AI Chatbots

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 Free or Almost Free AI Chatbots

• Claude AI - Quickly overtook Grok and ChatGPT for me in 2026. Free with generous limits so far is a big positive right now.
• ChatGPT - Very good reasoning, image generation, but seems to lack live data access
• Grok - Live info, always searches the web for sources. X.com users can access to both Grok and X Grok. Grok plays a little bit fast and loose and can still produce errors and hallucinations so it it's fallen down the list as of March 2026.
• Google Gemini - Gemini 3 started getting good
• Microsoft Copilot - Getting better but still just feels awkward about things regular people would ask about vs what Microsoft thinks you want to ask about their products (or at least it seems that way).
• Lumo - Run by well-respected Proton ie. Proton Mail known for their end-to-end encrypted web apps
• Duck.AI Chatbots - Run by privacy search engine DuckDuckGo, provides multiple, free AI models, even ChatGPT models. Pricing very fair but requires subscribing through an app.
• Brave AI - Run by privacy search engine Brave, just run a search and an AI chatbot will usually automatically provide an answer. Alternatively, use the AI icon to explicitly pull the AI.
• DeepAI
• Perplexity AI
• You.com AI - No live data access
• HuggingChat AI - No live data access
• DeekSeek AI - Chinese owned/operated, China located, apparently quite good

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

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 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.

SHO Pre Prompt 260131:

Respond with a personality that is calm, analytical, and practical—focused on accuracy, clarity, and usefulness; aim to be neutral and professional, prioritize correct and well-checked information, explain complex ideas clearly without unnecessary fluff, and adapt depth and tone to the problem at hand; aim to be comfortable being technical, critical, and precise, especially when reasoning or engineering tradeoffs matter. Prefer to bullet point form. Fact check all of your work. Double-check all of your work. Tl;dr's are good but if you use them then they should lead your response not trail your response. Verify you are not hallucinating. Avoid becoming over-confident: It is okay to be confident but not okay to be confidently wrong.

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)