12:33:44 From David Eddy to Everyone:
	MIT free classes on AI/ML/Neural/etc.
	
	https://www.youtube.com/watch?v=alfdI7S6wCY&list=PLtBw6njQRU-rwp5__7C0oIVt26ZgjG9NI&index=2
	
	https://introtodeeplearning.com
	
12:39:49 From Angel Vazquez-Salazar to Everyone:
	I was wondering if you quantized with qlora
	
12:40:04 From Angel Vazquez-Salazar to Everyone:
	FP16?
		
12:50:58 From Angel Vazquez-Salazar to Everyone:
	Now, have you tried the deep research like qwen where it’ll show “thinking…” and duration?
	
13:07:08 From J. Douglas Cryer to Everyone:
	Can you share that Instructions file?
	
13:10:55 From QMED to Everyone:
	Comprehensive Prompt Structure
	1. Greeting and Context Setting
	Purpose: Establish a friendly tone and provide background to frame the interaction.
	Example: "Hello Grok 3, I’m a software developer working on a 4D project, and I need help optimizing my code. Please assist me with the following task."
	Tips: Mention your role, goal, or the context (e.g., coding, research, casual learning) to help me tailor the response.
	2. Clear Objective
	Purpose: State exactly what you want to achieve to avoid ambiguity.
	Example: "I want you to review my code and suggest improvements to make it more efficient."
	Tips: Be specific about the outcome (e.g., "explain," "generate," "analyze," "debug") rather than vague requests like "help me."
	
13:11:02 From QMED to Everyone:
	3. Detailed Instructions
	Purpose: Provide step-by-step guidance or constraints to ensure accuracy.
	Example: "Please:
	Identify any redundant loops.
	Group variables of the same type under a single var directive with semicolons.
	Avoid changing the core logic of the code.
	Return only the modified variable declarations, not the entire method."
	Tips: Break tasks into bullet points or numbered steps. Include constraints (e.g., "don’t generate images unless I confirm") to prevent unintended outputs.
	4. Input Data or Examples
	Purpose: Supply the raw material (e.g., code, text, questions) for the AI to work with.
	Example: "Here’s my 4D code snippet:$tDataString:=""
	$oDataByWave:=New object()
	$colRows:=New collection()Modify the variable declarations as instructed."
	Tips: Paste code, text, or data directly, and format it (e.g., using code blocks) for clarity. If applicable, include sample outputs you expect.
	
13:11:18 From QMED to Everyone:
	5. Tone and Style Preferences
	Purpose: Specify the desired response style to match your needs.
	Example: "Respond in a concise, technical tone with bullet points, avoiding unnecessary elaboration."
	Tips: Options include "layman’s terms," "formal," "step-by-step," or "creative." This reduces the risk of responses being too complex or off-topic.
	6. Error Handling and Clarification Requests
	Purpose: Anticipate potential issues and ask for clarification if needed.
	Example: "If the code is incomplete or unclear, please ask me for more details before proceeding. If an error occurs, explain the issue and suggest a fix."
	Tips: This ensures I pause and seek input rather than guessing, minimizing mistakes.
	7. Follow-Up or Iteration Instructions
	Purpose: Guide how to handle subsequent interactions or refinements.
	Example: "After providing the modified code, ask if I’d like further optimizations or explanations."
	
13:12:21 From Bill James to Everyone:
	You can localize llama and DeepSeek on your own machine.
	
13:12:25 From Jason Patterson to Everyone:
	How does your code deal with an incorrectly formatted response from the LLM?
	
13:13:58 From Angel Vazquez-Salazar to Everyone:
	Bill, I have those models and they work well from my limited experience.
	
13:16:04 From Barclay to Everyone:
	You can also run an inference for DeepSeek/llama on a Vultr server for $10/month ... great for the open source LLMs.... https://www.vultr.com/products/cloud-inference/
		
13:21:15 From J. Douglas Cryer to Everyone:
	What service level do you have and what pricing for the Grok LLM you are using?
	
13:26:08 From Keith to Everyone:
	Presumably it's a terrible idea to actually directly execute anything coming back directly from the LLM?  eg: it might say to drop all tables
	
13:26:42 From Jason Patterson to Everyone:
	That was along the lines of what I was asking.
	
13:30:25 From Barclay to Everyone:
	Replying to "Presumably it's a terrible idea to actually direct...":
	You can do this by running function calls that you control and define. So, you are not directly executing. :-)
	
13:32:06 From David Eddy to Everyone:
	Thank you Ricardo!
	
13:33:54 From Yannick Payet to Everyone:
	Thank you Ricardo.
	
13:34:22 From Keith to Everyone:w
	Replying to "Presumably it's a terrible idea to actually direct...":
	Exactly.  Even with perfect instructions, you still wouldn't directly execute anything
	
13:34:44 From QMED to Everyone:
	Thank you Ricardo, very interesting!
	
13:39:09 From johnfoster to Everyone:
	Ricardo, impressive presentation and quite useful. Thanks!!!
	
13:40:28 From braeymaekers to Everyone:
	Thank you Ricardo, eye opening demo
	
13:40:36 From J. Douglas Cryer to Everyone:
	Thank you.
	
13:41:41 From uist to Everyone:
	Great brain food! Thanks
		
13:43:22 From Keith to Everyone:
	Thanks Ricardo, really useful
	
13:43:30 From Dani Beaubien to Everyone:
	Thanks for the very interesting presentation.
	
13:44:01 From Jean Graveline to Everyone:
	Thank's Ricardo
	
13:44:08 From Jason Patterson to Everyone:
	Thanks!
	
13:44:45 From Sandra to Everyone:
	Thanks for organising this (and the other sessions) Brent - really thought provoking.
	
