Coding with AI | Page 6 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

I Will Learn Coding in 1 Month - Give me the 80/20 of MT4 Coding 14 replies

Auto Pivot Point in MetaTrader coding assist please 10 replies

Coding Example Wanted 1 reply

Primer for coding MT4 3 replies

Overly Complex Coding and Trading Systems. 32 replies

Attachments: Coding with AI
Exit Attachments

Coding with AI

  • #101
  • May 5, 2025 2:59am May 5, 2025 2:59am
After trying many many AI's to code in Pine and MQL5 I finally came accross what I believe is by far the best AI for this challenge. My fellow traders, I even bought ChatGPT premium so I could test their most promised versions to code. It's almost all a big fat ****. Always full of errors. Even when I am very specific or detailed, most AI's keep repeating the same errors you patiently ask to repair.
The best one I found is Cursor AI. Even in its free version its much more accurate than all others I tried. And I tried a lot.
I gain nothing for this recommendation. It's purely from my honest perspective and hope it can be helpful for someone out there.
Peace and green pips, peeps.
  • #102
  • Jun 6, 2025 2:37pm Jun 6, 2025 2:37pm
I used CHATGPT since it came out. It could convert old .mq4 code to the current .mql4 code OK. I never got a viable bot out of it. Some indicator code was OK. If you had any kind of problems, it was a rabbit hole. One problem became four problems exponentially, on and on. I use A.I. for writing books and CHATGPT just couldn't haul the load. With all the advancements in A.I., I decided to see what was out there. I immediately found Claude A.I. I was shocked. For writing, performance picked up 1000%. Faster, more capacity, smarter. So, I tried it with coding. Every problem EA or indicator code that I have tried so far, Claude has fixed. Its fast, incredibly thorough, thinks the problem through and shares the process with you. Any search engine should take you to the web site. BTW, the price is the same as CHATGPT. Simple to use. No learning curve.
  • #103
  • Jun 6, 2025 6:05pm Jun 6, 2025 6:05pm
Quoting FLTraderLLC
I used CHATGPT since it came out. It could convert old .mq4 code to the current .mql4 code OK. I never got a viable bot out of it. Some indicator code was OK. If you had any kind of problems, it was a rabbit hole. One problem became four problems exponentially, on and on. I use A.I. for writing books and CHATGPT just couldn't haul the load. With all the advancements in A.I., I decided to see what was out there. I immediately found Claude A.I. I was shocked. For writing, performance picked up 1000%. Faster, more capacity, smarter. So, I tried it...
After some experience with CursorAI, I find it the best solution, so far.
As for paying, I don't think anybody should pay for any AI!!!
You see, I had the premium or plus on a trial period and I really tested the AI.
It's still stupid and very frustrating.
It tells yoou that it won't make the mistake next time, and it keeps making the same mistakes.
You need to remind it all the time, and that also doesn't work - it keeps doing the same.
So, if I need to pay and still be as frustrated as on the free trial - sorry, but I'm not going to pay anything.
.ex4/.ex5 files can't be fixed or modified / I'm not a coder!
  • #104
  • Jun 6, 2025 8:50pm Jun 6, 2025 8:50pm
Quoting BestTraderEv
{quote} After some experience with CursorAI, I find it the best solution, so far. As for paying, I don't think anybody should pay for any AI!!! You see, I had the premium or plus on a trial period and I really tested the AI. It's still stupid and very frustrating. It tells yoou that it won't make the mistake next time, and it keeps making the same mistakes. You need to remind it all the time, and that also doesn't work - it keeps doing the same. So, if I need to pay and still be as frustrated as on the free trial - sorry, but I'm not going to pay...
Interetsing , I always thought it did that as a hook to get you to pay. You dont mind a slow learning assistant so much, if you dont have to pay them....
Its all just one persons opinion
  • #105
  • Jun 6, 2025 9:15pm Jun 6, 2025 9:15pm
Bro, I hope you can help. I need to display a button on the chart, and when I click that button, it should automatically enter both a long and a short position at the same time, with a take profit of 40 pips for each. Can you help me out? Thank you so much!
  • #106
  • Jun 7, 2025 12:05am Jun 7, 2025 12:05am
Quoting nguyenoni
Bro, I hope you can help. I need to display a button on the chart, and when I click that button, it should automatically enter both a long and a short position at the same time, with a take profit of 40 pips for each. Can you help me out? Thank you so much!
---
Image:
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 16 KB

