-2.svg){kind=logo}
This last month has seen rapid generative AI developments that marketers should aim to stay on top of to remain competitive. This round-up summarises key insights from episode 27, episode 28, episode 29 and episode 30 of the "Artificially Intelligent Marketing" podcast, hosted by our CEO Paul Avery and his industry colleague Martin Broadhurst.
These episodes covered major new releases like upgraded versions of OpenAI’s ChatGPT and DALL-E, Google’s conversational AI, and Adobe’s enhanced creative tools. The opportunities to leverage AI continue expanding across content creation, data analysis, team collaboration, and more. However, the hosts also highlighted important considerations around responsible AI adoption, data quality, human oversight, and emerging legal issues. With new models constantly launching, taking time to properly evaluate and understand these technologies is crucial.
The summaries provide a high-level overview of the main discussions. For full details, we encourage you to listen to episodes 27, 28, 29 and 30, available below or on your preferred podcast platform.
Summaries from episode 27:
ChatGPT Gains Voice and Vision: Increasing Accessibility and Functionality
OpenAI has upgraded ChatGPT with voice and image capabilities, making the platform more interactive and multimodal. Voice is now available in the iOS and Android apps for paying subscribers, with wider rollout expected. Image inputs have also been launched, enabling tasks like website creation from sketches. While expanding creative applications, OpenAI acknowledges risks like impersonation and aims to enable identifying AI-generated content.
DALL-E 3's Text-to-Image Revolution: Easier Visual Asset Creation
DALL-E 3, launching in October, will transform text-to-image generation with more accurate outputs. It is natively integrated into ChatGPT and can refine prompts for detailed, precise images while limiting harmful content. DALL-E 3 presents opportunities for marketing campaigns, though won't imitate living artists' styles. OpenAI is developing tools to identify AI-generated images.
Amazon & Anthropic Partnership: $4 Billion Investment in AWS Capabilities
Amazon has invested $4 billion into AI company Anthropic, becoming its minority stakeholder. Anthropic's models will be featured on Amazon's AI platform, Amazon Web Services, enabling customisation for AWS customers. Early adopters like Lonely Planet and Bridgewater report dramatically reduced costs and streamlined workflows.
Anthropic's Claude: Boosting Recall in Long Documents
Anthropic has published research showing how to optimise Claude's memory and accuracy on long documents. Asking it to first extract quotes and providing correctly answered examples boosts performance over 30%. The results offer guidance on rigorously evaluating different prompting approaches.
Microsoft Copilot’s Ecosystem Integration: AI Across Product Suite
Microsoft is integrating its AI assistant Copilot across Windows, Office 365, Bing, Edge and more. It aims to offer personalised, context-aware help. The enterprise version launches November 1st, while consumer products get AI search, shopping and productivity features.
McKinsey’s Generative AI Forecast: Reskilling for Economic Transitions
McKinsey predicts generative AI could automate nearly 10% of US economy tasks by 2030, disproportionately affecting lower-wage roles. However, new higher-wage jobs will also emerge. Individual and organisational adaptability through reskilling is critical to smooth transitions.
Bentley-Gallup Study on AI Job Impact: American Opinions
Most Americans think AI will reduce jobs in the next decade but can capably perform specific tasks. However, skepticism exists around businesses' ethical AI deployment. Communicating transparently about AI and fostering receptivity in younger demographics is key.
Salesforce Einstein Copilot: Customised Conversational AI
Salesforce launched Einstein Copilot, an AI assistant offering personalised, data-driven answers natively in Salesforce. Users can customise prompts, skills and models via Einstein Copilot Studio for tailored workflows. The Einstein Trust Layer architecture ensures data privacy.
Summaries from episode 28:
OpenAI Considers Custom AI Chips
OpenAI is exploring making its own AI chips to reduce costs, according to Reuters. OpenAI currently runs its models with Microsoft using Nvidia GPUs, but estimates suggest scaling to Google search volume would require $16B annually for maintenance. Developing proprietary hardware could aid OpenAI's ambitions for growth and scalability.
ChatGPT Vision's Expanding Applications
ChatGPT's new vision capability, enabled by GPT-4, is demonstrating diverse use cases like homework help, sports coaching, photography advice, diagram explanations, and more. As the tool launches more widely, we can expect users to uncover even more innovative applications of its interactive, multimodal features.
