-2.svg){kind=logo}
The rapid pace of AI development continued through the final month of 2023, with major new releases coming from tech giants and startups alike. This monthly roundup summarises key insights from the latest three episodes of the "Artificially Intelligent Marketing" podcast. Episode 35, episode 36 and episode 37 covered AI highlights including Google's new ChatGPT rival Gemini, Meta's image generation leap with Imagine, and the reopening of paid ChatGPT Plus subscriptions.
These episodes also explored emerging best practices around optimising AI through techniques like prompt engineering. As the hosts discussed, crafting high-quality prompts is proving crucial to steering advanced models like GPT-4 and reliably extracting their potential. On the application side, integrations with Google Cloud and Instagram demonstrate how combining AI with existing data and interfaces unlocks more transformational capabilities.
However, the shows also covered controversies around manipulated demo materials from Google and the evolving legal landscape as artists revive copyright lawsuits. As AI proliferates across marketing and beyond, thoroughly evaluating new technologies and considering responsible adoption remain critical. Listen to the full episodes of 35, 36 and 37 below for Martin and Paul's in-depth analysis on these AI developments and their implications.
With generative AI in a stage of rapid iteration, staying up to date through resources like this podcast continues providing valuable perspective. This monthly summary aims to highlight the key topics for marketers to keep pace with the AI landscape as we enter 2024.
Summaries from episode 35:
ChatGPT's Momentous First Year
ChatGPT by OpenAI reached 100 million users just two months after launching in November 2022, compared to years for other platforms like Facebook and Instagram. It launched as a free research preview, but quickly grew in popularity. Key milestones include launching ChatGPT Plus in February 2023, powering features in Microsoft Bing, announcing GPT-4 in March, and expanding capabilities like voice, image generation and APIs. However, it has experienced challenges like outages, bans in Italy, and the firing/rehiring of CEO Sam Altman.
Synthetic Media Company Unveils Instant Digital Avatar
Startup Hey Gen has raised $5.6 million and launched new near instant custom avatars. Previously creating avatars required filming oneself in a professional studio, but improvements to their algorithm now allow avatars to be made from mobile phone quality video. This expands access to their technology that can create synthetic versions of users by writing scripts and having the avatars speak them.
Stable Diffusion Offshoot Enables Real-Time Image Generation
Stable Diffusion XL Turbo by Stability AI can now generate images in real-time as users type text prompts. This is enabled by more efficient processing, going from 20-50 steps previously to a single step output. It is currently available in a research preview, with potential applications in creative tools and interactive experiences.
Hype Builds Around New Generative Video Startup
AI video generation startup Pika has unveiled impressive technology in demos, touting the ability to create high-quality animated videos from text prompts. They have raised $55 million as they prepare to open up access to users. If the actual tool lives up to the demo, it could significantly advance text-to-video capabilities.
AI Excels at Complex Strategy Game Requiring Human Negotiation
Meta AI's Cicero has achieved human-level play in the strategy game Diplomacy. Impressively, it combines strategic thinking with natural language processing to cooperate and negotiate with humans. It ranked in the top 10% of players on Diplomacy websites, showing ability to model complex human psychology and diplomacy.
Music Generation System Becomes More Controllable and Shareable
Stable Diffusion's audio generation tool Stable Audio received major updates, including input audio capabilities and more editing parameters. This allows greater control over existing creations, and additions like sharing tools aim to improve collaborating on generatively created music.
Artists Revive Copyright Lawsuit Against AI Firms, Citing Damages
Visual artists have amended their lawsuit alleging AI image generators misused their work, now citing search result dominance of AI art as evidence of commercial harm. This revives their claim after earlier arguments were dismissed, evolving the approach artists are taking in attempting to hold AI firms liable for damages.
Summaries from episode 36:
In this intriguing episode host Martin Broadhurst welcomes Ankur Pandey, CEO and Co-founder of Longshot.ai, an AI-driven content marketing suite offering a range of tools from content generation to optimisation. Ankur, an experienced entrepreneur in the AI space, shares valuable insights on the evolution of content marketing influenced by AI advancements. Below is a summary of this episode.
Adoption and Acceptance of AI in Content Marketing
Ankur reflects on the journey of AI in content marketing, highlighting the shift from skepticism to widespread acceptance and implementation by content marketers. He notes the overwhelming acceptance compared to a couple years ago when marketers were more skeptical due to overpromises from AI tools. Ankur says AI is now universally used in workflows by serious content marketers.
AI and Search Engine Optimisation (SEO)
The conversation delves into the integration of AI with SEO strategies, particularly considering Google's helpful content update and EAT guidelines. Ankur discusses how Longshot.ai aligns with these developments. He notes reduced confusion around AI content and SEO, with content quality mattering more than it being AI-generated.
Longshot.ai's Features and Functionality
Ankur provides an overview of Longshot.ai, emphasising its capabilities in producing long-form content, managing brand voice, and optimising content for SEO, plagiarism, and readability. He highlights features like workflows to create human-like, on-brand content.
FactGPT and Innovations in AI Content Creation
The episode explores Longshot.ai’s FactGPT feature, which addresses challenges in creating recent, factual, and context-specific content. Ankur explains FactGPT's ability to incorporate user data and focus content on trusted websites, motivated by users' needs.