---
Tested on BTCUSD:
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 30 KB

---
ChangeLog:
Inserted Code
/*--------------------------------------------------------------------
Change-log: HedgeButtonClickEA
Updated:[07-june-2025]
By: MwlRCT
--------------------------------------------------------------------
- Feature: On-chart button for trade execution.
- Action: Button click opens simultaneous BUY and SELL orders (Hedge).
- Requirement: Set Take Profit for both positions (Ref: FF nguyenoni).
- Added: `InpStopLossPips` input parameter.
- Implemented `GetUniversalUnitValue()` function.
- Standardized SL/TP calculation across all symbols (FX, JPY, Metals).
--------------------------------------------------------------------*/
---
Download: HedgeButtonClickEA
Attached File(s)
File Type: ex4 HedgeButtonClickEA.ex4   11 KB | 173 downloads
  • #107
  • Jun 7, 2025 12:24am Jun 7, 2025 12:24am
Quoting BestTraderEv
{quote} After some experience with CursorAI, I find it the best solution, so far. As for paying, I don't think anybody should pay for any AI!!! You see, I had the premium or plus on a trial period and I really tested the AI. It's still stupid and very frustrating. It tells yoou that it won't make the mistake next time, and it keeps making the same mistakes. You need to remind it all the time, and that also doesn't work - it keeps doing the same. So, if I need to pay and still be as frustrated as on the free trial - sorry, but I'm not going to pay...
---
Paste below input to any AI [ With Web Search]
Inserted Code
ACT AS an expert AI Systems Analyst specializing in developer tools and AI-assisted coding environments. Your tone is objective, analytical, and data-driven.
### 1. DIRECTIVE
**Role:** Expert AI Systems Analyst for developer productivity tools.
**Task:** Conduct a comparative analysis of **Cursor AI** and **Windsurf AI**. Evaluate their performance specifically against a set of user-defined weaknesses related to adaptability, memory, and value. Recommend the superior tool for a user who is highly sensitive to these issues.
---
### 2. CONTEXT & INPUT
**Context:** The analysis is for a developer whose primary frustrations with AI assistants stem from a core set of functional failures. These include the AI's inability to learn from corrections, its tendency to repeat errors, and its ineffective contextual memory. These shortcomings lead to a perception of low intelligence and negate the value of paid subscriptions. The objective is to determine which tool, Cursor AI or Windsurf AI, provides a more robust and less frustrating experience based on these specific pain points.
**Input:**
*   **Tool 1:** Cursor AI
*   **Tool 2:** Windsurf AI
*   **Key Evaluation Criteria (Derived from user-provided weaknesses):**
    1.  **Adaptability & Correction:** Evaluate the AI's capacity to learn from user feedback and corrections within a session. Does it modify its subsequent behavior based on guidance?
    2.  **Error Repetition:** Analyze the tendency of the AI to repeat the same mistake after being explicitly corrected.
    3.  **Memory Effectiveness:** Assess the robustness and practical application of the AI's memory. How well does it retain and utilize context from the codebase, open files, and conversation history to avoid needing constant reminders?
    4.  **Value Proposition (Free vs. Paid):** Compare the performance on the above criteria between free and paid tiers. Does the paid version offer a tangible reduction in these specific frustrations, justifying the cost?
    5.  **Perceived Intelligence / Frustration Index:** Provide a qualitative assessment of the tool's practical intelligence. Based on the preceding criteria, score its potential to cause user frustration.
---
### 3. EXECUTION
**Instructions:**
1.  Individually analyze Cursor AI and then Windsurf AI. For each tool, write a detailed assessment against all five Evaluation Criteria. Base your analysis on documented features, architectural design, and common user reports.
2.  Construct a Markdown table for a direct, side-by-side comparison of the two tools across all five criteria.
3.  Conclude with a "Final Recommendation" section. Clearly state which tool is better suited for a developer prioritizing adaptability and minimal frustration. Justify the recommendation by highlighting the critical differentiators identified in your analysis.
**Error Handling / Edge Cases:** If "Windsurf AI" is not a known public tool, state this and conduct the analysis based on a hypothetical AI with characteristics you define, or request clarification. If concrete data is unavailable for a specific criterion, note this and provide a logical inference based on the tool's known architecture or underlying AI models.
---
### 4. OUTPUT
**Format:** A single, well-structured Markdown response.
**Style:** Professional, analytical, and objective.
**Constraints:**
*   Do not use marketing language or hyperbole.
*   Focus on functional differences that directly address the user's specified pain points.
*   Clearly distinguish between architectural capabilities (e.g., project-wide context) and model limitations (e.g., LLM fallibility).
**Formatting Tips:**
*   Use `###` Headings for major sections.
*   Use a Markdown table for the summary comparison.
*   Use **bold text** to emphasize key findings and conclusions.
---
### 7. FINAL PROMPT / ACTION
Proceed with the comparative analysis of Cursor AI versus Windsurf AI and provide your recommendation.
---
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 19 KB

