Catch up on select AI news and developments from the past week or so (in no particular order):
OpenAI outlines plan to shift into a for-profit public benefit corporation. OpenAI has announced plans to transition into a for-profit public benefit corporation (PBC) by 2025 to raise capital for advancing artificial general intelligence. The new structure will place control in a for-profit arm, while the existing nonprofit will retain equity but lose oversight. The nonprofit will focus on initiatives in healthcare, education, and science. This mirrors competitors Anthropic and xAI that operate as PBCs. However, Elon Musk and Mark Zuckerberg have filed motions to block the transition, citing concerns about governance and control. Importance for marketers: OpenAI's restructuring could accelerate AI advancements, providing marketers with more powerful AI tools for automation, content creation, and customer engagement.
DeepSeek-V3 open-source AI model surpasses Meta's Llama and Qwen. DeepSeek has unveiled DeepSeek-V3, an ultra-large open-source AI model with 671 billion parameters that outperforms Meta's Llama 3.1-405B and Qwen 2.5-72B in benchmarks. Its mixture-of-experts architecture activates select parameters, improving efficiency. The model excels in Chinese and math-centric tasks, challenging even closed systems like GPT-4o. Despite its $5.57 million training cost—significantly lower than competitors'—DeepSeek-V3 is accessible through Hugging Face and GitHub. Importance for marketers: DeepSeek-V3 expands affordable AI options, allowing marketers to integrate cutting-edge AI for analytics and content creation without relying on costly proprietary models.
Sam Altman seeks feedback from users to shape OpenAI's 2025 roadmap. OpenAI CEO Sam Altman used X to gather user suggestions for improving OpenAI's products in 2025. Suggestions included family accounts with child-friendly guardrails, better voice-chat features, and enhancements to Sora, OpenAI's text-to-video model. This follows a turbulent year involving legal disputes with Elon Musk and record-breaking $6.6 billion in funding. Altman's outreach underscores a focus on user-driven innovation and product refinement. Importance for marketers: OpenAI's openness to feedback signals potential improvements in AI tools marketers rely on, fostering more tailored solutions for content generation and customer engagement.
Apple's iOS 19 to feature advanced Siri with ChatGPT-like capabilities. Apple's iOS 19 will introduce a more advanced Siri powered by large language models, enabling more complex interactions similar to ChatGPT. The update, expected in 2025, will gradually roll out across devices but may not fully launch until iOS 19.4 in 2026. The update ensures compatibility with all iPhones currently running iOS 18, including older models like the iPhone XR. Importance for marketers: Enhanced voice assistants will create new opportunities in voice search optimization and conversational marketing, shifting how brands engage with audiences through AI-powered assistants.
OpenAI offers unlimited Sora access to Plus users during the holidays. OpenAI is temporarily providing ChatGPT Plus users unlimited access to Sora, its AI video generator. Typically capped at 50 videos per month, the holiday promotion aims to boost user engagement and creative experimentation. CEO Sam Altman attributed the offer to lower GPU demand during the holidays. This follows high demand for Sora's launch, which led to temporary sign-up restrictions. Importance for marketers: Unlimited Sora access enables marketers to experiment with AI video content, opening new possibilities for dynamic ad creation and multimedia campaigns.
Microsoft integrates non-OpenAI models into 365 Copilot to reduce costs. Microsoft is diversifying AI models for its 365 Copilot product, incorporating in-house and third-party models to reduce dependency on OpenAI. The initiative aims to cut costs and improve performance, addressing enterprise concerns about speed and expense. Microsoft is training smaller models like Phi-4 and customizing open-weight models to enhance efficiency. OpenAI remains a partner for advanced frontier models. Importance for marketers: Microsoft's diversification could lead to more cost-effective and flexible AI tools within the 365 ecosystem, improving productivity and marketing automation.
Tencent's open-source AI video model spurs creative experimentation. Tencent's HunyuanVideo AI video generator allows users to run uncensored, high-quality models locally, resembling Stable Diffusion's impact on AI image generation. Early tests show HunyuanVideo competes with models like Google's Veo 2 and OpenAI's Sora, though output quality varies. The model's open nature fuels hobbyist interest and raises ethical questions about content misuse. Importance for marketers: Open-source AI video tools enable cost-effective video production for marketers, allowing faster ad creation and personalized video campaigns.
