The following review is based on my actual experiences using ChatGPT 4 Plus as a marketing professional and writer. These observations are based on factual outputs and offer little futurist conjecture about how ChatGPT. I see no value in adding to the overwhelming ChatGPT hype wave.
However, I do offer pragmatic suggestions for marketers and writers who want to use the tool in its current incarnation based on my experiences. This should be more useful.
Obviously, ChatGPT contributed significant parts of this post, specifically the examples. If you just want to read the takeaways scroll past the examples.
1) Great Outcomes from Prompting with Source Data
First, when prompted to write in a specific style or a type of business document, the new ChatGPT Plus 4 is more concise and better than its predecessor. Add in some good messaging and source material ChatGPT Plus 4 is phenomenal at taking pre-drafted messaging and writing tactical executions from it.
I have taken messaging generated for a campaign, and have used the platform to create several executions, including targeted emails, email variants, formal language, web copy, and an article. Then I edit the documents for style, extra phrases (more on that in a moment), and awkward generative AI Frankensteining of phrases.
I tell it to write a specific type of output for a business function. Others use the “Act as” prompt to get it to write for a business function. For example, “Act as a tech start CTO.”
Usually, this takes about 15–20 minutes for a 250–500 word document. The more directive my prompting is, the better a job ChatGPT Plus 4 does. Understanding how to prompt directly impacts your outputs.
I would share an example, but these have been created for my employer. Apologies.
Takeaway: ChatGPT is good at taking already created messaging and concepts to inform iterations of text drafts. But you must prompt the machine well.
2) The Killer App? Summarizing Long Text
I find ChatGPT to be most helpful when it is used to summarize long-form text. Some use cases include summarizing social media posts or providing an executive summary of the content that gets to the point with less hyperbole.
Let’s show you how this works by summarizing Accenture’s recently released generative AI report.
Me: Please summarize this report on generative AI. Give me five takeaways in under 150 words. https://www.accenture.com/us-en/insights/technology/generative-ai
1) Generative AI is a rapidly emerging technology with the potential to revolutionize industries by synthesizing novel content, designs, and solutions. It utilizes advanced algorithms to learn patterns and generate new, previously unseen data based on those patterns.
2) The technology has broad applications, including creating realistic images, designing innovative products, optimizing supply chains, and generating human-like text. Industries like healthcare, manufacturing, and entertainment can greatly benefit from these capabilities.
3) To successfully implement generative AI, organizations should focus on three key factors: data quality, model selection, and a strong ethical framework. Ensuring access to high-quality data and choosing the right model will maximize the technology’s potential.
4) Organizations need to address ethical concerns related to generative AI, such as copyright infringement, privacy, and the spread of misinformation. Implementing transparent AI practices and following guidelines can mitigate these risks.
5) Companies should leverage generative AI to drive innovation, improve decision-making, and enhance customer experiences. By embracing the technology responsibly and ethically, organizations can unlock its transformative potential and remain competitive in the market.
Takeaways: 1) If you cannot or do not want to spend time reading and summarizing long text ChatGPT is a killer app. It is ideal for folks that need to create short summaries, and for people who want to understand elements of a topic without becoming a subject matter expert.
2) I can see why the academic community might find this application to be disturbing, particularly with GPT 4’s ability to consume 25,000 words in one prompt. So much for an honest book report. However, I have had issues prompting ChatGPT Plus with the latter word count so perhaps this is a coming feature.
3) Training Data Outputs Are Still Suspicious
ChatGPT Plus 4 still does not have better training data. Whatever web source material it uses from prior to September, 2021, still has not been verified, and the engine does not show you those sources.
At least it offers some disclaimer as the below output shows when I asked it who Geoff Livingston was. However, I have never published a photography book called “The District”. But I did publish a photo book that was primarily photographed in Washington, D.C. Close, but no cigar.
“As of my knowledge cutoff in September 2021, Geoff Livingston is a digital marketing and communications strategist, author, and photographer. He has worked with various organizations and clients to develop and implement marketing campaigns, primarily focusing on social media, content marketing, and online communications. Livingston has written several books, including “Now Is Gone,” “Welcome to the Fifth Estate,” and “Marketing in the Round,” which offer insights into the world of digital marketing, social media, and public relations. In addition to his marketing work, Geoff is also a passionate photographer and has published a photography book called ‘The District.’
“Please note that information about individuals may change over time, so it’s always a good idea to verify the most current information available.”
Takeaways: 1) ChatGPT 4 (just like the free ChatGPT 3.5 version) cannot be trusted as an accurate source of information. Any outputs generated by the bot need to be verified through additional searches and/or a subject matter expert.
2) This is the primary reason why ChatGPT integrations into sales CRMs are so dangerous. Untrained sales and marketing professionals cutting and pasting unfiltered ChatGPT text is a recipe for disaster.
4) ChatGPT 4 Does Not Provide Most of Its Sources
When prompted, ChatGPT 4 does not provide web URLs for its sources. So even when you dig in, there are issues.
ChatGPT 4: As an AI language model, I do not have the capability to directly browse the web or show you the specific sources in real-time. My knowledge is based on the data I was trained on, which includes a wide range of online sources up to September 2021. This includes websites, articles, and other text-based sources that discuss Geoff Livingston and his work. To find the most current information and sources, I recommend conducting an online search for “Geoff Livingston,” which will provide you with the most up-to-date information on his work and accomplishments.
However, it can cite popular sources that are widely known. Its ability to cite a source is on the most general level.
