The Artificially Intelligent Marketing Podcast Episode 9: EU copyright laws, code interpreter plugin, and text-to-image generation

08 May 2023| by Paul Avery

The Artificially Intelligent Marketing Podcast Episode 9

As you may know, there is LOTS going on in the world of Artificial Intelligence (AI) at the moment, and much of it is very important for life science marketers to know about. That’s why our CEO Paul Avery has joined forces with fellow industry enthusiast, Martin Broadhurst, to sift through the AI news, carry out the research, and test the AI tools… so you don’t have to. They bring all this information to you each week in the Artificially Intelligent Marketing Podcast, released every Friday.

The ninth episode offers insights on a variety of subjects, including:

You can tune in to the episode below, or by subscribing on your favourite podcast platform. We've also provided a succinct summary of the podcast in the rest of this blog post, in case you prefer to read up on this topic rather than listen. You can also use the links above to skip ahead and navigate to specific sections of the blog.


Before delving into the episode's main stories, the podcast hosts Martin and Paul discussed various short topics that caught their attention in the realm of AI this week. Here's a summary below:


Story 1: EU proposes new copyright laws for AI

AI and its implications for copyright have been under scrutiny for some time. The European Union (EU) has proposed new copyright laws that particularly impact those of us in the marketing field looking to utilise generative AI tools to create brand content.

The EU's effort to regulate AI technology has been in the works for a couple of years. The new regulations come as a response to the rise of generative AI and pressing issues in areas such as copyright protection, user safety and user data processing. For example, AI technologies, including large language models and text-to-image generators, consume a vast amount of data for training… much of it may well be copyrighted. The EU will now mandate that creators who train models using copyrighted data disclose this fact.

This development offers a moderate approach compared to calls for an outright ban, which some have suggested. For content creators and marketers, it calls attention to copyright considerations when using generative AI tools. Wired magazine, for instance, vowed not to use text-to-image generation until copyright issues were resolved.

This stance by Wired may significantly impact the marketability of AI models trained on copyrighted data. It highlights the competitive advantage tools like Adobe's Firefly, which boasts non-infringement of copyright in its training data, could have. It leads us to think that companies with large repositories of non-copyrighted data, such as Shutterstock, Getty Images, and Adobe, could create more compliant AI models.

While smaller businesses may risk using tools that don't limit copyrighted content in their training data, large brands might opt for less risky routes, even if it means compromising on the quality of generated images. 

An interesting development in this space is the approach taken by Grimes, the pop artist. Grimes has expressed that fans are free to create music using AI versions of her voice, offering to share any resulting revenue 50-50. Innovative business models like this could influence how copyright laws evolve in the AI era. It's a development worth watching closely.

Story 2: OpenAI releases code interpreter plugin for ChatGPT

OpenAI has launched a fascinating new plugin for ChatGPT called Code Interpreter. This tool allows users to analyse and interpret data using only natural language prompts. If you've been eager to get your hands on such plugins, the wait could be worthwhile as this one has been described as being capable of delivering the value equivalent to that of a junior-level data analyst.

The operation of this plugin is fairly straightforward. You simply upload your data file in CSV format, then ask ChatGPT to look within the data for trends and insights. The tool can even plot charts and visualise data based on your queries. This is a departure from traditional data analysis tools that require proficiency in SQL and other data manipulation techniques. With this plugin, you upload your data, pose a question, and get an answer. You could even request a report. It truly brings the power of AI and automation to your fingertips, making data analysis much easier and faster (even for non-technically trained people).

One fascinating example of the plugin in action that we spotted involved a user asking the Code Interpreter to extract key trends from a large dataset and then write an abstract for a paper based on the insights. The paper was subsequently drafted using ChatGPT as well!

For marketers, particularly those without any expertise in data manipulation using SQL or other coding languages, this plugin could be a game-changer. It has the potential to democratise data analysis, making it more accessible, especially for small to mid-size businesses that might not have the capacity or resources to analyse their data effectively (e.g. hire a data analysis team). The Code Interpreter plugin can make this much easier, by essentially providing a virtual junior data analyst, making it possible to uncover valuable insights in your data.

While the promise of the Code Interpreter plugin is thrilling, it's also crucial to be mindful of potential pitfalls. A significant consideration with AI tools like ChatGPT and this plugin is the possibility of AI hallucination – will it identify trends that aren't truly present? The reality is that while this plugin might aid senior data analysts by boosting their productivity, it may not fully empower non-developer data analysts as much as one might hope. We’ll have to wait and see until we can get access to the plugin…

Another thing to keep in mind is that being proficient at asking great questions about your data is a skill that's still vital, even with AI assistance… a tool is only as good as the questions you ask it. What’s more, your ability to probe deeper into the 'why' behind the trends and insights could be the key to extracting the most value from your data. In this regard, having experts that can effectively pilot these tools will still be indispensable.

Story 3: Stability's DeepFloyd IF brings text-to-image generation with readable text

Stability AI has unveiled DeepFloyd IF, a new tool that promises high-quality image generation from text prompts, including legible and sensible text in images, an issue that has plagued most existing tools so far. Stability AI's launch video showcases impressive examples of shop signage and graffiti. 