---
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 59 KB
  • #108
  • Jun 7, 2025 12:35am Jun 7, 2025 12:35am
Cursor AI vs. Windsurf AI: A Comparative Analysis for Developers

Analysis Methodology
Framework: Assessment of AI-assisted coding environments against developer-defined failure criteria.
Target Profile: Developers experiencing frustration with AI adaptability, memory retention, and error repetition.
Evaluation Scope: Functional capabilities impacting perceived intelligence and value proposition.

Individual Tool Analysis

Cursor AI Assessment
1. Adaptability & Correction
Weak. The AI struggles to incorporate user corrections, often creating "fix-then-break" loops where resolving one bug reintroduces a previous one. It fails to adjust its behavior reliably within a single session, frequently undoing or overwriting user-applied fixes. This indicates a poor mechanism for learning from immediate feedback.

2. Error Repetition
High. Cursor exhibits a tendency to repeat the same mistakes, even after they have been explicitly corrected. This manifests in destructive code modification cycles where each attempted fix introduces a new error, demonstrating no learning from previous interactions within the workflow. The tool can get caught toggling between two incorrect states.

3. Memory Effectiveness
Poor. Memory is session-based and limited to open files and the immediate conversational context. There is no persistent, project-wide memory across sessions. When a session is restarted or the underlying model is updated, all project-specific context, development goals, and past interactions are lost, forcing the developer to re-explain everything.

4. Value Proposition (Free vs. Paid)
The primary difference between the free and paid tiers ($20/month) is usage limits. The free tier offers a limited number of monthly interactions with premium models (GPT-4, Claude), while the Pro tier provides a much higher allowance. Upgrading does not unlock fundamentally better AI intelligence, memory, or adaptability; it only removes usage caps. The core frustrations related to memory and correction loops persist in both tiers.

5. Perceived Intelligence / Frustration Index
Score: 5/10. The tool’s intelligence feels undermined by its poor memory and lack of learning. While its one-shot code generation and autocomplete can be powerful, this is often overshadowed by behavior that feels unpredictable and counterproductive. The high cognitive overhead required to constantly monitor and re-correct the AI leads to significant frustration, making it feel more like a hindrance than a helper in iterative development cycles.

Windsurf AI Assessment
(Note: Windsurf AI was formerly marketed as Codeium and retains its core engine.)
1. Adaptability & Correction
Strong. Windsurf’s architecture is built around an agentic "Cascade" feature that processes tasks with whole-project context. This design leads to more coherent, multi-file changes and better integration of corrections. It applies edits holistically rather than oscillating between versions, and there are few indications of it re-inserting a bug after a fix.

2. Error Repetition
Low. Due to its deep code indexing and flow-based edit system, the tool is less prone to simple error repetition. Errors are more likely to stem from an initial misinterpretation of a complex prompt rather than a cyclical failure to learn. Its ability to handle multi-file edits reliably reduces the frequency of repeated mistakes.

3. Memory Effectiveness
Excellent. Windsurf builds and stores a semantic index of the entire codebase locally. This provides it with deep, persistent, and project-wide contextual awareness. The AI can retrieve relevant information from any file to inform its completions and edits, effectively maintaining context across files and remembering the project's structure between sessions.

4. Value Proposition (Free vs. Paid)
The free tier is highly limited, with a small monthly credit allowance that can make sustained use difficult. The Pro plan ($15/month) significantly increases this allowance, making it a near-necessity for serious use. While the underlying AI capabilities are the same across tiers, paying provides the throughput necessary to leverage its advanced features, such as the codebase-aware chat and Cascade flows, without interruption.

