Generative AI For Business
At the end of 2022 I wrote that the story that ChatGPT was a "Code Red" for Google's search business, seemed overwrought.
After all, "Google could choose to develop an advanced version of ChatGPT, offer it as a search interface option, and monetize it."
That's essentially what they did in May 2023, announcing the new SGE (Search Generative Experience) at the annual Google I/O conference.
In the process they also provided the first major change to Google Search's visual interface in many years, not counting the continual evolution of layout features. That secondary effect was a good idea. ChatGPT's interface was sleek and evoked proximity and one-on-one conversation. By comparison, Google Search began to seem a bit stale.
In February of this year Bing rushed to incorporate ChatGPT in its search interface.
ChatGPT 4 and the Plugin Marketplace
Since then we've seen the release of GPT-4 and the marketplace of ChatGPT plugins that extend the chatbot's functionality.
You have to upgrade to ChatGPT Plus ($20/month) to gain access to the marketplace. We've tried out plenty of the content/marketing plugins. In our view, they're nearly all (except for maybe 1 or 2) disappointing and useless.
We found one that does a decent job of creating social media blurbs if you feed it a prompt like: "Create a short social media blurb to promote an article about (topic)." It will return a lively blurb with relevant emojis and hashtags.
But if you refine your prompt ("add relevant emojis and hashtags"), and skip the plugin, you'll get the same result.
In other words, some of these plugins seem to have simply crafted good prompts which get triggered by keywords like "social media blurb."
We liked 2 of the apps that will summarize PDFs. But Anthropic's Claude does a stellar job at this and it's free. So why pay $20/month?
Lastly we found a ChatGPT app that did an excellent job summarizing a video, then it proceeded to hallucinate repeatedly. A few weeks later we tried it again and got successive error messages.
There are also apps that will create diagrams and charts. But why? The results are ugly and low-res. You can get better charts for free from Excel or Google Sheets or from any number of low-cost solutions like Canva. In addition, interactive data visualization tools, like Tableau, have been on the market for years.
The ChatGPT marketplace seems immature, as would be expected at this point. In some ways, it's reminiscent of the Alexa Skills marketplace, which quickly filled up with clever apps that no one needed.
Start by identifying problems worth solving in your business.
Note: ChatGPT Plus users get access to GPT-3.5 and ChatGPT-4. The latter is slow, understandably, because it's used for more advanced calculations or has to interface with plugins. (Plugins can only be used with GPT-4.)
AI has been used in marketing for decades now. To see how or if you might benefit from the new crop of generative AI tools, you'll need to dive beneath the surface chatter.
First, be clear on what problems are worth solving in your business and what opportunities you're seeking. Then as you learn about the capabilities of generative AI, you'll be prepared to sift that information.
Generative AI in Business and Marketing
For starters, there are opportunities to deploy LLMs in video translation, communications, analytics, categorization and summaries of documents and files, and customer support.
Here are some examples.
- Synchronicity Labs has used ChatGPT and other technologies to create what appears to be a superior video translation product (in closed beta). This tool generates the translation in the voice of the original speaker, dubs it to the video while maintaining any other audio tracks, like background music or sound effects, and convincingly lip syncs the translation to match the movement of the speaker's mouth. This tool has potential widespread application in localization, as well as in the film industry, education, government, corporations, and regional broadcasting. (It could eventually upend the film dubbing business.)
- ElevenLabs generates quality audio from text, and can do it in any voice or language, including your own. (Synchronicity Labs uses ElevenLabs in their tech stack.)
- Those with massive document troves to summarize will want to check out Anthropic's Claude. This tool can also be used to summarize PDFs, which can save research time. (Rather than reading 50+ pages a researcher can get a succinct and accurate summary and decide whether to read the entire document.)
- We've been impressed with Assembly.ai's pristine transcripts of video and audio files. Also see LeMUR, part of Assembly.ai's API. LeMUR is used to build apps on voice data. It's not surprising that CallRail jumped at the opportunity to build with LeMUR.
- An example of an app built on voice data is Fireflies.ai. It records and transcribes meetings, organizing content by sections and filters like "Questions" and "Tasks." You can voice-trigger tasks to be set up in project management platforms like Trello. Fireflies can also send out summaries and log call notes under relevant contacts in a CRM. Interestingly, some business users indicate Fireflies, by improving communication, is shortening their sales pipeline.
AI in general will continue to make it possible to refine processes in new ways, increase automation, and build resilience against unexpected factors.
Revisiting the Hype
As we mentioned in our last article, the hype surrounding ChatGPT's release was intense. This was particularly notable in content/marketing applications.
Social media was filled with claims you could get rich quick by having AI write all your articles and video scripts.
As I noted elsewhere, if even your pet parrot can peck a button and get an insta-article, why would the marketplace value it?
Nevertheless, prominent SEO pros reacted to ChatGPT with gloomy warnings on the future of the profession.
SEO experts amplified the mania by writing articles cataloging everything ChatGPT could do. For example, it could sort and categorize keywords by search intent. It could generate sitemaps, robots.txt files, and schema markup. Yes, but who does these things by hand anymore anyway?
Case in point: For years we've had tools that could not only generate schema but insert the code in exactly the right place with zero errors. And every marketing professional already has a keyword research tool like Semrush (founded 15 years ago) or Ahrefs (founded over a decade ago).
Next, top SEO publications heralded every marketing/SEO app in the ChatGPT marketplace. That's a clever tactic for ranking high for "ChatGPT apps" but these advertorials have low information value.
Soon these publications will chastise you for trend chasing. And they'll offer advice on avoiding this.
Every 6 months the tech industry enters a brief episode of mass collective hallucination where everyone is ardently convinced that an epochal tech revolution has just begun, that a "civilizational" technology is starting to take over. Remember web3? Remember LK-99? Remember LLMs?
When asked whether he thought LLMs were not transformative, Chollet replied:
LLMs are fine, but they aren't anywhere within 100 light years of what they were touted to be by your average tech industry NPC (a civilizational technology, the beginning of AGI, about to automate all jobs, obsoleting software engineering, etc.). It would be like saying "the metaverse hype was legit because there's a niche market for VR headsets to play games". None of what was forecasted actually happened.
Of course, a tool or action that improves your processes by 1% can be transformative in scaling up your business, thanks to the aggregation of marginal gains.
And tools like Synchronicity Lab's video translation, will no doubt attract those seeking acquisitions.
These are two reasons why business owners, aware of the possibilities, are scouring generative AI for possibilities.
Lastly, just because there is a lot of hype surrounding an emergent technology doesn't mean it won't be consequential, although hurdles to tranformative impact remain.