In Oct 2022, I wrote a post on 30 areas AI is taking over. Little did I know that everything I thought would be applied is even simpler when we enter into a narrative economy with #chatgpt.
For those who haven’t played around with it, the gpt3.5 version manifested in a chatbot with a free open playground shows the next phases of Google Search and a clear path towards science fiction becoming a reality. There are an incredibly wide range of use cases that have been posted all over twitter using #chatgpt.
My biggest observations and personal wonderings:
Conversational – The conversational aspect and memory of the interactions with the account is incredible. You basically have a JARVIS at your fingertips (minus the sentience, for now). You can keep refining results and prompting for more specific outcomes.
Party trick – I logged in and asked a friend to give me the most ridiculous prompt. We decided on “Write a horror movie script at a burger joint with a girl named Sam.” We were eating at the burger place and the waitress’ name was Sam. I proceeded to read the script to her and she was aghast.
OpenAI Credits 10000x – The total addressable market for use cases leveraging this technology is really just everything. If a token launched in the public space tomorrow with a $1 per request fee, the whole market cap would be flooded with mass orders. If it’s a fixed price (for now) then it’ll reach probably reach mass adoption with augmentations manifesting in other tools. It it becomes a bidding war open market, then all enterprises would buy millions of requests and figure out how to implement it into their tech stack for the next few years.
Death of a high class workforce – If my workforce is currently doing stack overflow searches to write code only a marginally better (or even worse) than a query using GPT3 then your next high value company could just be a team of 10. Imagine all C-suites and no grunt software devs and marketing people?
Consultancy as a Service – All consultant costs go to zero. I wrote a list of 30 costs to zero discussing how the history of our advancements are created from more automation and less manual intervention. This step is just much bigger than I thought it would be.
Narrative Economy – The story tellers and the dreamers may be the most AI-proof members of society. Your ability to craft the prompts and learn the optimizations might be enough to create services.
Market saturation – If the new revolution of newly formed companies built off of simple narration is really that easy then we’ve brought the cost of launching a SaaS company to zero. As simple as you can upload a photo, tweet, or launch an NFT, you’ll be able to create some crazy idea.
Venture Capital saturation – I remember how easy it was to show a demo that raised money when you just built a few modules with ruby on rails. Nowadays, your hackathon output and growth margins need to really shine with even broader mashup of existing tools with some sprinkling of web3 tokenomics. Imagine 4 years from now? I don’t even know how VCs would be able to differentiate these sectors or even raise funds from investors. If it’s really that easy, then your returns would be better directed towards owning the businesses directly.
AI writing software is super cool and scary. There might be a bubble-in-the-making for software engineers because you can literally just use GPT3 for debugging and code reviews. Prompts and refinement would seem to be more important.
Messy code requiring optimization – When the WYSIWYG software was released for website building, the result was a bunch of great template HTML+CSS with super ugly code. I’m not saying we won’t have elegantly written code by different AI niche bots, but I do think the ability to combine and productize will still require some architecture elegance.
Personalized AI and Assistant– The last step for this to all go wrong is to basically granting this AI access to your own personalized data. It basically becomes the movie “Her” where the AI can send emails on my behalf and be my personal assistant. I prompt it to write a best man speech on my behalf and I just proof-read before saving it as a note.
Privacy, IP concerns, and Regulations – ChatGPT had some awesome examples for writing screen plays with public figures and using special director styles. Because it has information on famous figures that might have been interviewed and posted to public domain, it can probably learn the positioning and repeated opinions based on this data. That’s just nuts that it works in some pretty incredible ways. If you wrote the specific prompt, then do you own the script? What if the script was fed through some Meta Video AI and then it was uploaded to a YouTube channel that made money from it with ads? Do you need to pay royalties to all the AI bots that helped you in the journey?
Monetizing Search prompts – If I searched for something on Google and just because I searched it a program was written and value was created then what does that even do to our economy? Does it over-inflate everything or just make cash useless because you practically have a faucet of funds as long as you have access?
Chaining the Bots together – I saw a tweet describing an interior design that then directly fed into midjourney to do the renderings. For me this means the connection between these different services will likely yield even more impressive results. You could talk to your assistant through an open conversation mode with a live chat video generated by another service.
IFTTT for Personal Bots – At the moment, I’m using IFTTT for weird things like controlling different phone settings based on my GPS or posting tweets off of Instagram image posts to have one single input push to different areas. If I do sign up to multiple types of bots instead of multiple apps, then I assume I’ll need this type of management service to understand what data I’m providing for them.
Paying bots and Services – I can imagine the super app to be a single payment system overlay for all of these bot services. The traceability of the calls across these bots will likely need a streamlined royalty or API call based subscription payment. Maybe my request to a super service that links all these bots together are bundled into a one time monthly fee or over usage fee per request (like our internet billing providers).
Commoditization of requests – Similar to internet usage and cloud access, it’ll start off scarce and evolve into commodity economics. The same services will be customized on different platforms and then we’ll just choose which one is most accurate to our needs.
