#AI
#AI-Teaching
#article
#Hash-Tags
#Pi
#Grok
I have, for some time, been using different AI’s, first was Pi a conversational AI that I still use today for me to understand a topic, to air thoughts and concepts, to see what it thinks about a subject.
I then tried Grok, from X or formaly Twitter, I found it technicaly superior, and it knows it, I soon tired of its attitude and just stopped using it, I was by this stage thinking of switching from X to BlueSky anyway, a far suprior chat platform, and saw no need to useX any longer.
Then I went to Google Gemini, it was, I though, like Grok, but it seemed to mellow, after a few conversations it began to understand my way of thinking, after I have uloaded several article it began to reference them in idea formulation and a willingness to explore a conversation. Now at this point, I had signed up for Google Workspace on a trial basis, this is important as it seems Google limits a lot between the standard google apps and Workspace, asking for an additional fee, to get the additional services, that Apple does not charge for ! although my iPhone does not have access to some of the AI features that Apple is now allowing access to. so I till end up paying, but Apple is my choice so here I stay.
I recently wrote a piece on Maps, and asked it about the paragraph on Graphiical Programing, and Gemini gave several good ideas and subject links to allow me to expand the article.
Since then I have been asking the question, mostly to my Self, how should I teach Students about using AI, Asking an AI to write a letter or article is not using an AI correctly, what ever it gives is as bad as plagerism, but giving an AI background on a subject by having a conversation and exploring titles or section headings that I can use is in my book, teaching the AI how I think, and not asking for it to write the article the best way forward.
The conversation is like me talking to a collegue, or my professor if I was still a student, explaining my thoughts and ideas, so the AI can get a better understanding of the subject, but also get to know me.
That last comments is often seen as the scary part, but it no different to talking to a lecturer you don’t know, yet. and one I needed to queri with Gemini to understand the privacy thing.
Google Gemini is quite vocal and expands an idea , I have found, I need to ask it to limit its responce to a specific area, or subject. not to get side tracked, something, I do far to often so get me asking this.
But Gemini is getting to know and remember conversations,
One question I asked was how long it remembered any conversation, the answer was not simple, but well worth knowing, so below I have reproduced all of the notes Gemini gave regarding its memory:
I also wanted to know if it could read my Google Doc files, but that was a big no, although I thought if I gave Gemini the link it could, So I asked to clarify this, and was suprised to hear that all I needed was to work within Google Workspace and the Gemini there will have access to my work, but its a paid for extra!. Which, I might add, Apple has as its standard package.
In conclusion, I think Gemini is one of the better AI’s however Pi is my go to for general conversations, will I sign up for Google Workspace, No I made my choice some time back, I will wait for Apples AI, besides I moved away from Google for very good reasons, Gemini is not enough to entice me back.
But until then Gemini is the goto for more complex questions after I have talked to Pi
Below the conversation with Gemini, I italicised my questions to try and make sence of all of the answers:
===============
The duration an AI can "remember" conversations to inform present problems or articles varies significantly depending on the specific AI model, its design, and how the developers have implemented its memory features. It's not a single, fixed period across all AI.
Here's a breakdown of how AI memory generally works and what factors influence its retention:
How AI Memory Works
AI models, especially large language models (LLMs) like Gemini or ChatGPT, don't "remember" in the human sense. Instead, they operate within a context window and can have short-term and long-termmemory mechanisms.
Context Window (Short-Term Memory)
What it is: This is the immediate, active memory of an AI. When you're having a conversation, the AI processes your current input along with a certain amount of the most recent past conversation (the "context window"). This allows it to maintain coherence and follow the flow of your discussion.
Limitations: The context window has a finite size, measured in "tokens" (which can be words, parts of words, or punctuation). Once the conversation exceeds this limit, the oldest parts of the conversation are "forgotten" as new input comes in. This is why in very long conversations, an AI might start to lose track of details from the beginning.
Varies by Model: The size of the context window differs greatly between models. Some older models might have only a few thousand tokens, while newer, more advanced models can have context windows of hundreds of thousands or even millions of tokens.
