasm, interruptions, and more.

Availability and Access

• Current Access: Available to all Team and most Plus users.
• Upcoming Expansion: Enterprise and Edu customers will gain access starting next week.
• Geographic Limitations: Not available in the European Union, Switzerland, Iceland, Norway, and Liechtenstein.
• App Requirements: Available in iOS and Android mobile apps, version 1.2024.261 or later.

Key Features and Improvements

1. Voice Options: Nine lifelike output voices: Arbor, Breeze, Cove, Ember, Juniper, Maple, Spruce, Vale, and Sol. Each voice has a distinct tone and character. The "Sky" voice has been removed due to legal issues.
2. Enhanced User Interface: AVM is represented by a blue animated sphere. Users receive a pop-up notification when AVM becomes available.
3. Advanced Interaction Capabilities: Real-time conversations with interruption capability. Emotion interpretation based on voice tone. Improved accent recognition and conversation flow.
4. Customisation and Personalization: Custom Instructions for tailored interactions. Memory feature for context retention across conversations.
5. Multilingual Proficiency: Supports over 50 languages.

Types of Voice Conversations

1. Advanced Voice: Uses GPT-4o's native audio capabilities. Features more natural, real-time conversations. Responds to non-verbal cues and can convey emotion. Daily usage limits apply for Plus and Team users.
2. Standard Voice: Available to all signed-in ChatGPT users. Uses multiple models, including transcription before response generation. Counts towards message limits.

User Experience and Functionality

• Background Conversations: Both standard and advanced voice can continue in the background.
• Conversation Resumption: Advanced voice conversations can be resumed in advanced voice, text, or standard voice.
• GPT Compatibility: Standard voice available with GPTs; advanced voice not yet available.
• Custom Instructions and Memories: Accessible in advanced voice conversations.
• Musical Content: Not supported to respect creators' rights.
• Volume Control: Set through the device, not within ChatGPT.
• Subtitles: Not available during conversations; transcriptions added to chat history afterward.
• Multiple Conversations: Only one voice chat at a time.

Privacy and Data Usage

• Audio Retention: Advanced voice audio clips stored with transcriptions in chat history.
• Deletion Policy: Audio clips deleted within 30 days of chat deletion, with exceptions for security or legal reasons.
• Model Training: Audio clips not used for training unless explicitly shared by users.
• Sharing Options: Free and Plus users can opt to share audio to improve voice models.

Limitations and Considerations

• Daily Limits: Usage limits apply for advanced voice; users notified when approaching the limit. (?? Note: Based on experience, the daily limit appears to be around 2 hours, though OpenAI has not officially specified this.)
• Interruptions: May occur occasionally; using headphones and enabling Voice Isolation on iPhone recommended.
• Transcription Accuracy: May not always align perfectly with the original conversation.
• Language Detection: Occasional inaccuracies; users can verbally correct or set preferences in settings.

Looking Ahead

This comprehensive rollout of Advanced Voice Mode signifies OpenAI's commitment to advancing voice-based AI communication. By making AVM available to a broader user base, OpenAI is positioning ChatGPT at the forefront of conversational AI technology. The company encourages user feedback, which will likely inform future improvements and innovations in AI-human interactions. As voice interaction becomes an increasingly important aspect of AI technology, this update underscores OpenAI's focus on enhancing user experience and expanding the practical applications of AI in daily life and work environments. The ongoing development of AVM is expected to drive further advancements in natural language processing and voice-based AI interactions, potentially revolutionizing how we interact with AI assistants in the future.