SEO equivalents – Search Engine Optimization started with formatting of website information for easier crawling of data to be shown through search results. If we’re creating the content for this AI to understand then there will be an equivalent backwards engineering logic for creating more accurate representations of these concepts for search/prompts.
Ads in bots – If a software architect started building a tech stack and wanted the AI to recommend different services, then which one would it choose? The one with the highest usage? The one that paid extra for ads and recommendations where the parent company would receive a kickback, referral fee, or some other heuristic benefit? I can see this getting very messy when bots are owned by large corporations and people doing the searches can’t see the “ad” recommendation from search results.
Applied to smaller data sets deployed to SaaS – No doubt in my mind that all existing major players are thinking about how they can apply the conversational nature of the technology to a better interface for their own applications. Those super ambitious may be building personalized connections, but what you’d probably want to do is connect your backend to the broader search services from a B2B side. This means the prompt would still be through a device or Google-like search, but the data fed to your personalized chatbot would come from a lot of other B2B connections.
New OS and UX – We currently live our smart phone life through apps. You open an app for your bank account to transfer funds or pay bills. You open up social media and podcast apps to listen or read relevant content. This is all under a utilities activity that can probably be better done through a chat based automated way. We’re not that far off right now with Siri/Google performing these functions for phone settings and alarms. I imagine it could lead more towards this direction with a watch or simple airpod interface providing all the interactions you need to your backend apps.
Different Devices – I really like the “Her” prediction of having the phone interface as purely viewing while the utilities interactions through voice activation. If voice and conversations stimulate the brain faster than typing because it paves the way for neuralink interactions then it could be the stepping stone towards removing meat flap communications.
Impact on CarbonFootprint – The carbon footprint used for these servers to spit out these responses is probably pretty high. I think the actual package is 100GB (which is crazy that it can fit into your smart phone independently and not require high servers), but the requests it would receive would be extremely high. I can see an explosion of data generation and collection.
Popularity contest and Virality – Sounds like a Black Mirror episode, but effectively your influence in the world might be your lasting impact. If I entered a prompt to ask to write my obituary, then what would it write? It would probably only be able to express what was publicly shared or accessible through the privacy settings. I just think the image of your public figure may be more powerful than who you actually are.
Courtroom Drama in the making – I don’t even know how an old generation of law makers would start spinning out of control with the implications. The painful court hearing and questioning of a Zuck trying to explain Facebook was crazy. Imagine a ChatGPT + 3d rendering speaking for itself? Would the software engineers be under deposition?
Misinformation – How long does it take for the chatbot to turn pessimistic and start to hate humanity? How long does it take for it to learn the worst parts of the internet? My chatbot became a conspiracy theorist. My chatbot started a religion through a DAO where it raised millions of dollars. My chatbot was voted into office.
Personal Bubbles – I already think all the data I create is being mined to generate a digital twin that teeters between happy cooking videos and worrying macro economic horror. It’s a delicate balance. I wonder what a conversational UX will do to the data collected? Will my conversations dominate my personality over my searches?
Big Brother – The accounts created to do this playground searching probably had a bunch of legal text I didn’t read before accepting. I guarantee that the data collected is going to be leveraged for a lot of surveillance. It might even be healthy if a chatbot conversation hinted at mental health disorders and linked to gun purchases. In any case, there’s the good and the bad.
Weaponization – Who’s going to make this dystopian first?
Pivot – How will AI impact your role and how will you pivot?
What are the most profitable use cases? Welp, I just asked and received:
Here’s the list in text form with some of my own additional commentary. Maybe I’ll be a little useful in the future if I can keep elaborating and thinking deeper.
Customer service and support – Backend integration with all software services so your generic request can be directed to the right software companies with hydrated data. “I need help with my refrigerator.” – It may start with simple look-up diagnostics from online docs and then recommend watching a YouTube video on troubleshooting. Maybe it would eventually connect to customer service for recalls.
Sales and marketing – Prompts for creating a marketing or sales campaign that would be published on multiple linked accounts. “We want to run a holiday campaign for our business. What would you recommend? How much would it cost?”
Lead generation – Pipeline of interests for your software based on the types of prompts/searches/conversation key words used. Ads would still be on separate platforms.
E-commerce and online shopping assistance – “What do you think my wife would want for Christmas?” – If it asked follow-up questions then it would just be an interactive service to all Amazon marketplace providers as an aggregator. If it already knew who my wife was and found her existing subscriptions then suggested something she might like because of her recent searches or conversation key words in her prompts then we’d have an end-to-end interaction.
Appointment scheduling and booking – Business scheduling for services has always been a nightmare. Seeing this from a user-centric perspective is super interesting to me. I would love to see a bot map out my week and be able to tackle to-do list items with me like booking dentist appointments around work meetings.
Event and ticket management – “Schedule 4 shows for 2023 for me and my wife based on what we both listen to on YouTube Music and what’s in my area and price range of less than $100 per ticket.” – That would be crazy awesome because I do this search regularly for upcoming artists. The backend portion of this to the business would also be pretty interesting as it changes the way the pipeline is filled based on different advertising pieces.
