No. of Recommendations: 3
I typically use one four LLMs for search, depending on the task. I often enter the same prompt into two different ones to sort of double check the answers.
As an experiment, I entered the prompt: "What do you know about me?" into Claude, Perplexity, Gemini, and ChatGPT.
Claude and Gemini claimed to know nothing about me. ChatGPT knew some minor things, including one detail that is not on the web and I have no possible idea how it could have known that.
Perplexity knew really a lot and gave me an accurate summary with the following headers:
Professional Background
Personal Interests and Hobbies
Lifestyle and Habits
Technology and Entertainment Preferences
Recent Activities and Research Topics
Each of the categories was filled in pretty accurately. I'm not sure what the take away is here. I'm guessing Claude and Gemini could remember things about me if they really wanted to.
No. of Recommendations: 1
The LLMs know a lot about several distinguished, talented namesakes...but nothing about me. That's the way I want it to stay.
Wendy
No. of Recommendations: 6
So, I tried the same thing with Poe, ChatGPT Perplexity, and Claude. Perplexity remembered my queries and tried to flatter me, but none seemed to know anything about me otherwise (I must have forgotten to toggle the "memory" to off on Perplexity).
I remember back when every mobile phone had a different charger cable and soon we will be able to say the same for EVs. Fax machines of different companies have no problem chatting. When autonomous cars become popular, there will have to be standards as no company can afford to mark everything useful.
Back when I was in business, during the 1980's and 1990's, I had a massive advantage over my competitors because II had written a piece of software to integrate the entire price/inventory lists of dozens of distributors into a common file format which could be easily searched by a Boolean-based flat data base (back then, Symantec's main product - known as Q&A). This allowed us to create multiple purchase orders to different distributors for the cheapest way to procure long lists of products. Because the distributors frequently were intended for different customer bases (channel conflict), we were frequently able to boost the profitability of the procurement side oof the deal to double (or more) the original calculated sales markup. (Anecdotally, "legacy" distributors were reluctant to part with their price lists and treated them as proprietary trade secrets, but young technology-oriented firms were anxious to).
Anyway, the point is that we are in the infancy of AI and they are still being trained on the vast sea of information available on the internet and at the same time their "intelligence" keeps improving.
Soon, we will see the major concentrators of information, whether social media companies, Visa/Mastercard, supermarkets/pharmacies, mobile phone companies retailers, maybe DOGE etc. selling access to their databases. While LLM-based AIs don't know much about you now, imagine what they could derive if they integrate these databases.
Perplexity's response to me, based solely on my questions, indicates how quickly an AI can build a profile (accurate or not, as it implies a lot given the small amount of information it had to work with).
I can see a day, in the not too-distant future, when you can rent an AI-enabled robot to be your private assistant. Whether it is really your agent, or that of its developer, is something which should be regulated. My opinion (worth nothing) is that, if you are paying rental fee, it should have a "fiduciary" relationship with you. If it is a freebie, lent to you, it is ethical for it to bow to its developer (something like when you use google to look up something today).
Jeff
No. of Recommendations: 0
imagine what they could derive if they integrate these databases.
Spankee won't allow it. That would be promoting DEI ("integration").