Italy fines OpenAI €15 million over ChatGPT privacy violations. Italy's data regulator fined OpenAI €15 million for processing personal data without legal grounds and failing to implement adequate age verification for ChatGPT users. OpenAI disputes the fine and plans to appeal, calling it disproportionate. The regulator also ordered OpenAI to launch a public awareness campaign about its data practices. Importance for marketers: The fine highlights increasing regulatory scrutiny of AI, emphasizing the need for marketers to monitor data compliance and ethical AI deployment.
UK creatives reject AI copyright exemption proposal. A coalition of UK writers, musicians, and filmmakers has strongly opposed the government's proposal to allow AI companies to train on copyrighted material without prior consent. The plan would require rights holders to opt out rather than granting AI firms proactive permission. Creatives argue this undermines intellectual property rights and call for licensing agreements and compensation instead. The government suggests stricter rules could disadvantage UK AI developers, though opposition persists across media sectors. Importance for marketers: Marketers using AI-generated content must stay informed about evolving copyright laws, ensuring compliance and respecting IP rights in content development.
Google unveils Gemini 2.0 Flash Thinking for enhanced reasoning. Google introduced Gemini 2.0 Flash Thinking, an AI model designed to solve complex tasks by simulating structured reasoning steps. The model uses large datasets and processes visual and textual inputs. Gemini 2.0 aims to improve interpretive AI, directly competing with OpenAI's models. Early demos show the model solving physics problems and analyzing detailed queries by breaking them into smaller parts, showcasing advanced problem-solving capabilities. Importance for marketers: AI models capable of structured reasoning could enhance marketing data analysis and improve customer personalization, offering deeper insights and more accurate predictions.
Google to introduce conversational AI answers in Search. Google plans to launch an AI Mode for Search, providing chatbot-like conversational responses alongside traditional search results. This feature, integrated with Gemini, will present users with links and interactive prompts, streamlining complex queries. The move signals Google's push to enhance search experiences through generative AI, following similar efforts by OpenAI. AI Mode is expected to roll out in 2025, reshaping how users engage with search engines. Importance for marketers: Conversational AI in search will drive SEO shifts, requiring marketers to optimize content for AI-driven responses, adapting to new formats for better visibility.
Tencent's AI video model empowers hobbyists with open-source generation. Tencent's HunyuanVideo AI video generator allows local, uncensored video creation, echoing the disruptive impact of Stable Diffusion in AI imagery. The model runs on consumer GPUs and can be fine-tuned, offering new creative potential for hobbyists. While its open nature fuels innovation, it raises concerns about misuse, including generating explicit content. Early tests show comparable performance to commercial models but highlight ongoing quality inconsistencies. Importance for marketers: Open-source AI video tools offer marketers low-cost alternatives for custom video content, accelerating campaign development and reducing production expenses.
Google criticized for using inexperienced workers to fact-check Gemini AI. Reports reveal Google instructed contract workers to evaluate Gemini AI outputs without requiring expertise in the subject matter. This shift replaces prior guidance allowing raters to skip prompts they lacked expertise in. Critics argue this reduces the accuracy of AI evaluations, potentially undermining the reliability of Gemini. Google maintains that raters primarily provide feedback on format and style rather than content accuracy. Importance for marketers: Ensuring AI accuracy remains crucial for marketers. This highlights the importance of verifying AI-generated outputs, especially in content marketing and automated campaigns.
OpenAI begins testing new 'o3' reasoning AI models. OpenAI has started internal testing of o3 and o3 mini, its latest AI models designed to tackle complex tasks with enhanced reasoning. The models are expected to outperform existing versions and enter public beta in early 2025. These advancements follow OpenAI's launch of o1 models, which showed significant improvements in coding, math, and scientific tasks. The o3 models aim to solidify OpenAI's lead in the AI space. Importance for marketers: More capable AI models will improve automation and analytics, allowing marketers to streamline workflows and enhance data-driven decision-making.
