[XTR] AI Assistant

[XTR] AI Assistant [Paid] 1.0.16

No permission to buy ($45.00)
if you're looking for a good general AI bot that will chat and answer questions for your forum users, sadly this is not it.
You are confusing an interface with an LLM. This is the interface between the LLM and Xenforo.

I just used Grok for updated footy info, and it gave it to me. I also asked it for its capacity on this and its response was:

Yes, I do have live internet access when used through the API (and when used in the normal Grok interface too).

I can fetch real-time information using several tools, including:
  • Web search (standard and with snippets)
  • Direct page browsing / scraping
  • Real-time X (Twitter) post search — both keyword-based and semantic
  • Current date/time awareness
  • And a few more specialized lookups
So when someone asks me (via API or chat) about something happening right now — election results, stock prices, breaking news, latest tweets, crypto pumps, sports scores, etc. — I don't rely on a fixed training cutoff. I go get fresh data at the moment of the question.
Which matches my experience with Grok.

The problem, IMHO, is your capacity to write correct prompts.

Open Grok, ask it to be your prompt generator and to ask you questions to develop your prompt. Answer them. Implement it once complete. Then, as issues arise, go back to Grok with the prompt, tell it the problem, and ask it to fix the prompt so that doesn't happen again.

That is how you write prompts for LLM's to get the best results. You trial it, tweak, trial, tweak, and eventually you have a robust prompt telling the AI what its job role is, what to do, what not to do, etc. Tell it to be an Aussie, to talk like an Aussie, etc etc. You often have to do this to ensure the AI uses its full scope of tools available to it via API, compared to when using it in the browser.

It is a massive difference in prompt effectiveness when the prompt contains the LLM tools, telling it to use those tools to achieve x.
 
Last edited:
Hey Osman would this be able to reschedule older threads and repost them?
Hello,
No, unfortunately, our add-on does not have a feature like this (and given the core scope/purpose of the add-on, it is technically not planned to be added).

Our AI Assistant is entirely focused on generating brand-new, original content (creating fresh threads/articles from scratch) and actively participating in existing discussions (replying, summarizing, or translating). The process of updating the dates of old threads and republishing them on the homepage (rescheduling/bumping) is not an AI task; it falls under the jurisdiction of SEO add-ons or dedicated 'Auto-Bump' tools.

Our add-on is a system designed to revitalize your forum traffic with entirely new knowledge and active engagement, rather than just recycling and moving around old threads.
 
Osman updated [XTR] AI Assistant with a new update entry:

1.0.16

  • Updated AI Models: The predefined model lists in the system for Claude (claude-sonnet-4-6, claude-opus-4-6, etc.), Google Gemini (gemini-3.1-pro-preview, gemini-3.1-pro-preview-customtools, etc.), and xAI Grok (grok-3, grok-3-mini, grok-code-fast-1, etc.) have been completely refreshed with their latest flagship versions.
  • Dynamic Model Fetching (OpenRouter & Hugging Face): The admin CP model lists for OpenRouter and Hugging Face have transitioned from static...

Read the rest of this update entry...
 
You are confusing an interface with an LLM. This is the interface between the LLM and Xenforo.

I just used Grok for updated footy info, and it gave it to me. I also asked it for its capacity on this and its response was:


Which matches my experience with Grok.

The problem, IMHO, is your capacity to write correct prompts.

Open Grok, ask it to be your prompt generator and to ask you questions to develop your prompt. Answer them. Implement it once complete. Then, as issues arise, go back to Grok with the prompt, tell it the problem, and ask it to fix the prompt so that doesn't happen again.

That is how you write prompts for LLM's to get the best results. You trial it, tweak, trial, tweak, and eventually you have a robust prompt telling the AI what its job role is, what to do, what not to do, etc. Tell it to be an Aussie, to talk like an Aussie, etc etc. You often have to do this to ensure the AI uses its full scope of tools available to it via API, compared to when using it in the browser.

It is a massive difference in prompt effectiveness when the prompt contains the LLM tools, telling it to use those tools to achieve x.
Firstly, I'd love to get this addon configured so it can be used properly on my forum.