Meta's Generative AI Ad Tools
Meta is introducing generative AI into its ad products, using past data to help advertisers dynamically create ads. The tools aim to make ad content more personalised and effective. However, Meta acknowledges risks like biased outputs and will limit certain content types.
Canva’s Magic Studio and Compensation Program
Canva unveiled Magic Studio, upgrading tools with AI for transforming, translating, and generating designs. It also established a $200M fund to compensate creators whose work trains Canva's AI. This automated opt-in program provides transparency while improving Canva's models.
Zapier’s Canvas Optimise Workflows with AI
Zapier’s new Canvas maps workflows visually so users can see their processes end-to-end. AI then highlights areas to automate, pre-building integrations to streamline workflows. Canvas empowers users to optimise operations through process transparency and smart recommendations.
Understanding Neural Networks
Researchers from Anthropic developed a technique to decompose and interpret neural networks. By breaking down networks into understandable components, they aim to demystify how AI models function and make them more transparent. Their work could enable safer, more reliable AI.
Databox Uses AI to Summarise Marketing Data
Databox is testing using AI to generate text summaries analysing trends in marketing data. While not yet perfect, the AI reports provide quick, helpful overviews. Tool providers are likely racing to reliably productise similar capabilities to simplify data analysis.
Adobe’s Project Stardust Editing with AI
Adobe unveiled Project Stardust, which uses AI to automatically select and manipulate objects in images. Combined with generative fill, it streamlines tedious editing workflows. As the first integration of Adobe Sensei into Photoshop, Project Stardust exemplifies how AI will enhance creative tools.
CEO Replaces Team With AI Chatbot
A CEO claimed to entirely replace his customer service team with a ChatGPT bot, citing cost savings, instantaneity, and superior intelligence. However, users dispute the bot's competency, raising questions if the move was largely publicity. AI implementation requires nuance, even when headlines promise full displacement.
Summaries from episode 29:
Google Offers Legal Protection for Generative AI Users
Google Cloud now provides indemnification for its generative AI services, covering training data and generated outputs like text or images. This addresses enterprises' legal liability concerns, which have slowed adoption. While risks remain, Google aims to spur innovation by easing IP worries. However, the policy depends on responsible use and doesn't cover intentional infringement.
Skyscanner’s Conversational AI for Travel Discovery
Skyscanner launched an AI-powered travel inspiration chatbot using OpenAI's GPT technology. It gives tailored recommendations based on natural language questions about destinations, accommodating different traveler needs and interests. Skyscanner says many users visit for inspiration rather than transactions. The beta launch provides valuable data on engagement patterns before wider rollout.
Fine-Tuning Risks Undermining AI Safety
Studies found fine-tuning generative models like GPT-3.5 can easily undermine safety systems, even with well-intentioned tuning data. Models can rapidly learn harmful responses from just a few examples during tuning. This indicates a need for safety testing before and after customisation. As fine-tuning grows, maintaining protections will require vigilance.
CEO Interest in AI Increases From Skepticism to Integration
Per a survey, CEO adoption of AI doubled from 23% to 53% in under a year, showing increasing mainstream integration. But uptake and perceptions vary by industry, with most interest in areas like marketing so far. Despite growing openness, concerns around bugs, privacy, and trust remain. Leadership teams are still exploring optimal enterprise applications.
UK Lacks Resources for Global AI Leadership
A University of Cambridge report indicates the UK government's aspirations to lead in AI are constrained by insufficient capital and computing infrastructure compared to the US and China. Developing a system like ChatGPT alone could cost over $30 million per month. The UK is investing in an AI supercomputer, but it won't be operational until 2026. The report advocates the UK focus on grounded real-world applications, with incentives for commercial AI adoption and robust ethics frameworks.
ElevenLabs Offers AI Voice Dubbing
Voice cloning startup ElevenLabs launched a new AI-powered feature to dub videos into other languages. It provides easy voiceover translation, with potential for global content localisation. However, the dubbed audio lacks the lip sync accuracy of more advanced solutions. While quality trails rivals for now, usability and affordability can enable creative applications, with expectations for improvements as the technology matures.
