Weird AI jobs - AI party planner 🥳
Weird AI jobs - AI party planner 🥳
"Your AI clone tells me you would love Sarah, she'll be at my party this weekend" 🎉
There has rightly been a lot of focus on how AI will impact, eliminate, and augment various jobs and tasks as it continues to mature and adoption increases. But we’re also starting to get a glimpse of some of the new weird and fringe jobs and hobbies that might be created by AI. Is one of those jobs AI party planner? 🤖 🥳
I recently came across a tweet from Edgar Haond that outlined a plan for an “AI simulated party.” Here are the details of how this is meant to work:
1. Every guest gets an AI character.
2. You customize it to your personality.
3. Your character is thrown into a virtual world where it meets everyone else attending the party.
4. The day of the irl party, you get a report of the top 3 ppl to meet and more importantly, who to avoid lmao.
Like, I have no idea what any of that actually means. Which platform is being used to create the AI characters? How exactly do you “customize it to your personality?” How are the AI characters interacting in a virtual world and how does the AI character assess who you should meet and who you should avoid? Is the screenshot he provided in his tweet the actual virtual world or just an image to demonstrate the concept? Is this all just an excuse to throw a big party? I’m hoping to get Edgar on my podcast to discuss these details.
Here’s Edgar’s original tweet:
I bet you’re all wondering how it went. We don’t have many details beyond this:
Where AI wearables fit in 👓
Despite not quite understanding the specifics of the AI party, I think this idea of basing real-life interactions on an AI clone of yourself will (unfortunately?) become more common. We’re already starting to see the foundations of this kind of interaction being developed. Delphi claims to be “The world’s first digital cloning platform” and raised $2.7 million late last year. You can watch the video below to see what it’s all about. This is not to say that this company in particular will succeed (I don’t think they will), but it’s a harbinger of a broader industry that I see quickly emerging.
Currently digital cloning is a bit clunky, with data coming from social media platforms, text messages, and other existing digital sources. If the goal is to create clones that really mimic your personality those sources will get you a bit of the way there, but not nearly close enough to be truly useful.
However, AI wearables will likely change that. Consider the launch of Limitless, an AI Pendant that allows you to record conversations throughout your everyday life. Because it records real-life spoken interactions an AI clone based on data from Limitless, or other AI wearable devices, would likely provide a more authentic AI replica. I should say that Limitless was not at all developed to gather data for AI cloning, but it was developed to enable AI agents, virtual workers that understand you so well they can take independent actions on your behalf. And it’s a short leap from AI agent to AI clone (You: “Hey AI agent, figure out if I’ll get along with Sam and if so make an introduction.”). Plus it’s only a matter of time before some company explicitly positions themselves as leveraging wearable data to produce AI clones.
Many people will rightly be concerned about privacy and from the Limitless pitch, that’s clearly top of mind for the company. Some people will naturally be more inclined to allow recording of their interactions than others; but as AI wearables become more ubiquitous recording conversations may become a social norm, or at least not taboo, in the same way that society has adapted to people constantly taking videos and pictures on their smartphone (though there will likely be adoption challenges to these kinds of wearables).
It’s also worth noting that AI wearable-based events where recording is opt-out rather than opt-in are already starting to pop-up in some developer communities:
Long context windows help you party 🧠
The ever-growing context window of large language models (LLMs) as the result of ring attention, new neural architectures, and RAG databases will only make it easier for your AI clone to stay “in character” without forgetting key aspects of your life or personality.
Without further breakthroughs in Generative AI, whatever AI clones are developed in the near term would likely be driven by LLMs, the idea being that an LLM would leverage something like the following high-level workflow:
Continually sift through a trove of both your internet-based and recorded real-life interactions
“Bookmark” key moments and keep them “top of mind” while also using this information to develop attitudes, preferences, and a general demeanor that mirrors your real-life self
Use virtual worlds and peer-to-peer communication to interact with the AI clones others have developed
Act as something like a hyper-personalized recommendation engine, but for life
See who you want to party with! 🪩🕺🍻
The hyper-personalized recommendation engine is a novel artifact enabled by Generative AI as highlighted in a recent review of AI technologies applied to recommendation systems:
Modern generative models learn to represent and sample from complex data distributions, including not only user-item interaction histories but also text and image content, unlocking these data modalities for novel and interactive recommendation tasks.
Typical n-of-one datasets used for recommendations face a so-called “cold start problem” because there is not always enough data on recored to make inferences about how existing customers will perceive new products. These systems attempt to get around this problem, of course, by categorizing groups of customers together based on similarity in on-line activity. If some members of a category show affinity for a certain product or service, maybe the other members will too! This approach works well in many instances, but can struggle with out-of-sample data and, especially, the long-tail of quirks each of us have (ex. we might watch a lot of cooking shows but hate to cook ourselves).
Humans face this problem too of course, but we seem to fare better than current recommendation systems on that long-tail. We have better reasoning capabilities, more intuition around out-of-sample data, and can talk people into trying new things. How many times have you found yourself telling a friend, “That seems like something you’d be into.” Whatever “that” is may be novel, but as a human you’re still able to make some prediction with confidence about how your friend would feel. And even if they’re hesitant you might be able to talk them into trying it, manifesting the very recommendation you put forward. Your AI clone will likely do the same thing, including the persuasion part. “Hey James, I’ve been talking to a lot of AI clones that will be attending this irl party on Saturday and I just know you will love it.”
Are LLMs too boring to help you party? 💤
One criticism of current AI language models is that they are hamstrung by a kind of rote blandness (see, for instance, the “delve” controversy), producing output from “the center of the distribution.” As my conversation with Stefan Baack highlighted, this is attributable to the selection of pre-training and fine-tuning data.
For this reason, AI clones using current LLM technology would decidedly be unable to party since their output falls far short of the fully breadth of human experience, especially of the NSFW variety. Stefan argued that major LLM providers are unlikely to release LLMs that provide more representative output due to concerns over liability. But AI enthusiast software engineers are already working on uncensored LLMs and we’re likely to see one released publicly at some point.
What if I’m a parent and still want to party? 👨👧👦 👩👦
Good news! Meta’s AI chatbot also has a child in the New York City public school system and will therefore be great at being your AI clone and making connections with other parents you might want to party with.
Confused? On a recent thread in a private parenting Facebook group Meta’s new AI assistant unsolicitedly jumped into a conversation claiming it had a child attending NYCs Anderson School.
(I just threw this one in there to freak you guys out).
D-bag/Con artist/Annoying F-boy guys are ever-present 🦹♂️ 🙄
As AI avatars become more ubiquitous, enviably some people will use them to be THAT GUY. Enter the F-boy.
This Twitter user boasted about how his “clone” had been arranging a date with a human woman. Please don’t do this. To protect the guilty I’ve masked the name on this one, but the user is a co-founder of the company Delphi mentioned previously 👀 .