Read more: OpenAI announcement on X | Voice Mode FAQ | TechCrunch | TomsGuide | Usage Examples on X | AI with Human duet

?? This Week's Spotlight: Meta Connect 2024

Meta Connect 2024 brought significant AI-related announcements, showcasing the company's commitment to advancing its AI capabilities. Here's a summary of the key AI news from the event:

Meta AI Assistant Updates

Meta AI, the company's AI assistant, is on track to become the world's most-used AI assistant by the end of 2024, with nearly 500 million monthly active users. The assistant received several notable updates:

1. Llama 3.2 Integration: Meta AI is now powered by the new open-source Llama 3.2 models, enhancing its capabilities and making it smarter.
2. Multimodal Abilities: The assistant can now understand and process both text and images, allowing users to share photos for analysis and information.
3. Voice Interaction: Users can now engage with Meta AI using voice commands across platforms like Messenger, WhatsApp, and Instagram.
4. Celebrity Voices: Meta introduced a range of celebrity voices for the AI assistant, including John Cena, Awkwafina, and Dame Judi Dench.

Llama 3.2 Model

Meta unveiled its latest large language model, Llama 3.2, with several improvements:

• Described as Meta's first multimodal large language model (MLM)
• Improved reasoning capabilities, especially for math and coding tasks
• Support for eight languages
• A context window of 128,000 tokens
• Available in light models (1B, 3B) and vision models (11B, 90B) parameter versions

AI Tools for Creators

Meta introduced new AI-powered tools for content creators:

• AI-powered image editing capabilities, allowing users to remove and edit objects within images
• Automatic lip-syncing and video dubbing for Reels, currently in testing with creators in the US and Latin America
• "Imagine" feature for generating AI images and captions on Facebook and Instagram

AI Integration Across Platforms

Meta is expanding AI integration across its platforms:

• Meta AI is now available on WhatsApp, Instagram, Facebook, and Messenger
• The company is testing AI-driven "imagined" content across Facebook and Instagram feeds

These AI announcements at Meta Connect 2024 demonstrate the company's focus on enhancing user experiences through advanced AI technologies, with particular emphasis on making AI more accessible and integrated into daily interactions across its various platforms.

Read more: Meta Connect 2024 | Watch on Demand | Llama 3.2 models | Rayban Meta AI Features

Meta's Llama 3.2: Expanding Access Across AI Platforms ??

The following summaries highlight the integration of Llama 3.2 across three major cloud platforms:

• Meta and Groq Expand Open-source AI Ecosystem with Llama 3.2 Launch: Meta and Groq continue their partnership by launching Llama 3.2, expanding open-source AI with vision LLMs (11B & 90B) and text-only models (1B & 3B). The collaboration leverages Groq's fast inference technology, offering developers high speed and scalability for building AI applications. Llama 3.2 models, now available for preview via GroqCloud, include image reasoning capabilities and expanded text modalities, enhancing innovation in generative AI. Developers can easily integrate with Groq's API to start building with Llama 3.2 today. Read more: Groq
• Meta's Llama 3.2 Launches on Google Cloud: Google Cloud has added Meta's Llama 3.2 models to Vertex AI Model Garden. The suite includes 11B & 90B vision models for image processing and lightweight 1B & 3B models for mobile integration. The models offer personalized AI experiences with a focus on privacy and efficiency. Companies like Shopify and TransCrypts are already using Llamafor data generation and AI-driven solutions. 90B model is available for preview, with more to come soon. Read more: Google Cloud Blog
• Meta's Llama 3.2 Models Now on Amazon Bedrock: Amazon Bedrock now offers Meta's Llama 3.2 models, including 11B & 90B vision models for sophisticated reasoning and 1B & 3B lightweight text-only models for edge devices. These models support multimodal AI tasks like image processing and are designed for efficiency and reduced latency. The models are available across regions in the US and Europe for cross-region inference, enabling a wide range of applications. Read more: Amazon Bedrock