Apple faces criticism over AI-generated news errors. Apple's AI mistakenly summarized a BBC headline, inaccurately reporting that a suspect had killed himself. This prompted calls from Reporters Without Borders for Apple to suspend the AI feature, citing risks to public trust. Apple's AI has previously misrepresented news, highlighting concerns about accuracy in AI-generated content. Apple has not publicly addressed the issue but faces mounting pressure from media watchdogs. Importance for marketers: The incident highlights the need for rigorous oversight when using AI for automated content. Marketers should carefully review AI-generated materials to prevent misinformation.
OpenAI's "12 Days of OpenAI" brought a flurry of product launches, updates, and new features aimed at enhancing the ChatGPT experience for users and developers. Spanning from December 5 to December 20, the daily livestream series introduced advancements, including the powerful o3 reasoning model, new developer tools, expanded search capabilities, and enhanced voice interactions. Notable highlights included the launch of Sora Turbo for video generation, Apple's integration of ChatGPT into iOS, and the introduction of "Projects" for better chat organization. More details below:
Thursday, December 5: OpenAI launched its "12 Days of OpenAI" with major updates to ChatGPT and the o1 model.
- Full version of the o1 model released, improving accuracy by 34% and processing 50% faster.
- o1 now supports multimodal reasoning, allowing users to input images.
- ChatGPT Pro subscription tier introduced for $200/month, providing unlimited access to o1-mini, GPT-4o, and advanced features.
Friday, December 6: OpenAI expanded its Reinforcement Fine-Tuning Research Program.
- Developers and researchers can fine-tune models for domain-specific tasks.
- The program improves model accuracy by incorporating user feedback to refine reasoning.
- Applications are open for universities and enterprises, with public availability expected in 2025.
Monday, December 9: OpenAI introduced Sora Turbo, its advanced video generation model.
- Sora generates text-to-video and video-to-video content.
- Available to ChatGPT Plus and Pro users in the U.S. with up to 50 videos/month.
- Features a Storyboard tool for frame-by-frame video generation and an Explore page to view user creations.
Tuesday, December 10: Canvas, OpenAI's interactive workspace, was rolled out to all web users.
- Canvas is now integrated into GPT-4o by default.
- Users can run Python code, manage projects, and use Canvas with custom GPTs.
- Canvas helps streamline development and debugging processes.
Wednesday, December 11 Apple integrated ChatGPT with iOS 18.2, enhancing Siri and Writing Tools.
- Siri can redirect queries to ChatGPT with user permission.
- Visual Intelligence lets users perform real-world searches via their camera.
- Writing Tools gains a "Compose" feature for generating text and images.
Thursday, December 12: Advanced Voice Mode received major upgrades, including screen sharing and Santa Voice.
- Advanced Voice Mode now interprets visuals from phone cameras and screens.
- Users can enable "Santa Voice" by clicking the snowflake icon in ChatGPT's voice mode.
- Screen sharing rolled out to Plus, Pro, and Team users.
Friday, December 13: OpenAI introduced "Projects," an organizational tool for managing conversations.
- Projects allow users to group chats, add context, and organize files within ChatGPT.
- Available to Plus, Pro, and Teams users, with free user rollout planned.
Monday, December 16: ChatGPT search expanded to all users, not just Plus subscribers.
- Users can now browse the web within ChatGPT, enhancing mobile and voice experiences.
- Visual search results and maps integrated into ChatGPT's interface.
Tuesday, December 17: "Mini Dev Day" focused on developer updates.
- o1 exited preview, adding function calling, structured outputs, and vision capabilities.
- Realtime API saw price reductions and WebRTC support.
- Preference Fine-Tuning introduced to enhance desired model behaviors.
Wednesday, December 18: ChatGPT became available via phone calls and WhatsApp.
- Users can call 1-800-ChatGPT for free monthly access (15 minutes).
- WhatsApp users can text ChatGPT directly for answers.
Thursday, December 19: OpenAI launched a MacOS desktop app with new integrations.
- Desktop app now works with Apple Notes, Notion, Quip, and various coding apps.
- Advanced Voice Mode added to the desktop app for improved functionality.
Friday, December 20 OpenAI concluded with the unveiling of o3 and o3 mini models.
- o3 surpasses o1 in benchmarks, nearing AGI levels.
- o3 mini introduces flexible reasoning modes.
- OpenAI opened external safety testing for o3, with public launch expected in January.
You can find the previous issue of AI Update here.
Editor's note: GPT-4o was used to help compile this week's AI Update.