LLMs Usage FAQ - Revision history

Planetoid: /* How to Get Started with AI-Assisted Refactoring of Legacy Code? */

2026-03-05T05:41:49Z

How to Get Started with AI-Assisted Refactoring of Legacy Code?

← Older revision		Revision as of 13:41, 5 March 2026
Line 310:		Line 310:
	💬 Answer:		💬 Answer:

	Two talks in Mandarin from WebConf are well worth referencing:		Two talks in Mandarin from [https://webconf.tw/ WebConf] are well worth referencing:

	The first one is directly on point — the speaker shares a hands-on refactoring journey spanning decades of legacy system work, with collaborative notes and slides available. Although AI coding tools were part of the picture, the core message is that refactoring is fundamentally a software engineering discipline.		The first one is directly on point — the speaker shares a hands-on refactoring journey spanning decades of legacy system work, with collaborative notes and slides available. Although AI coding tools were part of the picture, the core message is that refactoring is fundamentally a software engineering discipline.

Planetoid: /* Same Model, But Does It Behave Differently via API vs. Web Interface (Subscription)? */

2026-03-05T04:19:58Z

Same Model, But Does It Behave Differently via API vs. Web Interface (Subscription)?

@@ Line 303: / Line 303: @@
 # Conversation Memory: For plain text conversations, a key difference is how context (conversation memory) is handled — the web interface manages it automatically on the provider's end, while the API requires developers to manage it themselves.
 # System Prompt: The web interface (e.g., Claude.ai) uses a system prompt maintained and regularly updated by Anthropic<ref>[https://platform.claude.com/docs/en/release-notes/system-prompts System Prompts - Claude API Docs]</ref>; with the API, the system prompt is entirely up to the developer.
 == Related articles ==

Planetoid: /* Does it make a difference to use the same model via API versus the web interface (monthly subscription)? */

2026-02-24T06:15:57Z

Does it make a difference to use the same model via API versus the web interface (monthly subscription)?

← Older revision		Revision as of 14:15, 24 February 2026
Line 293:		Line 293:


	== Does ~~it make a difference to use the same model~~ via API ~~versus the web interface~~ (~~monthly subscription~~)? ==		== Same Model, But Does It Behave Differently via API vs. Web Interface (Subscription)? ==
	📝 Question: Does ~~it make a difference to use~~ the same model via API ~~versus~~ the web interface (monthly subscription)?
			📝 Question: Does the same model behave differently when accessed via API vs. the web interface (monthly subscription)?

	💬 Answer:		💬 Answer:
	# The ~~monthly~~ subscription web ~~version~~ has a daily usage ~~limit;~~ once you hit it, you need to wait a few hours for it to reset. The API, on the other hand, ~~depends on~~ your budget ~~— it's more flexible~~, ~~allowing~~ more ~~requests~~ and ~~higher frequency~~.		# Usage Limits: The subscription-based web interface has a daily usage cap — once you hit the limit, you'll need to wait a few hours for it to reset. The API, on the other hand, is limited only by your budget, offering more flexibility such as higher request volumes and faster request rates.
	# Some parameters ~~can~~ only ~~be adjusted~~ through the API; the web ~~version doesn't necessarily expose them~~. Take Gemini as an example: even on the free or basic ~~plan~~, the web interface ~~has~~ far fewer adjustable parameters compared to Google AI Studio (which is closer to an API~~-style interface~~).		# API-Exclusive Parameters: Some parameters are only available through the API and may not be exposed in the web interface. Take Gemini as an example — even on the free or basic tier, the web interface offers far fewer adjustable parameters compared to Google AI Studio (which is closer to a raw API experience).
	# The ~~monthly~~ subscription web ~~version~~ is designed with general users in mind, prioritizing ease of use. If you need advanced features, ~~you can consult~~ the API documentation to see ~~if anything fits your needs~~.		# Target Audience: The subscription web interface is designed with general users in mind, prioritizing ease of use. If you need advanced features, it's worth checking the API documentation to see what's available.
	# For ~~pure~~ text conversations, ~~the main~~ difference ~~between the web version and the API~~ is~~: the web version has~~ context (conversation memory) ~~managed by~~ the provider, ~~whereas with~~ the API, developers ~~need~~ to manage it themselves.		# Conversation Memory: For plain text conversations, a key difference is how context (conversation memory) is handled — the web interface manages it automatically on the provider's end, while the API requires developers to manage it themselves.
			# System Prompt: The web interface (e.g., Claude.ai) uses a system prompt maintained and regularly updated by Anthropic<ref>[https://platform.claude.com/docs/en/release-notes/system-prompts System Prompts - Claude API Docs]</ref>; with the API, the system prompt is entirely up to the developer.