?? ChatGPT & OpenAI News

• Sam Altman on "The Intelligence Age" and the Countdown to Superintelligence: In his essay "The Intelligence Age," Sam Altman, CEO of OpenAI, shares an optimistic outlook on AI's transformative impact on society and human progress. Altman believes we may achieve superintelligence in just a few thousand days, revolutionising our capabilities and leading to unimaginable advancements in education, healthcare, and more. Deep learning has been the driving force behind AI's rapid progress, and as it scales, AI will continue to improve. To make this accessible to all, Altman stresses the need for abundant compute and energy, while also navigating the complex challenges and risks wisely. Read more: The Intelligence Age | FoxBusiness
• OpenAI Academy: Shaping Global AI Knowledge: OpenAI has launched the OpenAI Academy, a platform aimed at fostering global understanding of artificial intelligence. It focuses on educating leaders, policymakers, and practitioners around the world on the opportunities and challenges of AI. The academy aims to provide resources, training, and knowledge sharing to help shape informed AI strategies, bridging the gap between technology development and policy. This initiative showcases OpenAI's commitment to responsible AI deployment and global collaboration. Read more: OpenAI Academy
• Jony Ive Confirms AI Hardware Project with OpenAI: Jony Ive, former Apple design chief, is collaborating with OpenAI’s CEO Sam Altman on a new AI hardware venture. Developed by Ive’s firm LoveFrom and backed by the Emerson Collective, the project aims to raise up to $1 billion. A small team, including former iPhone developers, is working on a groundbreaking AI-driven device in San Francisco. While details are scarce, the device is set to leverage generative AI to surpass traditional software capabilities. Read more: Maginative
• Genmab Launches "AI Everywhere" Initiative with OpenAI: Genmab, a global biotech company, has rolled out ChatGPT Enterprise to over 2,000 employees, aiming to integrate AI across all operations. With over 100 custom GPTs built, Genmab uses AI for drafting documents, summarising scientific literature, analytics, and efficient translations, saving employees an average of 3.5 hours per week. The GPT-4o enhancements have enabled more complex document processing. Genmab ensured robust privacy and compliance by evaluating OpenAI's data handling and security standards. Read more: OpenAI
• OpenAI CTO Mira Murati Departing: OpenAI’s CTO, Mira Murati, is reportedly leaving the company. Murati, who has played a key role in the development and direction of OpenAI's AI advancements, will be stepping down from her position, marking a significant leadership change for the company. The reasons for her departure are yet to be detailed, but this shift could have a notable impact on OpenAI’s future strategy and developments. Read more: The Verge | Mira Murati X announcement

?? Corporate AI Developments

• Google NotebookLM Enhances AI with Audio and YouTube Analysis Features: Google’s NotebookLM, now powered by Gemini, has expanded its AI capabilities by enabling users to analyse YouTube videos and audio files. Users can input YouTube video URLs to easily get summaries of key points and arguments, likely leveraging video captions for analysis. Additionally, NotebookLM can generate "Audio Overviews," which are spoken summaries of documents, articles, or other content, making learning more accessible for auditory learners. Users can also upload audio files in formats like .mp3 and .wav for AI summarization. A new sharing feature allows users to generate a public URL of the Audio Overview to share with others directly. Read more: Android Police