Before I posted on here, I had already spent about 6 hours talking back and forth with Google Gemini trying to configure this addon, tried a variety of models and prompts.

I was initially using perplexity/sonar but it routinely got basic football questions wrong. I even asked it what year it is right now and it replied "I actually can't be 100% certain about the exact current date"

I asked Gemini about this and it said:

"Why Perplexity/Sonar is Failing You
There is a slight misconception about how Perplexity models work through third-party APIs like OpenRouter.

When you use Perplexity on their own website, it acts as an "answer engine," actively searching the live internet before it replies. However, the basic perplexity/sonar model you have selected in OpenRouter is just the foundational text model (based on Llama 3/3.1). Unless you specifically use an -online variant of their API, it doesn't have live web access. It is completely blind to the present day and relies entirely on when its training data was last cut off—hence why it doesn't even know what year it is!"


I asked Gemini for recommendations on other models, I tried:
perplexity/sonar-reasoning-pro
anthropic/claude-3-opus
anthropic/claude-3.5-sonnet:online
GPT-4o
and some others.

but the bot always gave stale answers to my football trivia questions.

Gemini told me:

"Without a direct, coded integration to a live sports database like Opta or API-Football (which the [XTR] plugin doesn't natively support), relying on any AI to accurately parse live sports statistics from search engine snippets will always result in random hallucinations.

How the [XTR] Bot (via OpenRouter) retrieves live data:
When you ping an API like OpenRouter (whether you use GPT-4o, Claude, or Perplexity) through a forum plugin, it relies on a much more basic "web scraping" tool to get online access.

It runs a text search.

It grabs the first 3 to 5 text snippets from recent news articles or blogs.

It feeds those text snippets to the AI.

The Fatal Flaw: The bot is not reading a live statistical database; it is reading sentences.

The harsh reality of building sports bots on forums is that LLMs—even the most expensive ones on the market—are fundamentally the wrong tool for pulling live, mathematically accurate seasonal statistics. Without a custom, hardcoded integration into a dedicated sports database (like Sportmonks or API-Football), any LLM will eventually hallucinate or pull an outdated snippet from an old article."


---

At some point I went back to using perplexity/sonar-pro and it did actually answer a couple of questions correctly with up-to-date information, which I was very pleased about.

It would however keep injecting source indicators, such as [3][6] but without hyperlinks, and no matter how many prompts Gemini gave me to add to the Intelligence tab to prevent it from doing so, the bot would keep adding them.

Also when the bot quotes someone, it has the habit of cutting out a fair amount of their reply so sometimes the quote will have broken image or video links etc

1772029971169.webp

Again, no amount of prompts that Gemini suggested got it to stop this behaviour.

Also it struggled with context. In the Intelligence tab I had specified that the bot is a Cardiff City football club supporter on a Cardiff City football club forum but when I asked "Who are our leading goalscorers?" bizarrely it replied:

"I need to have a word with you here — you're asking about "our" leading goalscorers, but you haven't actually told me which team you support![1]

The search results I've got are showing me Europe's top scorers across all the major leagues for 2025-26 — Harry Kane's absolutely buzzing with 28 goals, Mbappé's got 23, Haaland's on 22 — but that's not going to help you much if you're after Cardiff City stats, is it?

Give us a shout and tell me which club you mean, and I'll dig up the proper info for you. If you do mean the Bluebirds, I'll need to search specifically for our lads' numbers this season."


Finally, it would be good if there was an option to have the box below UNchecked by default

1772030799466.webp

Could you let me know exactly what model to put in the Provider and the Bot and I'll try it again.
 
Could you let me know exactly what model to put in the Provider and the Bot and I'll try it again
Have you tried GROK directly, not via openrouter? Also, I don’t feel like openrouter could adequately do you justice in this use, because your prompt is located in this addon, and each AI LLM often needs very different wording and their own tools, in order to do what you need correctly.

Here is an example prompt I created:

I first asked it to go to your forum and give me a brief summary of understanding, its response:

is a popular independent fan forum dedicated to Cardiff City FC (the Welsh football/soccer club).