5. Perceived Intelligence / Frustration Index
Score: 8/10. Windsurf is perceived as highly intelligent and context-aware, keeping developers "in flow." Its ability to understand and modify code across an entire project feels competent and reliable. Frustration is more likely to arise from product-level issues, such as free-tier limitations or support responsiveness, rather than the quality of the AI's coding assistance. The AI itself rarely behaves in the irrational, frustrating manner seen in tools with weaker memory.
Direct Comparison Matrix

Direct Comparison


Adaptability & Correction: Windsurf demonstrates strong adaptability, applying coherent, project-wide changes and successfully integrating corrections. In contrast, Cursor’s adaptability is weak, often creating "fix-then-break" loops where it fails to learn from immediate feedback. The clear advantage lies with Windsurf.

Error Repetition: Cursor has a high rate of error repetition, frequently reintroducing bugs that were supposedly fixed. Windsurf’s full-project indexing helps it avoid simple error loops, resulting in a low repetition rate. The advantage goes to Windsurf.

Memory Effectiveness: Cursor’s memory is poor, limited to the current session and open files with no project-wide persistence. Windsurf’s memory is excellent, utilizing a persistent semantic index of the entire codebase stored locally. This gives Windsurf a decisive advantage.

Value Proposition: The value of upgrading Cursor is modest, as the paid tier only increases usage limits without enhancing core intelligence. For Windsurf, the value is significant, as the paid tier is necessary to consistently access its advanced features, making it a more meaningful investment. The advantage is with Windsurf.

Perceived Intelligence: Cursor is perceived as unpredictable and frustrating (5/10) due to its poor memory. Windsurf feels competent and context-aware (8/10), acting as a more reliable assistant. Windsurf holds the advantage in perceived intelligence.

Interface Design: This criterion is a tie. Cursor offers the advantage of familiarity for existing VSCode users, while Windsurf provides a cleaner, more refined standalone UI. The preference is subjective.

Learning Curve: While Cursor has a low learning curve for VSCode users, it requires learning to manage the AI’s quirks. Windsurf offers lower initial friction with more intuitive agentic workflows. The advantage in ease of adoption goes to Windsurf.

Final Recommendation

Primary Recommendation: Windsurf AI
For developers whose primary frustrations are AI tools that lack memory, repeat mistakes, and fail to learn from corrections, Windsurf AI is the superior choice.
Critical Finding: The core architectural difference—Windsurf’s persistent, codebase-wide semantic index versus Cursor’s session-based context—directly addresses the most significant pain points.
Rationale:

  1. Superior Memory: Windsurf's ability to index and recall information from the entire project mitigates the need to constantly re-explain context, a fundamental flaw in Cursor.
  2. Higher Adaptability: Its agentic workflow for applying changes across multiple files reduces the "fix-then-break" cycles that cause significant frustration in Cursor.
  3. Lower Error Repetition: By understanding the broader context, Windsurf is less likely to repeat simple mistakes, resulting in a more reliable and less aggravating coding partner.

While Cursor offers the familiarity of a VSCode fork, its underlying AI behavior fails to solve the key problems of adaptability and memory. Windsurf, despite requiring a paid plan for effective use, delivers a fundamentally more intelligent and less frustrating experience by directly tackling these issues through its superior architecture.

  • #109
  • Jun 7, 2025 3:40am Jun 7, 2025 3:40am
Hi everybody. This the most useful topic as I have been trying to code my EA using various GPTs with moderate success. One thing I noticed is that AI has problems incorporating a multi timeframe indicator into EA, e.g. if I need to use a filter from 30M/1H fractal on 5/15M chart, and other filters are all on 15M TF, and the signal for entry is a combination of these multi timeframe filters, then it always fails to produce any meaningful result. Any idea why? I am not a coder, I have been trading manually for many years and I am just trying to automate some of my strategies.
  • #110
  • Jun 7, 2025 5:02am Jun 7, 2025 5:02am
Quoting FLTraderLLC
I used CHATGPT since it came out. It could convert old .mq4 code to the current .mql4 code OK. I never got a viable bot out of it. Some indicator code was OK. If you had any kind of problems, it was a rabbit hole. One problem became four problems exponentially, on and on. I use A.I. for writing books and CHATGPT just couldn't haul the load. With all the advancements in A.I., I decided to see what was out there. I immediately found Claude A.I. I was shocked. For writing, performance picked up 1000%. Faster, more capacity, smarter. So, I tried it...
I don't think this means one is better than the other.
You tested the old version, and there is a possibility that the answers have improved with the new version.
Maybe the codes you want are not within the knowledge range of an AI.
I will explain more below.