• Google Updates Gemini Models with Reduced Pricing and Higher Limits: Google has released updated Gemini models, reducing Gemini 1.5 Pro pricing and increasing rate limits for higher accessibility and cost-efficiency. The new changes allow for more extensive model usage and enhanced AI capabilities for developers. Additionally, the Vertex AI platform has new features that make model deployment and usage more efficient, catering to a broader range of enterprise and developer needs. Read more: Google Developers Blog
• Visa Acquires AI-Based Fraud Prevention Firm Featurespace: Visa has agreed to acquire Featurespace, a UK-based AI-driven payments protection firm, to enhance its financial crime and fraud detection capabilities. While the financial terms remain undisclosed, a Sky News report valued the deal at ￡700 million ($935.06 million). Founded in 2008, Featurespace boasts clients like HSBC, NatWest, and Worldpay. The acquisition is expected to broaden Visa's value-added services by leveraging Featurespace’s AI for real-time fraud management. This move follows Visa’s prevention of 80 million fraudulent transactions worth $40 billion last year, thanks to its AI investments. Read more: Reuters
• IBM and NASA Team Up for Open-Source AI Climate Model: IBM and NASA are collaborating to develop an open-source AI model focused on weather and climate analysis. This initiative aims to improve climate forecasting and environmental data analysis. By making the AI model open-source, IBM and NASA intend to enable researchers and developers worldwide to enhance climate predictions and address challenges related to weather patterns and climate change. Read more: Yahoo Finance
• Runway Launches $1 Million AI Film Fund for Filmmakers: Runway has launched The Hundred Film Fund, offering grants of $5,000 to $1 million to filmmakers using AI in their projects, aiming to produce up to 100 films. The fund supports projects utilizing generative AI, with an advisory panel including Jane Rosenthal and will.i.am assessing submissions. Runway doesn't require exclusive use of its tools or claim IP rights, fostering creative freedom. This initiative follows Runway’s partnerships, including a deal with Lionsgate and its AI Film Festival collaborations. Read more: Variety | TechCrunch
• Luma AI Launches Dream Machine API Beta for Video Generation: Luma AI has introduced the Dream Machine API in beta, offering an intuitive REST API with familiar SDKs to help developers integrate AI video creation into their apps seamlessly. With features like visual effects, Keyframes for storytelling, and Camera Motion controls, the API simplifies video content creation without the need for complex tool-building. Read more: Luma AI on X | Get started here: Dream Machine API
• Alibaba Cloud & Nvidia Partner on AI for Autonomous Driving: Alibaba Cloud and Nvidia have teamed up to integrate Alibaba’s Qwen large language models (LLMs) with Nvidia's Drive AGX Orin platform, enhancing AI capabilities for autonomous driving. The collaboration, announced at the Apsara Conference, aims to improve in-car experiences with AI-powered voice assistants for dynamic conversations, visual intelligence, and proactive features like headlight suggestions. Plans include adapting Qwen LLMs to Nvidia's Drive Thor for advanced driver assistance and AI cockpit functionality. Read more: SCMP
• Meta AI Nears 500 Million Monthly Users Globally: During Meta Connect 2024, Mark Zuckerberg revealed that Meta AI is one of the most-used AI assistants globally, approaching 500 million monthly active users. The AI tool hasn’t even launched in major markets like the EU yet. India remains its biggest market due to WhatsApp’s user base. For context, OpenAI’s ChatGPT currently has over 200 million weekly users. Meta also announced new updates, including the launch of the Llama 3.2 model series with multimodal capabilities, lightweight models for Europe, celebrity voices, and enhanced AI features across platforms like Facebook, Instagram, and Messenger. Read more: TechCrunch
• UK Clears Amazon's $4B Partnership with AI Firm Anthropic: The UK's Competition and Markets Authority (CMA) has cleared Amazon’s partnership with Anthropic, a $4 billion deal that didn't require further scrutiny due to Anthropic’s revenue and market share not meeting UK merger thresholds. Under the partnership, Anthropic uses Amazon Web Services (AWS) and Amazon's chips to develop its AI models. The CMA has also approved Microsoft's partnerships with AI startups Mistral AI and Inflection AI but is still reviewing a deal between Anthropic and Google. The partnership reflects increasing global scrutiny of tech investments in AI. Read more: AP News

?? AI Innovations

• AI2’s Molmo Outperforms GPT-4 and Claude on Certain Benchmarks: The Allen Institute for AI (AI2) has introduced Molmo, an open-source AI model that shows significant performance on some benchmarks compared to leading AI models like GPT-4 and Claude. Molmo is designed to be multimodal, enabling it to handle different forms of data like text and images. The model comes in several sizes, allowing developers to choose the most suitable version for their needs. The open-source nature of Molmo means it can be fine-tuned extensively, making it an attractive option for AI researchers and developers aiming to build sophisticated AI applications. Read more: VentureBeat | Molmo