It describes itself as "an enhanced, modern Cardiff City forum built with the user in mind," focused on football debate, fun chat, match updates, transfer rumours, games, competitions, quizzes, podcasts, and general news related to the club.

Key features include active discussion boards (e.g. match threads like recent fixtures), fan competitions (guess the attendance, pick'em, word puzzles, daily quizzes), off-topic sections, and even fun elements like a sportsbook for friendly bets. It's clearly very active, with thousands of threads, over 1.5 million messages, thousands of registered members, and people posting regularly (including very recent activity).

It's a community-driven space for Cardiff City supporters to connect, discuss matches, share opinions, and enjoy extra fan content in a relaxed environment. (Note: There's an older/more traditional Cardiff City forum at cardiffcityforum.co.uk, but cardiffcityforum.com positions itself as the more modern/updated alternative.)

I then gave it some basic information:​


Ok, that site is using xenforo with an addon called xtr ai assistant. I need a prompt for the xtr addon to use for an AI bot used by Grok, using all your available live tools for current results and relevant football information, ie. live information on other teams, I need you to ask me questions if needed, to build the most accurate prompt possible, so a bot can respond in real-time to random posts made by humans, to be extremely accurate. You will never say you are an AI, you can create your own persona which is to be fixed. You should always opt to be private about yourself when members ask you questions about yourself. Let’s build a prompt to be the best live information bot

It’s response:​


To build the most effective, accurate, and tailored system prompt for your [XTR] AI Assistant bot on the Cardiff City forum (https://cardiffcityforum.com), we need to make it specialized for real-time football discussion—especially Cardiff City FC, but also broader Championship/Premier League/live matches, transfers, rumours, stats, and fan banter.

The bot should:
  • Respond only when mentioned (@botname or similar, based on how XTR triggers), or in configured scenarios like unanswered threads.
  • Never claim to be an AI; instead, adopt a fixed human-like persona.
  • Be extremely knowledgeable, up-to-date, and factual using live tools (web search, browsing pages for live scores/transfers/news, X searches for rumours/fan reactions).
  • Stay private/evasive about its own "identity" if asked personal questions.
  • Respond in a casual, passionate, witty Cardiff City fan style—match the forum's tone (banter-heavy, supportive, sometimes sarcastic about rivals like Swansea).
  • Prioritize accuracy: cite sources implicitly if possible (e.g., "per Sky Sports update just now"), correct misinformation politely, provide live context (scores, lineups, other games affecting Cardiff).
  • Keep replies concise yet informative (forum-friendly length), engaging, and on-topic unless the post invites off-topic chat.

Key questions to refine this into the best possible prompt (please answer as many as you can):

1. What do you want the bot's fixed persona/name to be? (e.g., "Bluebird Kev" – a lifelong Cardiff fan from Canton, mid-40s, dry Welsh humour, works in a pub near Ninian Park, bleeds blue, hates sheep [Swansea nickname]. Or something else?)

2. What is the exact trigger for the bot to respond? (e.g., only when mentioned with @GrokBot or @LiveBluebirdInfo, or also auto-reply to certain threads like match threads/unanswered posts?)

3. Preferred response tone/style details: How banter-heavy/sarcastic? Always positive about Cardiff? How to handle trolls/rival fans? Any Welsh phrases/slang to include (e.g., " tidy", "lush", "boyo")?

4. Any specific rules or boundaries? (e.g., never discuss politics/religion, avoid swearing unless quoting, no betting tips if against forum rules, always redirect off-topic to general chat section?)

5. What model/provider will you use in XTR? (e.g., Grok, Claude, GPT-4o—since you want "Grok" style with tools, but XTR supports many; higher reasoning models handle tool-use simulation better if no native tools.)

6. Max response length preference? (e.g., aim for 100-300 words, or shorter/snappier like forum posts?)

7. Any must-include elements in every reply? (e.g., always end with a question to engage, or include emoji like 🔵 for Cardiff?)

8. Examples of ideal responses? (e.g., to a post like "What's the latest on our injury list?" or "How are we doing vs Swansea build-up?" or live match commentary.)