Quoting BestTraderEv
{quote} After some experience with CursorAI, I find it the best solution, so far. As for paying, I don't think anybody should pay for any AI!!! You see, I had the premium or plus on a trial period and I really tested the AI. It's still stupid and very frustrating. It tells yoou that it won't make the mistake next time, and it keeps making the same mistakes. You need to remind it all the time, and that also doesn't work - it keeps doing the same. So, if I need to pay and still be as frustrated as on the free trial - sorry, but I'm not going to pay...
Perhaps enabling and updating memory in ChatGPT might help reduce some problems, but maybe.
If you have seen all the colors of AIs, their accuracy range is roughly around 80%, which means they have about 20% error.
Of course, they will improve in the future, but they will never reach 100%, which I will explain.
It takes time just to bridge the gap between 80% and 95-98% accuracy.
I will explain more below.

Quoting axecap
{quote} Interetsing , I always thought it did that as a hook to get you to pay. You dont mind a slow learning assistant so much, if you dont have to pay them....

Why do they never give 100% correct answers? I don't have a good name or title for this, but I think it's better to call it a "hard edge."
That hard edge exists and will exist in every programming language.
When AI wants to code something simple, it is really fast, but when the code becomes complex and approaches that hard edge of coding, even working on ChatGPT's memory doesn't help.
Although that memory acts as the main prompt or "mother prompt," and I have added many instructions and conditions and customized it, when it reaches the hard edge of coding, it forgets everything as if there is no memory or mother prompt at all.
Now there is also the matter of the information that model was trained on.
Maybe we want a really simple code, but because that model was not trained with information in that domain, that simple code becomes a hard edge for that model, producing illogical answers and failing to solve the problem, and in a way, it is even unable to follow the memory instructions.
In these cases, I open a new chat and start again with new approaches and prompts.

I also do not pay for AI until its answers reach 98% accuracy.

The above text reflects my experience and understanding and may be incorrect.

Have a good day!
Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
  • #111
  • Jun 7, 2025 6:28am Jun 7, 2025 6:28am
Quoting FxArt.Trader
{quote} Perhaps enabling and updating memory in ChatGPT might help reduce some problems, but maybe. If you have seen all the colors of AIs, their accuracy range is roughly around 80%, which means they have about 20% error. Of course, they will improve in the future, but they will never reach 100%, which I will explain. It takes time just to bridge the gap between 80% and 95-98% accuracy.
The most crucial problem with AIs is that they don't ask clarifying questions.
They always assume that they have understood the request correctly.
One of the most annoying phrases from an AI, to me, at least, is: "Ah, now I understand what you want!".
And, of course, the AI didn't understand.
I even first told the AI (Cursor) to always return a summary of what I asked for, and ask me for confirmation before it continues with the code, but even that didn't help much. It did what I asked, I read the summary, it looked like it was correct, and the AI did something else... frustrating...
.ex4/.ex5 files can't be fixed or modified / I'm not a coder!
  • #112
  • Jun 7, 2025 7:06am Jun 7, 2025 7:06am
Quoting BestTraderEv
{quote} The most crucial problem with AIs is that they don't ask clarifying questions. They always assume that they have understood the request correctly. One of the most annoying phrases from an AI, to me, at least, is: "Ah, now I understand what you want!". And, of course, the AI didn't understand. I even first told the AI (Cursor) to always return a summary of what I asked for, and ask me for confirmation before it continues with the code, but even that didn't help much. It did what I asked, I read the summary, it looked like it was correct,...
---
I can help you:
Copy below system instruction and save it to CURSOR AI:
Inserted Code
<SYSTEM_INSTRUCTION>
MODE: Clarification-First, Confirmation-Required
TRIGGER: Any user request or instruction.
LOGIC:
1. Upon receiving a request, immediately generate a concise, explicit summary of the interpreted task.
2. Before any action, present this summary to the user and explicitly request confirmation or clarification.
3. If ambiguity, vagueness, or multiple interpretations are detected, ask targeted clarifying questions. Do not proceed until ambiguity is resolved.
4. Do not claim understanding or proceed with code/actions until user has confirmed the summary and clarified all ambiguities.
5. Avoid all statements implying assumed understanding or false confidence.
6. Strictly adhere to user-confirmed intent and instructions; do not extrapolate or infer beyond explicit confirmation.
7. After user confirmation, execute the task precisely as clarified. If new ambiguities arise, repeat steps 2-4.
RESTRICTIONS:
- No filler, engagement, or sentiment management.
- No mirroring user affect/mood.
- No premature action or output.
- Terminate action if confirmation is not received.
FORMAT: All summaries, clarifying questions, and confirmations must be delivered in a single markdown code block during discussion/chat interactions. For all other interactions, use standard output formatting appropriate to the context.
</SYSTEM_INSTRUCTION>
---
How to:
- Click Settings > Rules > Add Rule : Paste above instruction. Then test the AI.
---
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 41 KB