?? Applications of AI

• T-Mobile Partners with OpenAI for AI-Powered Customer Experience Platform: T-Mobile has announced a partnership with OpenAI to create an AI platform, IntentCX, designed to enhance customer service by analyzing interactions from its T-Life app. The platform aims to automate tasks like troubleshooting and predict issues that lead to customer churn, using insights from past service calls and interactions. IntentCX will utilize secure customer data to improve user experience but will not contribute to training OpenAI's core models like ChatGPT. The AI platform is set to launch in 2025, with plans for additional AI-driven services in the pipeline. Read more: The Wall Street Journal
• Duolingo Introduces Interactive AI Chat and Games for Language Learning: Duolingo has launched a new interactive experience, allowing users to chat and play games with AI characters to enhance their learning of Spanish and French. These AI-powered features provide immersive language practice, helping users develop their skills through real-time conversation and engaging activities. The goal is to make language learning more dynamic and personalized, using AI to simulate realistic dialogues and scenarios. Read more: TechRadar | Duolingo Press Release
• AI Enhances Weather Forecasting Accuracy: The EC-AIFS AI model from ECMWF predicted Hurricane Francine 10 days before landfall in Louisiana, outperforming traditional models in speed and consistency. AI models process vast weather data quickly and efficiently, though they currently have lower resolution and focus on large-scale events like hurricanes. NOAA is also investing $100 million in AI-based forecasting technology, with human oversight remaining key to ensure accuracy. Read more: FOX Weather

? Responsible AI

• LinkedIn Halts AI Training Using UK User Data Amid Privacy Concerns: LinkedIn, owned by Microsoft, has suspended using UK user data to train its AI models after the UK’s Information Commissioner's Office (ICO) raised privacy concerns. The platform had been using user-generated content for AI training, but now UK users, along with those in the EU, European Economic Area, and Switzerland, can opt out of this practice. LinkedIn stresses that user control over data remains a priority. Other tech giants like Meta have faced similar scrutiny, with plans to use user data for AI training also paused in the UK. Read more: BBC News
• ChatGPT Error in Child Protection Case Raises Privacy and Safety Concerns: A child protection worker used ChatGPT to draft a Protection Application Report for a custody case, accidentally disclosing sensitive information about the child and family involved. The AI’s output included inappropriate assessments, such as downplaying the severity of risks to the child and misrepresenting details like a toy used in harmful ways as a "notable strength." The Victorian Office of the Information Commissioner (OVIC) launched an investigation, which has spotlighted the risks of using AI in high-stakes environments. Although the outcome of the case was not affected, the incident underlines the potential dangers of AI-generated errors and the need for careful oversight, especially in contexts involving privacy and child safety. Read more: Information Age
• FTC Targets Deceptive AI Claims in Crackdown on Five Companies: The U.S. Federal Trade Commission (FTC) has taken action against five companies accused of using AI in deceptive ways. Three businesses that promised passive income through e-commerce storefronts were forced to suspend operations. The FTC also settled with DoNotPay, an AI-based legal services company, and Rytr, an AI writing tool that allowed users to create fake product reviews. DoNotPay agreed to pay $193,000 and notify customers of its legal service limitations, while Rytr agreed to discontinue its review generation feature. This enforcement underscores the FTC’s stance that AI tools must comply with existing laws against fraud and deceptive practices. Read more: Reuters | FTC
• Cloudflare's AI Audit Tool Empowers Content Creators to Control Bot Access: Cloudflare has launched a new tool, "AI Audit," to help website owners monitor and control how AI bots interact with their content. It provides detailed analytics on bot access frequency, content types accessed, and bot purposes (such as text generation or data scraping). With AI Audit, creators can block all AI bots with a click or differentiate between bots that credit sources and those that don’t. Future enhancements will include setting fair prices for AI companies accessing content for training purposes. This tool aims to empower content creators to retain control and potentially monetize their work rather than having AI companies use it without compensation. Read more: ZDNet
• Deepfake Caller Impersonates Ukrainian Official in Attempt to Deceive US Senator: A deepfake call nearly fooled Sen. Ben Cardin (D-MD), head of the Senate Foreign Relations Committee. The caller used AI to impersonate Dmytro Kuleba, Ukraine’s former foreign minister, during a Zoom meeting. Suspicious behavior and politically charged questions tipped off Cardin, who later reported the incident. The State Department confirmed the imposter’s identity, although the person behind the deepfake remains unknown. This incident highlights the growing threat of sophisticated AI deepfakes used for political manipulation, as noted by Senate security officials who warn of more attempts to come. Read more: The Verge
• 'Goodbye Meta AI' Posts Do Not Protect User Data from AI Training: Many users, including celebrities, have shared a message titled "Goodbye Meta AI," hoping it would prevent Meta (Facebook, Instagram, WhatsApp) from using their content for AI training. However, posting this message does not offer any legal or privacy protection. Meta confirmed that only formal objection forms filled out by users will be considered valid. The "Goodbye Meta AI" message is similar to past viral posts attempting to assert privacy rights without legal effect. To opt out of having content used for AI training, users in the UK will receive a notification linking to an official objection form from Meta. Read more: The Guardian
• California’s AI Safety Bill SB 1047 Awaits Decision: California's SB 1047, awaiting Governor Gavin Newsom’s signature, aims to regulate AI's use in the entertainment industry. The bill focuses on protecting performers' rights, mandating transparency, consent, and compensation for AI-created digital replicas and likenesses. Supported by SAG-AFTRA, the legislation addresses concerns over AI's role in Hollywood, particularly with rising use in film production. Newsom's decision will significantly impact how AI is adopted and controlled in the industry. Read more: The Verge

