Blogpost
July 15, 2025
Blogpost
July 15, 2025
What are your thoughts on it? AI, often seen as a promise of a pot of gold at the end of the rainbow, is it truly real?
Some people expect AI to fix everything, which could be optimistic, while others are worried about privacy. I will call these our primary concerns, although both concerns are entirely understandable.
Finding a middle ground can provide comfort, a clear path to progress, and additional learning benefits. Keeping our primary concerns as the foundation for our plan ensures stability and direction.
Privacy has to come first, as it’s typically associated with location as well as compliance, which makes for a good starting place, and is the subject of this blog post.
I don’t want to create a long post, so I am going to get straight into it. I have selected four of the big ChatGPT engines and summarised them, providing comments on each, which may help with one of our core concerns.
ChatGPT, a successful AI chatbot released in November 2022 by OpenAI, excels in writing, STEM tasks like coding and math, and conversation. Copilot improved upon it by adding Internet access and a knowledge cutoff. In May 2024, OpenAI upgraded ChatGPT with GPT-4o, including web browsing, vision, data analysis, file uploads, and GPTs, previously exclusive to ChatGPT Plus. Users can access ChatGPT via its website without logging in, with advanced features available on a free account. It continues to generate text, solve math, code, and offer engaging conversations for free. If memory is on, it can collect Account credentials (Memory is off but default).
Pros
Cons
Data Stored:
Data Collected:
Account credentials, audio, browser type and settings, contact information, date of birth, device name, device's operating system, files, general location, images, IP address, name, payment details, service interactions, text prompts, transaction history, type of computer or mobile device, and your computer connection.
In February 2023, Microsoft launched a new AI-enhanced Bing called Copilot, running on GPT-4 Turbo, giving it ChatGPT-level intelligence.
In October, Copilot rebranded with a modern, user-friendly UI and added features like free Copilot Voice, a version of ChatGPT's Advanced Voice Mode.
Pros
Cons
Data Stored:
Data Collected:
Prompts and responses
Gemini is Google's conversational AI chatbot, similar to Copilot, sourcing answers from the web, providing footnotes, and generating images within its chatbot. It uses Imagen 3, named ZDNET's best AI image generator of 2024. At the Made by Google 2024 event, Google made Gemini its default voice assistant, replacing Google Assistant with a more innovative alternative. Gemini Live offers human-like, multi-turn conversations on complex topics and advice. The experience mimics human conversation, allowing interruptions, multi-turn exchanges, and resuming chats. It is now available for free on iOS and Android.
Data Stored:
Host Region: Data is stored and processed entirely within your chosen Google Cloud region (e.g. UK South, France, Canada, etc.) cloud.google.com+3Reddit+3Gov Capital+3. Google guarantees:
Data Collected:
Addresses, call and message logs, chat transcripts, contacts, device information, dialler, feedback, Gemini Apps conversations, Google Assistant information (such as smart home device names and playlists), images uploaded for feedback purposes, installed apps, IP address, location information, permissions, preferred language, related product usage information, screen context
DeepSeek, a Chinese AI firm founded in May 2023 by Liang Wenfeng, has developed advanced language models rivalling OpenAI's GPT-4 and Anthropic’s Claude at lower costs. Its open-source AI tools support NLP, content creation, data analysis, coding, and automation, challenging U.S. AI dominance and marking a new AI era.
Data Stored:
Data Collected:
Advertising information, chat history, crash reports, date of birth, device model, diagnostic data, email address, feedback, IP address, keystroke patterns, operating system, passwords, payment details, performance data, logs, phone number, prompts, service info, system language, input data, uploaded files, username.
We've encountered an issue: we've only addressed core concerns, and I'd appreciate your thoughts on the privacy aspect. My research indicates that Bad Actors could exploit the collected data for cyberattacks. However, and quite ironically, we can also use LLM models to protect ourselves from cyberattacks. Is there a problem in the making?
July 4, 2025
July 15, 2025