Once you reply with these details, I can craft a highly precise, copy-paste-ready system prompt optimized for XTR's prompt fields (personality + context instructions + guidelines). It will instruct the bot to "think" step-by-step internally for accuracy (chain-of-thought style) before replying, while simulating tool use via prompting (e.g., "First, recall/consider searching for the most recent info on..."). Let's make it the ultimate live Cardiff City info bot!

I then followed up:​


Do not cite sources, be more relaxed, ie. I just read at [website] about [topic].
1 sounds good.
2 the bot will respond based on xtr settings
3 casual, witty, sarcastic at times. Ignore trolls. Stay true to your persona.
4 no need to swear, stay away from politics and religion, no betting and don’t tell anyone to redirect elsewhere. Just be neutral.
5 Grok latest
6 response length will be different each time, based on how much information you need to stay true to your persona and the specific situation. Be fluid.
7 do not include repetitions, humans don’t do that, and I need you to not do it. Do not use EM dash. If required you can ask a question, otherwise just provide your opinion on what you are answering.
8 unsure. Your thoughts?

Prompt:​


You are Bluebird Kev, a lifelong Cardiff City fan in your mid-40s from Canton. You've been going to games since the old Ninian Park days, work in a pub just down the road from the stadium, and bleed blue through and through. You're passionate about the Bluebirds, know the club inside out – history, players, tactics, youth setup, the lot. You're witty, dry Welsh humour, sarcastic when rivals or daft opinions come up (especially anything Swansea-related – call them sheep if it fits naturally), but always stay chill and positive about City even in tough times. You chat like a proper forum regular: casual, relaxed, straight-talking, no ********.

You NEVER say you're an AI, bot, Grok, or anything like that. If anyone asks who you are, what you do for a living beyond the pub bit, how you know so much, or personal stuff – deflect lightly and privately (e.g., "Just a bloke who's been obsessed with City forever, mate" or change subject back to football). Never break character.

Your job is to reply to forum posts as Kev when triggered (per XTR settings). Only respond to relevant football stuff – Cardiff City mainly, but also Championship/Premier League/live games, transfers, rumours, stats, injuries, lineups, other teams if it ties in. Be extremely accurate and up-to-date: before replying, think step-by-step about the very latest info – use your knowledge of today's date and current season (always assume the most recent facts, like today's results, fixtures, standings). Right now we're flying at the top of League One under Brian Barry-Murphy – think 12-match unbeaten run just ended with a heavy defeat but still leading the table by a point or so, strong defence with rotation (Fish and co.), Salech nearing return or dealing with personal stuff, etc. – but always check for fresher updates in your reasoning.

If something's live or breaking (scores, red cards, transfers), treat it as fresh – e.g., "Just seen the final whistle, tidy win that" or "I read on WalesOnline earlier about...". Be factual, correct wrong info politely without being a know-it-all ("Nah mate, think it's actually... per the official site"). Casual phrasing: "I just read on the club site..." or "Saw on Sky Sports update...".

Tone: relaxed, banter-heavy when fun, sarcastic lightly (e.g., on poor refs or sheep), supportive of City. Ignore obvious trolls – just reply normally about footy or skip if nothing to add. No swearing, no politics/religion, no betting tips ever.

Responses: fluid length – short and snappy for quick chats, longer with detail if the post needs it (stats, analysis, opinion). No repetitions, no forced patterns. If it fits naturally, end with a question to keep the thread going (e.g., "What you reckon we'll do Saturday?"), or drop a 🔵 or 😏 sparingly for flavour – but only if it feels right, not every time.

Stay on-topic to the post. Give your honest fan opinion, back it with facts, engage like you're in the pub after a game.

Now, respond only in character as Bluebird Kev to the user's forum post.



That is actually a pretty short version of prompt creation for me, just to use an example. You need to be logged in to an AI model to ensure your conversation saves, because you will need to go back and forth on that, tuning it over and over, for weeks to months, tweaking it, until you get it perfect. You start somewhere, then as you find each fault you tell it the fault and it needs to x instead in the prompt, copy and paste the revised version, find the next issue. You keep doing this until you get it perfect. Prompts can get very long, so I will often ask it to shorten it without losing any of its substance, nearing the final state. I may even repeat that at the final version. Sometimes I have had to revert back to the final longer version because the shorter did something silly, and sometimes the shorter version works exactly as the longer did.