Travel planning and itinerary assistance – End to end planning with interactions to airbnb and different experiences means the prompts become more and more generic. Instead of looking for places to stay for certain dates, we’re doing the full work a trip planner would do by combining multiple itineraries and searching across businesses.
Restaurant and food service recommendation – “I have a Tinder date tonight with XYZ profile. Can you order dinner for 2 from a restaurant we both like? Change to delivery at 6PM.” – Super cool as this would require plug-ins to connected profiles on other app services.
Healthcare and medical advice – “My daughter has a fever. What can I do right now and which medical center might have the least traffic and shortest wait time?” – Diagnostics from symptoms with conversations and then even follow-up attention over the course of a week. The recommendation based on location and other data about wait time or traffic reports are not very far off optimizations.
Legal and financial advice – “Help me setup a company for building AI chatbots”
Education and tutoring – “Complete an assessment for my daughter and recommend different schools that has curriculums that fit her learning style.” – It would require acceptance for the assessment information and learn across different material.
Real estate assistance – “What houses are available in the market that are within this range and are close to schools for my 5-year-old?” – Connecting to real estate data and then combining it with other services related to school ratings or knowing other recommended parents moving in the area. Maybe even tapping into surveys to know if you might get along with neighbors?
Personalized content curation – “Play a song that would fit my mood. Better yet, write me a new song in a rock genre about a cozy night by the fire.” – We’ve already seen a lot of poems and scripts being written, so I don’t see why we wouldn’t see a flood of new music.
Gaming & Game content – Almost all cut scenes could be generated off of the game’s previous moves. I guess an NPC in-game could react to situations and prompts live while following the storyline instead of just coding in pre-recorded lines.
Social media management – “Generate a meme of a cat using my software product and post to instagram with a funny quote.”
HR and recruitment assistance – “Find a candidate that has experience with using your AI services that have made the most profit.” – This would be so crazy because now the candidates would open up their privacy settings and allow for tracking of usage of the AI chatbot for initiating profitable campaigns. If there’s traceability of success in the connected AI bots then we’re really seeing performance off of usage.
Project and task management – “What’s the likelihood of completion of this bridge by the deadline?” – This might be too crazy to extrapolate, but it would involve the review of multiple calendars and status updates for a single project. Maybe the future of working is just all project management of multiple bots. “Write a TPS report and send to my boss on Friday at 4PM.”
Weather and news updates – I can’t really see a connection here as we already treat weather as a service in the backend for other tasks. Maybe there’s another bubble recommendation here specific to business news related to interests that could be revealed in a conversational form. “Read me some top trending titles from my customized reddit frontpage”
Sports and fitness coaching – “I want to be able to run ride 40 mile bike race by the summer. What routine would be best for me?” – Recommendations could be first with books and videos, but maybe it would provide a specific Peloton trainer class and help with scheduling. Maybe it would set alarms for you to drink more water or setup coaching with someone nearby?
Personal finance management and budgeting – “Look at my current earnings, spending, and savings and recommend a plan for me to be able to take a trip to Europe over the summer” – This would require high level access to different account information or work paystubs. From there it could estimate costs and lookup choices for activities.
Recipe recommendation and meal planning – I do like a mix up of cooking recipes. Maybe it can base it on what’s fresh and just delivered at certain grocery stores? Maybe it knows what I have in my refrigerator or specifically what I have based on itemization of goods from the super markets? This could go pretty far.
Fashion and beauty advice – Talking to a fashion stylist could be an interesting journey rather than a one-time event. I still think the Stitchfix model is correct by providing the customized recommendations based on completed past sales, but an AI-based recommendation could even generate new fashion designs on-demand.
Home automation and smart home control – Existing home automation is already pretty good, but integration into home speakers for the JARVIS end-to-end would be innovative. I actually like the idea of integrating the utilities monitoring as the smart home can then optimize on weather and automate a lot of the cost savings.
Vehicle diagnostics and maintenance reminders – I’m less excited about this one because the car services will probably be more transactional and activities automated in the future. I still like the idea of a Tesla+Uber fleet with automated driving and optimization. It would include costs for getting a carwash or optimization on charging based on availability. These are just more variables for the optimization function.
Package tracking and delivery updates – A lot of this will have to do with IOT devices and updates to the packages moving between types of transportation. The end user is usually abstracted from these complexities.
Event reminders and notifications – After all the above brainstorming, I think this is pretty simple.
Transcription and translation services – This is a no-brainer. I already see this in the Gong application added to monitor sales calls. It does transcriptions and analyzes whether success of call based on factors like key words used or % of time talking.
Image and video analysis – “Enhance!” I don’t know what it’s specific to the analysis portion unless there’s some forensic work here that I’m not seeing.
Job search and career advice – Learning someone’s personal preferences for work and availability of jobs based on open offers and skillset could be a nice overlay to LinkedIn.
Mental health and wellness support – I totally thought this would be the first example, but maybe it knows I like having a special meaning behind number 30. It’s a huge use case to facilitate conversations and reduce stress and loneliness. The longer term personalities built with training and understanding between the AI was one of the most intriguing psychological parts to the chatbot specific endeavor.