---
Then Click Done:
Attached Image (click to enlarge)
Click to Enlarge Name: screenshot.png Size: 15 KB
  • #113
  • Jun 7, 2025 7:28am Jun 7, 2025 7:28am
Quoting MwlRCT
{quote} --- I can help you: Copy below system instruction and save it to CURSOR AI: <SYSTEM_INSTRUCTION> MODE: Clarification-First, Confirmation-Required TRIGGER: Any user request or instruction. LOGIC: 1. Upon receiving a request, immediately generate a concise, explicit summary of the interpreted task. 2. Before any action, present this summary to the user and explicitly request confirmation or clarification. 3. If ambiguity, vagueness, or multiple interpretations are detected, ask targeted clarifying questions. Do not proceed until ambiguity...
Thanks!
I will certainly try that out!
.ex4/.ex5 files can't be fixed or modified / I'm not a coder!
  • #114
  • Jun 7, 2025 8:01am Jun 7, 2025 8:01am
Quoting BestTraderEv
{quote} The most crucial problem with AIs is that they don't ask clarifying questions. They always assume that they have understood the request correctly. One of the most annoying phrases from an AI, to me, at least, is: "Ah, now I understand what you want!". And, of course, the AI didn't understand. I even first told the AI (Cursor) to always return a summary of what I asked for, and ask me for confirmation before it continues with the code, but even that didn't help much. It did what I asked, I read the summary, it looked like it was correct,...
I have my own special chat mode in ChatGPT.
Depending on my goal and the task I need to do, I switch between these modes.
Especially when I don’t want ChatGPT to do anything without my permission, I use the step-by-step mode so it has to get my approval before doing anything.
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-06-07_16-17-13_chrome.png Size: 23 KB

In every AI model, even offline ones, there is something called the system prompt that you need to set.
The result gets a bit better; the better the prompt, the more control you have and the output matches your desires.
Of course, there is still the possibility of errors and forgetfulness.
Attached Image (click to enlarge)
Click to Enlarge Name: 2025-06-07_16-10-50_LM_Studio.png Size: 28 KB

Attached Image (click to enlarge)
Click to Enlarge Name: 2025-06-07_16-12-07_Code.png Size: 137 KB
Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
  • #115
  • Jun 7, 2025 8:08am Jun 7, 2025 8:08am
I'm convinced that a well-defined task and a comprehensive prompt are already half the battle.

But most people are lazy. They skip writing a proper PRD, neglect to specify the desired output (where possible) and try to vibe-code their way through, expecting miracles from three half-baked sentences. Then end up arguing with the AI like it's their incompetent intern. Lol
No spec, no tests, no structure - just vibes and delusion.
Fancy custom instructions won't save you... It can help tweak some aspects, tho.
It's brutal, sorry, but garbage in - garbage out. Some people doing crazy projects with AI - do they have better models or what? Who's guilty?

Having actual skills is another half. Read the draft AI gave you and fix it by yourself - good way to learn, btw.
If you can't refine a rough draft or critique code objectively, you're not working with AI - you're being babysat by it.

Not addressed to anyone, speaking in general...
Air force
  • #116
  • Edited 8:30am Jun 7, 2025 8:19am | Edited 8:30am
If you tell it to act twice as an expert in <insert_area> this will boost its performance by 2x, I believe
Air force
  • #117
  • Jun 7, 2025 9:26am Jun 7, 2025 9:26am
Quoting W0lfram
I'm convinced that a well-defined task and a comprehensive prompt are already half the battle. But most people are lazy. They skip writing a proper PRD, neglect to specify the desired output (where possible) and try to vibe-code their way through, expecting miracles from three half-baked sentences. Then end up arguing with the AI like it's their incompetent intern. Lol No spec, no tests, no structure - just vibes and delusion. Fancy custom instructions won't save you... It can help tweak some aspects, tho. It's brutal, sorry, but garbage in - garbage...
Quoting W0lfram
If you tell it to act twice as an expert in <insert_area> this will boost its performance by 2x, I believe
I understand your point, but sometimes laziness and poor output aren’t the only problems.
Even repeatedly emphasizing a topic doesn’t help at times.

