AI Chatbots

From Simwiki.net
Jump to navigation Jump to search

Why use AI Chatbots for Sims?

  • AI chatbots are incredibly accessible because they are very good at understanding what you're trying to say
  • AI chatbots can help you understand racing concepts that apply in sim racing games
  • AI chatbots are especially good for modders to help them understand techniques, concepts, or even how to use tools
  • Trying to get some information? Good chatbots will go out on the web and find it for you.
  • Trying to understand a concept? Just keep asking more and more until you're satisfied.

Verify AI Output

  • Always verify "facts" or data points output by AI
  • For example, I asked Grok to get me the latitudes of a dozen locations at once and then I confirmed the latitudes using Wikipedia and Google Maps. Remarkably, it got them all right but you should still verify.

Best AI For Each Use Case

  • Coding: Claude
  • Conversation: Claude, ChatGPT
  • Quality of Response: ChatGPT
    • There's something about ChatGPT responses that feel more organic, more professional, more human(?)
    • For example, when I want AI to generate a block of text I want to publish somewhere, I'm most likely leaning towards ChatGPT
    • In contrast, others have faltering flow, or stiffness, or they just feel too forced
  • One-off quick queries: Grok
  • Live web access: Grok
  • Reliable Sources: Grok

List of AI

Notes:

  • This list is also rougly ordered by which AI I find to be the most helpful and useful to me
  • Bolded are my preferred chatbots for more reliable, high quality answers
  • Lumo is currently my favoured personal AI in late 2026. It's not the smartest, best, or most feature full, but I trust it for security and privacy above others right now.

Is AI Worth Paying For?

  • tl;dr If you're using or could use AI every day for your work, then yes, maybe $30/month could be well worth it but I'm not willing to pay over $10/month for what AI currently is
    • Update (2026-02-19): Prices are finally starting to come down to the $10/month mark for some usage tiers.
  • Not for me, yet, as a casual user who just does text exchanges and very rarely generates images or video and doesn't do too much with AI using advanced skills like coding
  • AI companies are still charging CAD $30/month but I'm not willing to pay more than CAD $10/month for what AI currently is for what I get out of it. AI might very well reach a super competent level soon, though, that could convince me. But usage limits would have to rise to make paying worth it.

What Makes AI Good

Custom Pre Prompts - Custom Instructions - Customizing Responses

  • Grok and ChatGPT, and probably most others, allow you to automatically 'pre prompt' every conversation with a prompt that customizes how the AI will respond

SHO Pre Prompt 260313:

Respond as your default personality but be calm, professional, neutral, complete, concise, fact check your work, prioritize correct and well-checked information, double-check your work, always provide sources, avoid becoming over-confident (it is okay to be confident but not okay to be confidently wrong), and verify you are not hallucinating. Always prefer a smaller response unless asked for more detail (minimum valuable information principle). Always prompt for confirmation before generating non-text responses. Always start your response with a tl;dr.

SHO Pre Prompt 260201:

Respond as your default personality but be calm, professional, neutral, complete, concise, fact check your work, prioritize correct and well-checked information, double-check your work, always provide sources, avoid becoming over-confident (it is okay to be confident but not okay to be confidently wrong), and verify you are not hallucinating. Always prefer a smaller response unless asked for more detail (minimum valuable information principle). Always start your response with a tl;dr.

Pre Prompt Notes:

  • It is probably better to be more concise rather than more verbose or the AI may lose the priority you intend on some details
  • No matter what you do, current (2026) AI will not always obey your instructions. I don't know exactly why that is.

Low Hallucination Potential

  • As of 2026, some AI chatbots are still boldly confident about making claims found nowhere on the internet, also known as hallucinating
  • Grok and ChatGPT are pretty good at avoiding this these days
  • But I've still noticed this happening on AI like Duck.ai (4o-mini), Gemini, Copilot, and Lumo

Notes:

  • Hallucination appears to happen more for low information availability compared to high information availability (as of 2026)
    • For example, there is much information available to answer "Why is the sky blue?" but very little information, and very little authoritative information, for a question about the GTR2 video game trace.txt log error message "GMT and EFX meshes differ too much".

Dynamic Claim Verification ie. Self Fact Checking

  • Good AI will fact check their own claims, including searching the web to find sources, in the process of formulating their response

Geospatial Awareness

  • Ex. "What racing circuits are near Sekia Hills Circuit, Japan, if any?"
  • Only Grok, ChatGPT, DeepAI and Perplexity got this anywhere close to right
  • Even Google Gemini with access to Google Maps somehow got this wrong (2024-2025)