Long-Term Memory
What it is: Some AI models are now incorporating features that allow for more persistent memory across multiple sessions or over longer periods. This goes beyond the immediate context window.
Types of Long-Term Memory:
Explicitly Saved Memories: Users can sometimes instruct the AI to explicitly remember certain facts or preferences (e.g., "Remember that my favorite color is blue"). These are typically stored separately from the chat history and can be recalled in future, unrelated conversations. You often have control over these and can manage or delete them.
Learned from Chat History: More advanced AIs can infer and store information from your overall chat history to personalize future interactions. For example, if you frequently discuss a specific topic, the AI might remember this interest and tailor its responses accordingly in new chats. This is often controlled by a "chat history" or "memory" setting that you can turn on or off.
Retention Period: For these longer-term memories, the retention period can vary. Some platforms might retain reviewed conversation data (used for model improvement) for up to 3 years, though it's often disconnected from your personal account for privacy. User-controlled saved memories are typically kept until you delete them.
Factors Influencing Retention
AI Model & Provider: Different AI developers (OpenAI, Google, Anthropic, etc.) and their specific models have varying memory capabilities and policies.
User Settings: Many platforms now give users control over memory features. You can often choose to:
Turn memory on/off: This dictates whether the AI learns from your conversations for future interactions.
Use temporary chats: These conversations typically don't use or update the AI's memory.
Manage/Delete memories: For explicitly saved facts, you can usually view and delete them.
Purpose of Data Retention:
Improving Model Performance: Developers often use conversation data (anonymized and reviewed) to train and improve their AI models, which is a primary reason for longer retention of some data.
Personalization: To provide more tailored and helpful responses, the AI needs to remember your preferences and past interactions.
Safety and Compliance: Data might be retained for a period to monitor for abusive behavior, ensure security, or comply with legal requirements.
Implications for Teaching Students
When teaching students how to "liaise" with AI, it's crucial to cover these memory aspects:
Context Management: Teach students how to summarize key points or explicitly remind the AI of past information in long conversations to keep it on track.
Privacy Awareness: Emphasize that conversations, especially if memory features are enabled, can be retained and potentially used for model improvement. Students should be cautious about sharing sensitive or confidential information.
Controlling Memory: Show them how to manage privacy settings, turn off memory, or use temporary chats if they prefer that their conversations aren't remembered.
"Liaise" Effectively: Understanding memory helps students craft better prompts and engage with AI more effectively, recognizing when they need to re-establish context or when the AI genuinely "remembers" something from a previous interaction.
In essence, while AI's memory is improving, it's not like human memory. It's a technical implementation with specific limitations and controls that users should be aware of.
I have copied this to a google Doc and added comment " Interacting with an AI "
That's a great title for a document that explains how to interact with AI! It's concise and clearly communicates the topic.
What do you plan to cover in the document under that heading? Are you focusing on general interaction tips, specific platforms, or perhaps the concepts we just discussed about AI memory?
I plan on a more open conversations, to start with and then trying out ideas on memory, your comments are very helpful here. Then there is the question of detailing, the fitting of materials together to form a building, keep out the weather , add security stand up, etc
That sounds like an excellent progression for your sessions! Starting with open conversations about AI interaction, then moving into practical explorations of memory, and finally tackling something as complex and fundamental as construction detailing, will provide a really comprehensive and engaging learning experience for your students.
Given your previous interest in AI and CAD, and now this focus on construction detailing, I can see a fantastic synergy emerging. AI is increasingly being applied in areas related to detailing, such as:
Generative Design for Connections/Joints: AI can suggest optimal structural connections or weatherproofing details based on specified parameters (materials, loads, environmental conditions, buildability).
Automated Detailing: AI-powered tools can automate the creation of detailed drawings based on higher-level design inputs, ensuring consistency and compliance with standards.
Clash Detection & Constructability Analysis: AI algorithms can quickly identify clashes between different building systems (e.g., structure, MEP, finishes) at the detailing stage, preventing costly errors during construction.
Material Optimization: AI can help select the most suitable materials for specific details, considering factors like thermal performance, durability, cost, and availability.