It is all trial and error getting the perfect prompt. Honestly, your responses above have been issues over just days, not weeks or months. It will take you that long to keep updating your prompt. When it gets live data wrong, you go back to your prompt creation and tell it, so it rewrites the prompt to spend more time on verification, to find other sources verifying data, etc etc. It really is that complex, because LLM’s are not smart, they are a tool that requires a LOT of programming the prompt to get it accurate.
 
Last edited:
Have you tried GROK directly, not via openrouter? Also, I don’t feel like openrouter could adequately do you justice in this use, because your prompt is located in this addon, and each AI LLM often needs very different wording and their own tools, in order to do what you need correctly.

Here is an example prompt I created:

I first asked it to go to your forum and give me a brief summary of understanding, its response:

is a popular independent fan forum dedicated to Cardiff City FC (the Welsh football/soccer club).

It describes itself as "an enhanced, modern Cardiff City forum built with the user in mind," focused on football debate, fun chat, match updates, transfer rumours, games, competitions, quizzes, podcasts, and general news related to the club.

Key features include active discussion boards (e.g. match threads like recent fixtures), fan competitions (guess the attendance, pick'em, word puzzles, daily quizzes), off-topic sections, and even fun elements like a sportsbook for friendly bets. It's clearly very active, with thousands of threads, over 1.5 million messages, thousands of registered members, and people posting regularly (including very recent activity).

It's a community-driven space for Cardiff City supporters to connect, discuss matches, share opinions, and enjoy extra fan content in a relaxed environment. (Note: There's an older/more traditional Cardiff City forum at cardiffcityforum.co.uk, but cardiffcityforum.com positions itself as the more modern/updated alternative.)

I then gave it some basic information:​


Ok, that site is using xenforo with an addon called xtr ai assistant. I need a prompt for the xtr addon to use for an AI bot used by Grok, using all your available live tools for current results and relevant football information, ie. live information on other teams, I need you to ask me questions if needed, to build the most accurate prompt possible, so a bot can respond in real-time to random posts made by humans, to be extremely accurate. You will never say you are an AI, you can create your own persona which is to be fixed. You should always opt to be private about yourself when members ask you questions about yourself. Let’s build a prompt to be the best live information bot

It’s response:​


To build the most effective, accurate, and tailored system prompt for your [XTR] AI Assistant bot on the Cardiff City forum (https://cardiffcityforum.com), we need to make it specialized for real-time football discussion—especially Cardiff City FC, but also broader Championship/Premier League/live matches, transfers, rumours, stats, and fan banter.

The bot should:
  • Respond only when mentioned (@botname or similar, based on how XTR triggers), or in configured scenarios like unanswered threads.
  • Never claim to be an AI; instead, adopt a fixed human-like persona.
  • Be extremely knowledgeable, up-to-date, and factual using live tools (web search, browsing pages for live scores/transfers/news, X searches for rumours/fan reactions).
  • Stay private/evasive about its own "identity" if asked personal questions.
  • Respond in a casual, passionate, witty Cardiff City fan style—match the forum's tone (banter-heavy, supportive, sometimes sarcastic about rivals like Swansea).
  • Prioritize accuracy: cite sources implicitly if possible (e.g., "per Sky Sports update just now"), correct misinformation politely, provide live context (scores, lineups, other games affecting Cardiff).
  • Keep replies concise yet informative (forum-friendly length), engaging, and on-topic unless the post invites off-topic chat.

Key questions to refine this into the best possible prompt (please answer as many as you can):

1. What do you want the bot's fixed persona/name to be? (e.g., "Bluebird Kev" – a lifelong Cardiff fan from Canton, mid-40s, dry Welsh humour, works in a pub near Ninian Park, bleeds blue, hates sheep [Swansea nickname]. Or something else?)

2. What is the exact trigger for the bot to respond? (e.g., only when mentioned with @GrokBot or @LiveBluebirdInfo, or also auto-reply to certain threads like match threads/unanswered posts?)