ChatGPT 4: As an AI language model, I can provide information and sources that I have been trained on, up until my knowledge cutoff date in September 2021. I can mention well-known sources, books, research papers, or websites that are relevant to a particular topic. However, I cannot provide real-time or specific URLs for any given piece of information.
Figuring, it would be useful to get an example, I asked it to show me how it cites a source.
ChatGPT 4: Certainly! Let’s consider a discussion about climate change. One of the most well-known sources for information on this topic is the Intergovernmental Panel on Climate Change (IPCC)… Their Fifth Assessment Report (AR5), released in stages between 2013 and 2014, is often cited when discussing the scientific consensus on climate change and its impacts on our planet.
Takeaways: 1) Do what the ChatGPT bot tells you to do. Conduct your own online search.
2) Also, if you don’t have source material and are looking to understand a topic Wikipedia offers better general summaries because it sources its information. I have often thought ChatGPT endangered Wikipedia more than any other web service. Wikipedia is safe, for now.
3) Until OpenAI overcomes this issue and sources web data in real-time in a trusted manner, I see its overall usefulness as limited to supporting research summaries, starter text for articles and other tactical campaign writing, and very tailored writing to topics its training data is actually accurate on.
5) It can mimic prompted styles (including your own)
One of the biggest knocks on ChagtGPT is its, well, ChatGPT sound prose, which is often academic. My podcasting buddy Greg Verdino pointed this out to me with a unique experiment (cough). So I tried it using one of the photography blogs.
Me: Can you write an 500 word article about shooting cherry blossoms mimicking my style here: https://geofflivingston.com/2022/05/14/photo-essay-3-days-in-big-bend-national-park/
The result was somewhat close. Close enough that I could edit to sound like me in under 15 minutes. That’s not bad at all. There was one paragraph I found myself nodding to…
ChatGPT 4: When it comes to camera settings, there are a few key considerations to keep in mind. For one, you’ll want to use a shallow depth of field to create a dreamy, ethereal effect. This can be achieved by shooting at a wide aperture, such as f/2.8 or f/1.8. Additionally, you’ll want to use a fast shutter speed to freeze any movement in the blossoms or the surrounding environment.
6) But its native voice still sounds academic
My biggest knock on ChatGPT’s native writing style is its overwhelmingly academic windiness. I immediately suspect ChatGPT written prose when I see extra phrases, formal tone, unnecessary conjunctions, hyperbole, and adverb abuse. It’s like a bunch of TAs from Stanford trained it. Whoops.
Here are a couple of examples from the previous prompt, illustrating my point.
ChatGPT 4: The beauty of cherry blossoms is truly a sight to behold. These delicate, pink blooms bring a sense of joy and renewal to the world, and capturing their beauty in photographs can be a deeply rewarding experience. In this article, we will explore the art of shooting cherry blossoms and offer tips and tricks for capturing the perfect shot…
In conclusion, shooting cherry blossoms is an art form that requires a combination of technical skill and artistic vision. By carefully considering your settings, exploring different angles, and exercising patience and persistence, you can capture truly breathtaking images that celebrate the beauty of these delicate blooms. So go out and explore your world with your camera in hand, and see what beautiful images you can create!
I did try to coach it to create text with less embellishment.
Me: Can you cut the hyperbole and extra phrases?
ChatGPT 4: Shooting cherry blossoms can be a rewarding experience. Plan your shoot carefully and find interesting compositions. Use a shallow depth of field, fast shutter speed, and polarizing filter to capture the blossoms’ beauty. Experiment with different angles and be patient to capture the perfect shot.
And the pendulum swings completely in the opposite direction. I could probably coach it to produce the right output, but it would probably take about the same amount of time to simply write it or edit the actual output. Let’s call this a writer replacement fail.
Takeaways: 1) Advertising, journalism, and good content marketing featured to-the-point crisp text, short paragraphs, and snappy headlines. ChatGPT needs at least another version to generate raw text like that.
2) Writers and content creators who want to maintain a style or original voice should look at ChatGPT outputs as summarizing key points or rough drafts that they can then hone. If you want punchy, the last mile is still in the writer’s hands.
7) Visual prompting, hmmm.
One of the biggest promises of GPT 4 is its ability to take visual queues. As of this week, OpenAI has not released the algorithms potential in the ChatGPT Plus app, supposedly. For example, you cannot upload a file to ChatGPT Plus 4. That didn’t stop me from prompting it.
Me: Is this a good photo? https://www.flickr.com/photos/geoliv/52755570262
ChatGPT 4: As an AI language model, I cannot provide a subjective opinion about whether a photo is good or not. However, I can describe the photo you linked to on Flickr. The photo depicts a beautiful landscape shot of a mountain range with a rainbow in the background. The image is taken from a distance with the mountain range in the foreground and the rainbow in the background. The colors are vibrant, and the composition is well-balanced. The image has been well-exposed, with the foreground, mid-ground, and background all well-lit. Overall, the photo appears to be well-captured and visually striking.
This seems like a good review, except for one thing… The photo was taken in the heart of London, at night.
Takeaways: Yeah. I am not a big fan of OpenAI’s DALL.E image generator, and this output makes me skeptical about GPT 4’s visual prompting capabilities. Happy to be wrong about that, but something tells me computer vision algorithms are not OpenAI’s strongest offering.
All in all, GPT 4 offers ChatGPT more concise and better writing capabilities. As we have not seen all of the algorithm’s promised features yet, I cannot say how impactful ChatGPT Plus 4 will be. However, the current iteration is an iterative improvement.
I feel good about subscribing to and using ChatGPT Plus as a result of this new algorithm. This would not have been the case with 3.5, and its less concise and more erroneous outputs.