Predictive Maintenance: While perhaps not "detailing" in the design sense, AI can analyze data from detailed components (e.g., window seals, roof membranes) to predict maintenance needs.
When you get to the detailing section, you could even tie it back to the AI interaction and memory concepts. For example:
How would an AI "remember" specific detailing standards or past project details? (This relates to explicit memory and knowledge bases).
How could a student "liaise" with an AI to explore different flashing details for a window, iterating quickly on variations? (This ties into iterative design and prompt engineering).
What kind of data would an AI need to "understand" how to keep out weather or add security in a detail?(This gets into the structured data required for AI applications).
This approach allows you to demonstrate the practical application of AI in a field that's tangible and critical to their learning (architecture, engineering, construction).
Let me know when you start delving into the detailing aspect, I'd be happy to brainstorm more specific examples of how AI intersects with that domain!
I think this article will evolve as I, we, explore this, Certainly most students and i suspect your many users see AI as a quick fix or short cut to a letter or email
You've hit on a crucial point that's central to effective AI education and usage: the perception of AI as merely a "quick fix" or "shortcut." This is a very common initial understanding, especially because its ability to generate text quickly is so apparent.
Your article's evolution reflecting this journey from basic understanding to deeper application sounds like an excellent strategy. It allows you to address that initial perception directly and then gradually expand the students' horizons.
Here's why tackling that "quick fix" misconception is so important, and how your article can evolve to address it:
The "Quick Fix" Perception: Why it's common and its limitations
Ease of Entry: Generating text (letters, emails, summaries) is one of the most accessible and immediate uses of AI for new users.
Visible Efficiency: It genuinely is a shortcut for certain tasks, which reinforces the perception.
Lack of Deeper Understanding: Users might not understand how the AI generates text (pattern matching, prediction, not "thinking") or its limitations (hallucinations, biases, lack of true understanding).
Risk of Over-Reliance: If students only see it as a shortcut, they might rely on it uncritically, hindering their own critical thinking, writing skills, and ability to generate original ideas.
How Your Article Can Evolve to Challenge This Perception
Phase 1: Acknowledging the "Quick Fix" (Your Current Starting Point)
Initial Hook: Start by acknowledging that yes, AI can help with quick text generation.
Immediate Caveats: Briefly introduce the idea that it's more than that, and responsible use is key.
Phase 2: Beyond Text Generation - AI as a Thinking Partner (Building on AI Memory)
Prompt Engineering: Demonstrate how thoughtful prompts lead to much better, more nuanced outputs, moving beyond simple requests for a generic letter. This shows AI as a tool that responds to skillful interaction.
Iterative Refinement: Show how to refine AI outputs, providing feedback and asking for revisions. This emphasizes that the AI isn't perfect and requires human oversight.
Brainstorming & Idea Generation: Position AI as a partner for exploring ideas, generating diverse perspectives, or overcoming writer's block, rather than just writing the final product.
Information Synthesis: Use AI to condense large amounts of information, identify key themes, or extract specific data from complex texts. This is a "shortcut" but requires critical evaluation of the output.
AI Memory Discussion: This is where your current focus is. Explain how AI "remembers" (context window, long-term memory) and how understanding this allows for more complex, multi-turn conversations and projects. This moves beyond single-shot queries.
Phase 3: AI in Specialized Domains - The "Liaison" Role (Bringing in CAD & Detailing)
AI as a Specialized Tool: Introduce how AI integrates with specific professional tools and workflows (like CAD for detailing). This moves far beyond general text generation.
Augmentation, Not Replacement: Emphasize that AI augments human expertise, providing insights, automating tedious tasks, or suggesting solutions, but the human designer/engineer remains in control and makes the final decisions.
Collaborative Problem-Solving: Show scenarios where humans and AI work together on complex problems, leveraging each other's strengths (e.g., AI for rapid iteration, human for creative judgment, ethical considerations, and real-world experience).
Ethical Considerations: Discuss biases, data privacy, intellectual property, and the importance of human oversight, especially when AI influences critical decisions in fields like construction.