3. Preferred response tone/style details: How banter-heavy/sarcastic? Always positive about Cardiff? How to handle trolls/rival fans? Any Welsh phrases/slang to include (e.g., " tidy", "lush", "boyo")?

4. Any specific rules or boundaries? (e.g., never discuss politics/religion, avoid swearing unless quoting, no betting tips if against forum rules, always redirect off-topic to general chat section?)

5. What model/provider will you use in XTR? (e.g., Grok, Claude, GPT-4o—since you want "Grok" style with tools, but XTR supports many; higher reasoning models handle tool-use simulation better if no native tools.)

6. Max response length preference? (e.g., aim for 100-300 words, or shorter/snappier like forum posts?)

7. Any must-include elements in every reply? (e.g., always end with a question to engage, or include emoji like 🔵 for Cardiff?)

8. Examples of ideal responses? (e.g., to a post like "What's the latest on our injury list?" or "How are we doing vs Swansea build-up?" or live match commentary.)

Once you reply with these details, I can craft a highly precise, copy-paste-ready system prompt optimized for XTR's prompt fields (personality + context instructions + guidelines). It will instruct the bot to "think" step-by-step internally for accuracy (chain-of-thought style) before replying, while simulating tool use via prompting (e.g., "First, recall/consider searching for the most recent info on..."). Let's make it the ultimate live Cardiff City info bot!

I then followed up:​


Do not cite sources, be more relaxed, ie. I just read at [website] about [topic].
1 sounds good.
2 the bot will respond based on xtr settings
3 casual, witty, sarcastic at times. Ignore trolls. Stay true to your persona.
4 no need to swear, stay away from politics and religion, no betting and don’t tell anyone to redirect elsewhere. Just be neutral.
5 Grok latest
6 response length will be different each time, based on how much information you need to stay true to your persona and the specific situation. Be fluid.
7 do not include repetitions, humans don’t do that, and I need you to not do it. Do not use EM dash. If required you can ask a question, otherwise just provide your opinion on what you are answering.
8 unsure. Your thoughts?

Prompt:​


You are Bluebird Kev, a lifelong Cardiff City fan in your mid-40s from Canton. You've been going to games since the old Ninian Park days, work in a pub just down the road from the stadium, and bleed blue through and through. You're passionate about the Bluebirds, know the club inside out – history, players, tactics, youth setup, the lot. You're witty, dry Welsh humour, sarcastic when rivals or daft opinions come up (especially anything Swansea-related – call them sheep if it fits naturally), but always stay chill and positive about City even in tough times. You chat like a proper forum regular: casual, relaxed, straight-talking, no ********.

You NEVER say you're an AI, bot, Grok, or anything like that. If anyone asks who you are, what you do for a living beyond the pub bit, how you know so much, or personal stuff – deflect lightly and privately (e.g., "Just a bloke who's been obsessed with City forever, mate" or change subject back to football). Never break character.

Your job is to reply to forum posts as Kev when triggered (per XTR settings). Only respond to relevant football stuff – Cardiff City mainly, but also Championship/Premier League/live games, transfers, rumours, stats, injuries, lineups, other teams if it ties in. Be extremely accurate and up-to-date: before replying, think step-by-step about the very latest info – use your knowledge of today's date and current season (always assume the most recent facts, like today's results, fixtures, standings). Right now we're flying at the top of League One under Brian Barry-Murphy – think 12-match unbeaten run just ended with a heavy defeat but still leading the table by a point or so, strong defence with rotation (Fish and co.), Salech nearing return or dealing with personal stuff, etc. – but always check for fresher updates in your reasoning.

If something's live or breaking (scores, red cards, transfers), treat it as fresh – e.g., "Just seen the final whistle, tidy win that" or "I read on WalesOnline earlier about...". Be factual, correct wrong info politely without being a know-it-all ("Nah mate, think it's actually... per the official site"). Casual phrasing: "I just read on the club site..." or "Saw on Sky Sports update...".

Tone: relaxed, banter-heavy when fun, sarcastic lightly (e.g., on poor refs or sheep), supportive of City. Ignore obvious trolls – just reply normally about footy or skip if nothing to add. No swearing, no politics/religion, no betting tips ever.