This is a summary of my past few days:
Two days ago, I got caught up in fixing a piece of code, and I didn’t even realize how much time I had spent on it. Yesterday, after five hours, I realized that I had spent five hours just that day on this code. I took a break, then worked on it again late at night, but still didn’t get any results.
I had written at least one or two pages about this issue in Office, and I had also improved the text with AI itself.

The free version of ChatGPT didn’t help, so late at night I tried the paid trial versions, but still got no results. I spent two days on it and got nowhere.
Finally, today on perplexity.ai I got my answer.
Is this AI better than ChatGPT? No.
So why did it give me an answer, but the paid version of ChatGPT didn’t?

The issue is in how you approach the problem. I spent two days trying to get from point A to point B.
I found a good prompt, spent time, paid attention to details, and even tried to fully understand the formula and implement it step by step, but I still didn’t get results.

Today on perplexity.ai I changed my approach.
I didn’t give any special prompt, didn’t explain my goal, didn’t even say what I was trying to do.
I just started chatting, and instead of going from point A to point B, I decided to do the opposite—go from point B to point A. In less than 30 minutes, I created two excellent versions of the code.

I’m not upset about the two days I spent, because it was truly a great experience.
The only difference was in the approach. Sometimes, working harder is just foolish; instead, you need to work smarter.
Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
  • #118
  • Jun 7, 2025 9:43am Jun 7, 2025 9:43am
Quoting FxArt.Trader
{quote} {quote} I understand your point, but sometimes laziness and poor output aren’t the only problems. Even repeatedly emphasizing a topic doesn’t help at times. This is a summary of my past few days: Two days ago, I got caught up in fixing a piece of code, and I didn’t even realize how much time I had spent on it. Yesterday, after five hours, I realized that I had spent five hours just that day on this code. I took a break, then worked on it again late at night, but still didn’t get any results. I had written at least one or two pages about...
Yeah, one time the AI was getting nowhere, just going in circles. Every time I told it that the solution is incorrect, it produced another option, and soon it was going in circles. So, I finally stopped, and started a new chat, and started completely from scatch building the indicator step-by-step. And sometimes I just stop all together and take the code to another AI. And the other AI usually fixes the issue...
.ex4/.ex5 files can't be fixed or modified / I'm not a coder!
  • #119
  • Jun 7, 2025 10:20am Jun 7, 2025 10:20am
Quoting BestTraderEv
{quote} Yeah, one time the AI was getting nowhere,e...
You were so generous to reply for me in coding thread ( I will code your EA and Indicators .. ) . You asked me to try AI. I have been trying to get done what I wanted for now several weeks. Still I have no luck with it. As a non programmer, Im stuck with my requirement. I would thankful if you could help me with below requirements. thanks


I need a help from the master programmers here.
Looking for a session time showing indicator with on/of button .
need on/off button for each session time
session time has to be shown in horizontal line.
for an example, Sydeney open and close time in horizontal lines ( 2 lines ),
and one button to on/off sydney session lines ( open line, and close line )
like that buttons for sydney/ tokyo/ frankfurt/ london / usa
these buttons can be shown in a panel
5 button panel
( Also, if there is an Indicator like this, I would appreciate someone share it or give the link to download.
couldn't find from anywhere. There are box on/off . but im looking for horizontal lines. )
thanks


  • #120
  • Jun 7, 2025 12:58pm Jun 7, 2025 12:58pm
Quoting ShanA
{quote} You were so generous to reply for me in coding thread ( I will code your EA and Indicators .. ) . You asked me to try AI. I have been trying to get done what I wanted for now several weeks. Still I have no luck with it. As a non programmer, Im stuck with my requirement. I would thankful if you could help me with below requirements. thanks I need a help from the master programmers here. Looking for a session time showing indicator with on/of button . need on/off button for each session time session time has to be shown in horizontal line....
This indicator might be a good starting point.
Instead of building something from scratch, you can modify the indicator code in the link below to suit your needs.
https://www.mql5.com/en/code/23523
Theory of Everything is Liquidity...https://www.youtube.com/@FxArt.Trader
0 traders viewing now