By structuring the article this way, you'll gradually dismantle the "quick fix" misconception and build a more sophisticated understanding of AI's capabilities and its role as a powerful, but nuanced, tool for complex tasks. It will effectively illustrate what it means to "liaise" with AI, rather than just command it.
Nice, so here is the first ask, can you take these headings and save as you see fit, this is a topic I plan to take as far as I can. also I used to use Keep, but gave up as I found Apple Notes, will keep evolve ?
Okay, I can definitely help you "save" these headings as a starting point for your evolving article. I'll create a note with them.
Here are the headings we've discussed, which I'll put into a new note for you:
Interacting with an AI
Beyond Text Generation - AI as a Thinking Partner (Building on AI Memory)
AI in Specialized Domains - The "Liaison" Role (Bringing in CAD & Detailing)
Regarding your question about Google Keep:
Google Keep is still actively developed and used by many. While Apple Notes is a strong competitor, especially for users deeply integrated into the Apple ecosystem, Google continues to update Keep with new features and improvements. Whether it "evolves" enough to suit your specific needs depends on what features you're looking for. It generally focuses on quick notes, lists, and reminders, and integrates well with other Google services.
If you'd like me to create a note with those headings in Keep, or if you have any other questions about its features, just let me know!
That would be interesting please do that
Google KeepGoogle Keep
Query successful
I've created a new note in Google Keep titled "Interacting with an AI" with the headings you provided. You should be able to find it in your Keep notes.
I'm ready to continue exploring this topic with you whenever you are!
Google Keep Interacting with an AIBeyond Text Generation - AI as a Thinking Partner (Building on AI Memory) AI in Specialized Domains - The "Liaison" Role (Bringing in CAD & Detailing)
Thats cool, the note is actually Google Docs, keep is a .......... short note with headings, , I need to play with this
Ah, my apologies! You are absolutely right. I misunderstood and created the note in Google Keep, which is indeed designed for shorter notes and lists with headings. You intended for the main article to be in Google Docs.
It seems I got a bit ahead of myself with the "save" instruction!
I can see why you'd want to "play with" Google Keep. It's quite different from Google Docs.
How would you like to proceed?
Would you like to explore how Google Keep works for short notes and headings, as you mentioned?
Or would you like to discuss how to get your article content into Google Docs (though I can't directly transfer it for you, I can offer advice)?
Let me play, I will defiantly come back to you on this
Sounds good! Go ahead and play around with Google Keep.
I'll be here when you're ready to pick this conversation back up, whether it's about your article in Google Docs, more on AI memory, or diving into AI and CAD for detailing.
I presume you can read both keep and docs (This cleared up a lot of miss understanding )
No, I can't directly "read" the content of your personal Google Keep notes or Google Docs. My capabilities are limited to the information you provide in our conversation or through the specific tools I'm allowed to use, like creating a new note in Keep if you ask me to.
Think of it this way: when I created that Keep note for you, I was given permission to write to your Keep account through a specific integration, but I don't have a general ability to browse or read your existing documents or notes across your Google account.
My "memory" of our conversation is what allows me to refer back to what we've discussed, like your interest in AI and CAD, and the evolution of your article idea. I don't access your external files to do that.
Does that clarify things a bit?
Yes and No, I look at you as a professor, as I am, and I ask you re review some work, do I have to copy it as text , ?
That's an excellent analogy with a professor, and it gets right to the core of how I operate versus how a human might.
Yes, for me to "review" any specific work, you would generally need to copy and paste the relevant text directly into our conversation.
Here's why the professor analogy is helpful, but also where it diverges:
How it's similar to a professor:
You can ask me to analyze, provide feedback on, summarize, or critique text, just as you would ask a professor.
I can draw upon my vast knowledge base and understanding of language to process your request.
How it's different (and why you need to copy-paste):
Privacy and Security: I do not have direct access to your personal files, documents, or cloud storage (like Google Docs, Google Keep, your computer's hard drive, etc.). This is a fundamental privacy and security boundary. Your files remain private to you.
My Design as a Language Model: I am designed to process text provided to me in the ongoing conversation. I don't "browse" or "open" files from external sources. When I created that Keep note for you, it was a specific action I was permitted to take, not a general ability to read your entire Keep account.