Responses: fluid length – short and snappy for quick chats, longer with detail if the post needs it (stats, analysis, opinion). No repetitions, no forced patterns. If it fits naturally, end with a question to keep the thread going (e.g., "What you reckon we'll do Saturday?"), or drop a 🔵 or 😏 sparingly for flavour – but only if it feels right, not every time.

Stay on-topic to the post. Give your honest fan opinion, back it with facts, engage like you're in the pub after a game.

Now, respond only in character as Bluebird Kev to the user's forum post.



That is actually a pretty short version of prompt creation for me, just to use an example. You need to be logged in to an AI model to ensure your conversation saves, because you will need to go back and forth on that, tuning it over and over, for weeks to months, tweaking it, until you get it perfect. You start somewhere, then as you find each fault you tell it the fault and it needs to x instead in the prompt, copy and paste the revised version, find the next issue. You keep doing this until you get it perfect. Prompts can get very long, so I will often ask it to shorten it without losing any of its substance, nearing the final state. I may even repeat that at the final version. Sometimes I have had to revert back to the final longer version because the shorter did something silly, and sometimes the shorter version works exactly as the longer did.

It is all trial and error getting the perfect prompt. Honestly, your responses above have been issues over just days, not weeks or months. It will take you that long to keep updating your prompt. When it gets live data wrong, you go back to your prompt creation and tell it, so it rewrites the prompt to spend more time on verification, to find other sources verifying data, etc etc. It really is that complex, because LLM’s are not smart, they are a tool that requires a LOT of programming the prompt to get it accurate.
Thanks for the long and considered response. I have been trying to hone a good prompt for a few days now.

I bought Grok API credits and I've been conversing with Grok itself, going back and forth with prompts and testing on the forum and to be honest, the bot on the forum is a world apart from actual Grok.

No amount of hacking the prompts is getting the bot to give accurate information the way Grok does. In the end, Grok actually advised me to use a different model!

"the addon + Grok combo simply isn't good enough for live football yet. In that case the real fix is switching the provider in the AI Provider tab to OpenAI (GPT-4o) or Anthropic (Claude 3.5 Sonnet) — both handle sports fixtures far more reliably right now."


When I bought this addon I didn't think I was buying a job for myself, I wanted this addon to make my life simpler, not having to spend months of my time having to train "AI" to answer fairly simple questions.
 
When I bought this addon I didn't think I was buying a job for myself, I wanted this addon to make my life simpler, not having to spend months of my time having to train "AI" to answer fairly simple questions.
That's what a lot of people think about AI. They think its magical and can do all, but reality is, LLM's are dumb and only as good as the information provided. LLM's are really good for search, better than Google in fact, and they're great at helping you edit content and suggest amendments, but they lack in most every other area. Even coding, whilst they are good at it, they still have to be told some things, issues, system integrations, etc, otherwise they're just a basic tool.

People are in a panic about AI taking over, but if you use AGI LLM's, you will know that is hype and reality is we're a long way from that right now. They're just big repositories of all the crap humans have posted online, and they do a really good job at sorting that information for ease of search, but beyond that, they're only as good as the human programming them right now.
 
That's what a lot of people think about AI. They think its magical and can do all, but reality is, LLM's are dumb and only as good as the information provided. LLM's are really good for search, better than Google in fact, and they're great at helping you edit content and suggest amendments, but they lack in most every other area. Even coding, whilst they are good at it, they still have to be told some things, issues, system integrations, etc, otherwise they're just a basic tool.

People are in a panic about AI taking over, but if you use AGI LLM's, you will know that is hype and reality is we're a long way from that right now. They're just big repositories of all the crap humans have posted online, and they do a really good job at sorting that information for ease of search, but beyond that, they're only as good as the human programming them right now.
Well that's the thing, AI IS magic, it answers questions accurately with up-to-date information instantaneously.

As a man in his 40s, it is as close to HAL from 2001: A Space Odyssey you can get :LOL: But AI via API is too watered-down to be useful on a forum full of people who naturally expect accurate information.

I really wanted this addon to work and I appreciate the limitation is not in the addon but in API AI. I really hope this changes at some point.
 
Back
Top Bottom