Interestingly, the model deviates from the common diffusion technique used in previous image generation models in order to deliver better text-rendering results. Instead, DeepFloyd IF creates a smaller, 64x64 pixel image, and then scales it up. This different approach is beneficial in generating legible text as part of the image. 

For marketers, the introduction of text into images can significantly enhance the practical utility of text-to-image generation tools. One can easily envision the integration of this tool into platforms like Canva, enabling marketers to seamlessly utilise this technology. The anticipation is for Stability AI to offer a user-friendly interface, possibly through tools like Clip Drop, making the technology easily accessible.

The early iterations of DeepFloyd IF, despite being fascinating and promising, are still in the nascent stages. Yet, the progress made in addressing the text artefact issue in image generation is a significant stride. The possibilities for applications in areas such as logo generation, product imagery, advertising campaigns etc. are endless, and the tool promises to bring a wealth of creativity and utility to the marketing world.

Story 4: Anime film trailer made using text-to-video generation

A creator on Twitter recently shared an impressive project: an anime movie trailer created entirely through RunwayML's GEN2. Despite GEN2 being only a few weeks old at the time, the creator was able to produce a two-minute trailer that truly showcased the tool's capabilities.


RunwayML's GEN1 was essentially a video stylisation tool for augmenting existing video clips. But with GEN2, the game has changed, shifting towards text-based video generation. Although the generated anime trailer was not without its artefacts (for instance, a car that disappears into the road, and a character with only one leg), it's a remarkable demonstration of the potential of these tools, especially considering they have only been available for a few months at most.

This leap in progress from mildly photorealistic images with some abnormalities to a point where an entire anime movie trailer can be produced is truly awe-inspiring. It signifies the exponential speed of development in this technology, posing exciting opportunities for marketers and content creators.

With the continuous advancements in text-to-video technology, marketers can now consider producing video content that might have been too time-consuming or expensive before. This can especially be beneficial for animated content or B-roll footage for social media posts.

However, as these tools become more mainstream, it's the creativity of the idea and its ability to capture the audience's attention that will truly set marketers apart. It’s also worth noting that text-to-image and text-to-video tools are not a one-and-done deal, but each output requires constant iteration, tweaking, and refining to get to what you want – the role of the AI artist, if you will.

Tool of the week: Bing AI in the Microsoft Edge browser

This week, Paul wanted to share an interesting discovery that has the potential to greatly influence how he navigates the web and performs his daily work. Though it's not particularly new (it came out around mid-March), Paul has been trying out Bing's AI chatbot in the sidebar of Microsoft's Edge browser, and has been intrigued by the functionality it offers. 

Long-time Chrome users may be initially hesitant to test Edge's waters, but Bing's chatbot could be a game changer. Tucked neatly in the sidebar, the chatbot's capabilities are wide-ranging: generating text, creating content, writing blog posts, and summarising documents, all powered by ChatGPT-based AI. Combined with Edge's browsing prowess, the user experience is significantly enhanced.

Bing's AI sidebar stands out when integrated into a workflow. By opening documents in Google Drive or OneDrive directly in the browser, selected text can be quickly pushed to the Bing sidebar. From here, the AI can summarise, explain, revise, or expand the text. As a glimpse of Microsoft's Copilot in action, this functionality could be a vital tool for regular document or email editors within a browser.

The tool does have its limitations - it is text-centric and lacks data management capabilities or substantial interaction with PowerPoint or Excel. While it can generate images, this feature seems restricted to the primary browser version of the Edge environment.

Persuading users to transition between ecosystems is a formidable challenge. Yet, Bing's AI in Edge offers such added value that it could warrant abandoning an established Chrome setup. Given the potential security risks associated with third-party Chrome plugins, Microsoft's commitment to security and data management presents a compelling alternative.

After a few weeks of experimentation with Bing AI in Edge, the convenience of having AI features in the browser's sidebar is apparent. The AI outputs high-quality results and, backed by GPT-4, it provides user-friendly, UI-focused tools like a chat mode and a compose mode for drafting blog posts.

The full capabilities of the compose mode are yet to be thoroughly explored, but preliminary tests show it can generate a blog post of approximately 700 to 800 words. This limit could prove restrictive for those with larger text-generation needs, who may still require standalone chatbots like ChatGPT GPT-4.

How to subscribe to the Artificially Intelligent Marketing Podcast

We hope you found value in this blog post. However, it merely skims the depth of what Paul and Martin delve into in their podcast! So please check out the entire episode at no cost here, or simply locate it on your go-to podcast platform. Additionally, our library of past episodes is readily available for your listening pleasure here.

Listen to the podcast episode

AI disclaimer

The creation of this blog post was supported using AI to transcribe the original podcast audio, and GPT-4 (ChatGPT) to summarise the transcript into blog sections. These sections were then manually curated and edited by one of the human members of the BioStrata team. The blog image was created with the help of AI using Midjourney.