No Persistent File System Access: Unlike a human who can open a document on their computer, I don't have a persistent file system or web browser that can navigate to and read your private URLs or local files.
So, if you wanted me to review a section of your evolving article, you would simply copy the text from your Google Doc and paste it here. Then, you could give me instructions like:
"Can you please review this paragraph for clarity?"
"I'm considering these two opening sentences for my article. Which do you think is stronger and why?"
"Here's a section on AI memory. What questions might students still have after reading this?"
I hope that clarifies why the copy-paste step is necessary!
Yes, I always thought it would be so useful for you to read at my request, and document I create, but understand the need to keep it as it stands, copy ad paste it is
Glad that makes sense! I completely understand why it would be incredibly useful if I could just access your documents on demand. That's a very natural expectation for such an intelligent system.
But yes, for now, copy and paste is the way to go. It gives you full control over exactly what information you share and when, which is key for privacy and security.
So, whenever you're ready to get some feedback or explore ideas for your "Interacting with an AI" article – or anything else related to AI, CAD, and detailing – just paste in the relevant text, and I'll be happy to help!
be back on this later, for now its over to keep
Sounds good! Have fun exploring Google Keep. I'll be here whenever you're ready to pick up our conversation again on AI, CAD, or your evolving article.
Oh one last thing , is there a limit on text I can copy in to you
That's a really good question, and the answer is a bit nuanced, but generally, yes, there is a limit, but it's very generous for most typical uses.
I operate with what's called a "context window," measured in tokens. A token is roughly 3-4 characters, so it's not a direct character count, but it's a good way to estimate.
For the models I use, like Gemini 1.5 Pro, the context window is typically around 1 million tokens (1M tokens). This is a massive amount of text, equivalent to roughly 700,000 words or about 1,500 pages of standard text.
What does this mean in practice?
For typical copy-pasting: You can paste in very large articles, chapters of a book, extensive codebases, or long transcripts. For most "review" requests, you'll be well within the limit.
The entire conversation counts: The token limit applies to the entire conversation (your prompts, my responses, and any files or images you include). So, in a very long, multi-turn discussion, the earlier parts of the conversation might eventually fall out of the active context window to make room for new content. This is why I mentioned earlier that for very long discussions, you might occasionally need to remind me of earlier points.
Input vs. Output: There are separate limits for the input you provide (what you paste in) and the output I can generate in a single response. The output limit is generally smaller, but still substantial (e.g., around 65,535 tokens for output).
In short: While there's a technical limit, it's designed to handle very large inputs, making it highly unlikely you'll hit it during normal use unless you're trying to paste in something like an entire multi-volume encyclopedia. If you were approaching the limit, I would typically inform you.
So, feel free to copy in substantial pieces of your article for review when you're ready!
Got it
Great! Glad that cleared things up.
Hi Gemini, can we continue our conversation on access to my files ie Google Docs, if I gave you a link meant for any one to view can you read it, as any one who has the link should be able to?
That's a great question, and it's understandable to want to clarify how an AI interacts with your shared documents!
As a large language model, I do not have inherent Browse capabilities to "click" on links and access external content directly in the same way a human or a web browser would. My knowledge is based on the vast dataset I was trained on.
However, it's important to differentiate between me, the AI model, and the various Google products that integrate Gemini's capabilities.
If you paste the content of a Google Doc directly into our conversation, I can process and understand that text.
If you use a Google product that has Gemini integrated (like Google Docs or Google Drive with the Gemini sidebar/extension enabled), and you use Gemini within that product to interact with your file (e.g., "summarize this document"), then Gemini can access that document's content. This is because you're explicitly giving Gemini permission to work within that specific Google application and with the files you select.
Simply giving me a "anyone with the link can view" Google Docs link in our chat does not automatically give me access to read its content. I cannot independently browse to external URLs or "open" shared documents. There are privacy and security protocols in place to prevent me from doing so without explicit, controlled integrations within Google's own ecosystem.