AI Technology Choices and Development Strategies
Ankur talks about the decision-making process in selecting and integrating various AI models and technologies, including OpenAI and other homegrown solutions. He details Longshot.ai's testing procedures to determine the best AI for specific use cases.
Future Trends in Content Marketing and AI
The discussion turns to future possibilities in AI-driven content marketing, such as real-time content personalisation and the evolving nature of brand voice. Ankur anticipates innovations in user interaction with AI tools beyond improvements in the technology itself.
Advice for Content Teams Utilising AI Tools
Ankur shares practical advice for content teams, stressing the importance of understanding the nuances of AI tools and their integration into content marketing workflows. He advises examining priorities like veracity and brand voice when selecting tools.
Closing Thoughts
Ankur encourages content marketers to embrace AI advancements, emphasising the opportunity for growth and innovation in the field. He remains optimistic about the future of AI in content creation.
Summaries from episode 37:
Google's Gemini: The Challenger to GPT-4's Throne
Google recently unveiled its new generative AI model Gemini, which comes in three sizes - Nano, Thro, and Ultra. Ultra is said to outperform GPT-3 on several benchmarks, though some critics have questioned the benchmark evaluations. Thro is now available via API and delivers capabilities on par with GPT-3.5. While impressive, Gemini's launch video appears to have used manipulated prompts, damaging credibility. Overall, Gemini seems highly capable but may not exceed GPT-4 abilities. Its true multi-modal design could enable innovative applications when released in 2024.
Unlocking AI's Potential: Prompt Engineering
Microsoft research reveals prompt engineering can significantly enhance LLMs like GPT-4. Techniques like "chain of thought" prompting, dynamic few-shot prompting, and repeated prompting boost accuracy and reliability. A prompt strategy called Medprompt, originally for medical challenges, achieved record scores on diverse academic tests. Though current models are limited, prompt crafting allows more control in steering AI capabilities. As models advance, quality prompting will remain key to realising their potential.
Enhancing AI Recall: Claude 2.1
Tests by Anthropic expose Claude 2.1's inconsistent recall from its vast 200k context window. On a "needle in haystack" test, its accuracy was only 27%. However, simply prompting Claude to start with the most relevant sentence raised performance to 98%. This shows the dramatic impact prompt adjustments can have on reliability, despite advanced context windows. It also demonstrates Claude's strength with summarisation, if properly prompted.
Meta's AI Image Generation Leap
Meta launched Imagine, an AI image generator trained on 1.1 billion Instagram and Facebook images. It creates unique pictures from text prompts, currently only in the US. Imagine appears highly capable for a first release, though lacks true photorealism. Meta also unveiled Backdrop on Instagram, which removes backgrounds using its Segmentation Model. Overall, Meta is rapidly deploying its research into innovative consumer applications.
Instagram's AI-Powered Backdrop
Instagram launched a new feature called Backdrop that removes backgrounds from photos using Meta's Segmentation Model. Users can replace backgrounds with anything they want. This demonstrates Meta's push to bring its latest AI research into innovative consumer offerings.
ChatGPT vs Human Expertise: A Study
An MIT study found ChatGPT generated more persuasive creative content than humans alone for products and campaigns. Participants saw higher quality when told content was human-made, though AI inclusion didn't lower perceptions. For both products and messages, ChatGPT's quality surpassed humans, especially when used together. While top experts likely still exceed AI capabilities, this demonstrates advanced generative AI can outperform average human marketers.
McDonald's Tech Upgrade with Google Cloud
McDonald's entered a partnership with Google Cloud, aiming to apply generative AI across business priorities. With McDonald's customer data and digital interfaces combined with Google Cloud, personalised interactions via AI could significantly boost metrics like order size. Advanced applications of AI likely offer more transformative potential than standalone uses.
Mistral: Open-Source AI Advancement
Startup Mistral released an open-source AI model called Mistral 8x7B. It uses fewer parameters than GPT-3.5 yet beats it on benchmarks, while incorporating a mixture-of-experts architecture. Its open-source nature provides full user control, addressing data privacy concerns. As it demonstrates, open source models are rapidly advancing, offering innovative capabilities without restrictions of commercial LLMs.
ChatGPT Plus Subscriptions Reopen
After high demand kept the waitlist long, OpenAI reopened paid ChatGPT Plus subscriptions in December 2023. For those still eager to access the ad-free, priority version, spots are now open again. As new AI products launch at incredible speed, staying patient amid long waits is often necessary.
Explore AI in Marketing with the "Artificially Intelligent Marketing" Podcast
Discover key insights from Paul and Martin's dynamic podcast episodes in this article, offering a glimpse into how they navigate the intersection of AI and marketing. To dive deeper into the captivating world of AI's impact on marketing strategies, you're encouraged to listen to the full episodes, available here or on your preferred podcast platform. Moreover, an extensive backlog of episodes serves as a rich resource, showcasing AI's transformative role in marketing.
AI Note
The creation of this blog post was enhanced with AI tools. It started with an AI-assisted transcription of the podcast, followed by the collaborative efforts of Claude from Anthropic and GPT-4 (ChatGPT) to refine the transcript into succinct, engaging blog content. The BioStrata team further refined these contents for greater clarity and appeal. Additionally, the Clipdrop AI software played a crucial role in developing the blog's eye-catching featured image.