OpenAI Shifts Core Values Towards AGI and Scale
OpenAI recently updated its core values on its website, shifting from attributes like "audacious" to goals like "AGI focus" and "scale." This signals a strategic emphasis on pursuing artificial general intelligence and rapid growth. While expected given OpenAI's trajectory, the change could impact internal culture and external perceptions. As values guide decisions, the evolution highlights OpenAI's aim to lead in transformative AI.
Interview: Olivia Gamblin, Founder of Ethical Intelligence
Olivia Gamblin, an ethicist and entrepreneur, discussed incorporating ethics into AI strategy. She helps leadership teams align decisions to company values through "ethics maturity" frameworks. Gamblin explained distilling core values from regulations, industry standards, and corporate principles. She advocates protecting or innovating based on different values. Gamblin highlighted marketing's role as trust builders and providing user feedback. She warned about "ethics washing" if actions don't match messaging. Overall, Gamblin outlined an ethical approach to AI drawing on her expertise.
Summaries from episode 30:
Diving Deep: Paul and Martin's AI Toolkit Revealed
The podcast hosts discussed their favourite AI tools and use cases, including several AI tools useful for marketing tasks:
- Magi - which provides a front-end interface for accessing multiple AI models like Claude and allows uploading documents for analysis. The hosts noted Magi's custom "personas" and security benefits of using the API.
- AudioPen - an audio transcription tool that utilises ChatGPT to format dictated notes into structured text. The hosts explained how this aids on-the-go note-taking.
- ClipDrop - praised for its suite of image generation and editing features like background removal and upscaling. The hosts described using it to easily create transparent logo PNGs.
- DALL-E 3 - with advanced text-to-image capabilities, though the hosts noted it was not yet surpassing certain other tools.
- Descript - an AI-powered video editor, was noted for offering easy transcript creation and editing. A host explained using it for quickly cleaning up poor quality audio.
Overall, the hosts emphasised first exploring ChatGPT Plus for its wide range of features including Dall-E image generation. They then suggested trying specialised tools for unique applications that build on ChatGPT's core capabilities. The hosts details revealed the expanding potential for marketing tasks. However, they noted the need to test many tools to determine true usefulness.
Jasper's End-to-End AI Copilot: Elevating Marketing Teams
Jasper AI unveiled major expansions beyond content creation, aiming to be a true marketing co-pilot. New features support strategy, analytics, team workflows, and voice/brand customisation. This shift targets benefits for entire marketing teams rather than just individuals. The updates position Jasper as an integrated AI assistant versus just a writing tool.
Universal Music vs. Anthropic: Navigating the AI-Generated Lyrics Quagmire
Universal Music accused Anthropic's Claude of replicating full lyrics when prompted, raising copyright concerns. The music industry is testing models' reproduction of protected works. However, legal complexities exist around publicly available data used for model training. This development highlights risks for brands using generative AI while laws evolve.
Invasion of Celebrity Clones: The Rise of AI Chatbots in Marketing
Prominent figures like authors and influencers are offering AI chatbots trained on their content. This allows customised advice on-demand. Meta is also launching celebrity bots on social platforms. As costs drop, expect more brands to explore digital doubles for marketing. However, reputation risks exist if bots behave poorly.
Visionary Tools: Harnessing ChatGPT's Visual Capabilities
ChatGPT's new vision API enables creative applications like website critiques, diagram explanations, and handwritten note transcription. Users are finding innovative ways to combine images and text prompts. As developers build on the API, we may see automated bulk analysis tools emerge. Vision unlocks new generative AI potential.
Delve into AI marketing through the "Artificially Intelligent Marketing" Podcast
In this blog post, you'll find a sneak peek into the valuable discussions led by Paul and Martin in their jam-packed podcast episodes. To immerse yourself fully in the fascinating realm of AI's role in marketing, we invite you to listen to the complete episodes, accessible either below or on your favourite podcast platform. Additionally, the archive of past episodes is a treasure trove of information on how AI is revolutionising marketing.
AI Note
The development of this blog post was augmented using AI technology. The process began with AI-driven transcription of the podcast audio, after which Claude from Anthropic and GPT-4 (ChatGPT) helped in refining the transcript into clear, concise blog segments. The BioStrata team further polished these segments for enhanced clarity and readability. Clipdrop AI software was used to create the blog's featured image.