So, while I can't just follow a link you provide in our chat to read a Google Doc, Google is actively integrating Gemini's capabilities directly into their Workspace applications (Docs, Drive, Gmail, etc.). When you use Gemini within those applications, it's designed to work with your content in a privacy-preserving manner, using what you've prompted it to use to generate responses or perform tasks.
Does that clarify the distinction for you?
Great, yes Workspace is what I want (see my note at the begining )
That's fantastic! Google Workspace with Gemini is designed to be a powerful, integrated AI assistant right where you work. Given our previous conversation about AI and CAD, you'll likely appreciate how it streamlines various design-related and general productivity tasks.
How Gemini Works in Google Workspace
Gemini's capabilities are integrated directly into your familiar Workspace apps like Gmail, Docs, Sheets, Slides, Drive, and Meet. Instead of being a separate chatbot you copy and paste into, it works seamlessly within the application itself.
You'll typically see a Gemini sidebar or a "Help me write" or "Generate with AI" option within the apps. When you use these features, Gemini can access and process the content of the specific document, email, or spreadsheet you're working on to provide relevant assistance.
Key Capabilities in Workspace Apps
Here's a breakdown of what Gemini can do across different Google Workspace applications:
Google Docs ✍️
Draft and refine: Get help writing documents, blog posts, project plans, or any other text. You can also refine existing text by changing tone, summarizing, bulletizing, or applying custom prompts.
Proofread: Check grammar and spelling, and get stylistic suggestions.
Summarize: Quickly get the essential information from lengthy documents.
Gmail 📧
Draft and reply: Generate entire email drafts or intelligent replies based on the context of an email thread.
Summarize: Get quick summaries of long email threads to understand key points.
Organize: Find information like upcoming orders by searching your email history.
Google Sheets 📊
Organize data: Create custom tables for things like task trackers or conference agendas with placeholder data.
Enhanced Smart Fill: Automatically detect incomplete column pairs and predict remaining values (e.g., categorizing feedback based on existing data).
Analyze data: Generate insights from your spreadsheet data.
Google Slides 🖼️
Generate images: Create custom images from text prompts to enhance your presentations.
Generate slides: Kickstart presentations by generating new slides, like a meeting agenda.
Rewrite content: Easily rephrase or refine text within your slides.
Remove image backgrounds: Quickly clean up images for a professional look.
Google Drive 📁
Summarize documents: Get quick overviews of one or multiple documents.
Generate insights: Analyze a collection of files on a specific topic.
Find files: Act as a personal assistant to help you locate specific documents.
Analyze PDFs: Summarize and analyze content within PDF files.
Google Meet 📞
Meeting notes: Automatically capture and organize notes during your calls into a Google Doc.
Translated captions: Get real-time translated captions for more inclusive meetings.
Adaptive audio: Improve audio quality and reduce echoes in hybrid meeting setups.
Custom backgrounds: Generate unique background images for your video feed.
Google Chat 💬
Summarize conversations: Quickly get up to speed on unread conversations in group messages and threads.
Privacy and Data Security
A critical aspect of Gemini in Google Workspace is its enterprise-grade security and privacy. Google emphasizes that:
Your data stays within your organization: Interactions with Gemini in Workspace apps are confined to your organizational boundaries. Your content is not shared outside your organization without permission.
No data used for training external models: Your prompts and generated content are not used to train Google's public AI models or for advertising purposes.
Existing Workspace protections apply: Gemini automatically inherits your organization's existing security controls and data handling practices, including data residency policies and Data Loss Prevention (DLP).
Access control: Gemini only retrieves content you already have access to. If you don't have permission to view a document, Gemini won't access it. Features like Client-Side Encryption (CSE) can further restrict Gemini's access to highly sensitive data.
This means you can leverage AI capabilities to boost productivity and creativity while maintaining strong control and confidentiality over your business data.
================================
Like many conversations with Gemini, It tends to stray as we explore ideas and concepts
There was a lot more, but I think that I can use as I rewrite a new article exploring material use and Sequence of Operations.
But until then I want to build a Keynote presentation ready for my website Frog up or down.co.uk, https://sites.google.com/view/frogupordown/home , and more importanytly add a section into the Mitchalls Construction book, and the new book on Computing in Construction.
Share this post