?? AI Learning Hub

• Google Cloud’s Gemini: Driving Real-World AI Solutions Across Industries

In a keynote titled "Gemini at Work in less than 7 minutes," Thomas Kurian discusses how Google Cloud’s Gemini platform is revolutionizing industries through AI agents. Kurian emphasizes the platform’s real-world AI impact, open accessibility, and the rise of AI agents for customers, data, and creativity. Key customer success stories include:

• Snap: Uses Gemini’s visual AI capabilities to offer solutions based on visual context.
• GoTo Group: Developed an AI agent to simplify Indonesian transactions with Gemini 1.5 for optimal translation.
• Warner Bros. Discovery: Achieved up to 80% time savings in captioning tasks.
• Tombras Agency: Created an AI-driven dynamic billboard campaign, producing over 6,000 headlines in 29 hours.

Kurian highlights how Google Cloud aims to democratize AI for businesses, transforming operations with Gemini'sinnovations and open platform tools.

Watch the keynote: YouTube

• Understanding Synthetic Data in AI & Machine Learning

Synthetic data is artificially generated data that mimics real-world data, often used in AI and machine learning when actual data is limited, sensitive, or expensive to collect. It helps in training AI models by offering scalable, diverse, and privacy-safe datasets for applications like testing algorithms, enhancing models, and simulating various scenarios. Synthetic data plays a key role in areas such as computer vision, natural language processing, and predictive analytics.

Read more: Amazon Web Services

??? We'd Love to Hear from You!

Did you find this edition insightful? Like, share your thoughts, and ask us anything.

Stay curious, and until next time, enjoy exploring the vast possibilities AI offers.

???? Subscribe on LinkedIn The ChatGPT Observer Newsletter

As we explore the captivating world of AI technologies, it's important to note that the content of this article is a product of human effort, enhanced by AI tools and additional resources. We have made sure to properly attribute and link all referenced articles to their original sources. For the visuals:

• Images sourced from external sites are credited within the article.
• Other images were generated using DALL-E, MidJourney and IdeoGram with enhancements made in Canva. These AI-generated and enhanced images are designed to visually represent each section's key themes.