	== Related articles ==		== Related articles ==

Planetoid: /* Is There AI Software That Works Without Internet Connection */

2026-02-24T06:07:42Z

Is There AI Software That Works Without Internet Connection

@@ Line 290: / Line 290: @@
 (1) Use cloud models to search and compile data, then have local models perform secondary verification
 (2) For data that requires special privacy protection, it must be processed by local models
 == Related articles ==

Planetoid: /* Related articles */

2026-02-17T13:43:11Z

← Older revision		Revision as of 21:43, 17 February 2026
Line 296:		Line 296:
	* [[Troubleshooting of Google Gemini]]		* [[Troubleshooting of Google Gemini]]
	* [https://arxiv.org/html/2502.11959 STRIVE: Structured Reasoning for Self-Improvement in Claim Verification]		* [https://arxiv.org/html/2502.11959 STRIVE: Structured Reasoning for Self-Improvement in Claim Verification]
			* [[Generative AI glossary]]

	== References ==		== References ==

Planetoid: /* Is There AI Software That Works Without Internet Connection */

2026-02-12T05:11:59Z

Is There AI Software That Works Without Internet Connection

← Older revision		Revision as of 13:11, 12 February 2026
Line 271:		Line 271:
	== Is There AI Software That Works Without Internet Connection ==		== Is There AI Software That Works Without Internet Connection ==

	📝 ~~Problem~~: Is There AI Software That Works Without Internet Connection?		📝 Question: Is There AI Software That Works Without Internet Connection?

	💬 Answer: Offline AI Software Options		💬 Answer: Offline AI Software Options

	~~I've compiled~~ a [[Run_LLMs_locally \| list of software that can run AI models offline]]. I recommend trying Ollama or LM Studio.		Actually, there are quite a few software options that can run AI models offline ([[Run_LLMs_locally \| list of software that can run AI models offline]]). I recommend trying Ollama or LM Studio.

	~~Consideration 1: Adjust Performance Expectations~~		Considerations for Using Local Models

	~~If you're accustomed to cloud-based AI models, you may find it challenging to adapt to local model performance.~~		(1) Adjust Expectations for Local Models
	~~You'll need to lower your expectations~~ for ~~local models.~~

	~~Consideration 2: Model Selection and Practical Applications~~		If you're accustomed to using cloud-based AI models, you'll find it more difficult to adapt to local model performance. You need to lower your expectations for local models.
	~~Currently~~, ~~the most commonly used~~ local model ~~is gpt-oss:120b~~. ~~According~~ to ~~Jerry from Asia Pacific Intelligence, it lacks knowledge about Taiwan-specific topics. In my testing, it performs adequately~~ for ~~basic tagging tasks~~.

	Practical Work Applications:		(2) Model Selection and Practical Applications
	~~Assign~~ relatively simple tasks, such as:
	(1) Use cloud models to search and compile data, then have local models ~~verify it again~~		Currently, the local model I use most often is gpt-oss:120b. Jerry from Asia Pacific Intelligence shared at a Generative AI meetup that it lacks knowledge about Taiwan-specific topics. In my actual testing, it performs quite well for basic tagging tasks.
	(2) For ~~privacy-sensitive~~ data that must be processed ~~solely~~ by local models
			Practical Work Applications: I assign relatively simple tasks, such as:
			(1) Use cloud models to search and compile data, then have local models perform secondary verification
			(2) For data that requires special privacy protection, it must be processed by local models

	== Related articles ==		== Related articles ==

Planetoid: /* Solving ChatGPT Generated File Not Found Issues */

2026-02-12T05:02:22Z

Solving ChatGPT Generated File Not Found Issues

@@ Line 267: / Line 267: @@
 Please provide a CSV file using a Traditional Chinese filename.
 </pre>
 == Related articles ==

Planetoid: /* How to Resolve AI Models Generating Incorrect Website Links? */

2026-02-09T03:23:12Z

How to Resolve AI Models Generating Incorrect Website Links?

← Older revision		Revision as of 11:23, 9 February 2026
Line 208:		Line 208:

	💬 Response: If you want to use the original models rather than smarter reasoning models, you can ask the AI to add a preliminary step before giving conclusions: "Please extract and number relevant webpage paragraph text related to the question's answer, then answer the question based on these paragraphs." This can reduce the probability of hallucinations in less capable models.<ref>[https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/reduce-hallucinations Reduce hallucinations - Anthropic]</ref><ref>[https://the-learning-agency.com/the-cutting-ed/article/hallucination-techniques/ Improving AI-Generated Responses: Techniques for Reducing Hallucinations - The Learning Agency]</ref><ref>[https://www.godofprompt.ai/blog/9-prompt-engineering-methods-to-reduce-hallucinations-proven-tips 9 Prompt Engineering Methods to Reduce Hallucinations (Proven Tips) - Workflows] "Step-Back Prompting is a technique where you ask the AI to review its previous response and make sure it is accurate." </ref>		💬 Response: If you want to use the original models rather than smarter reasoning models, you can ask the AI to add a preliminary step before giving conclusions: "Please extract and number relevant webpage paragraph text related to the question's answer, then answer the question based on these paragraphs." This can reduce the probability of hallucinations in less capable models.<ref>[https://docs.anthropic.com/en/docs/test-and-evaluate/strengthen-guardrails/reduce-hallucinations Reduce hallucinations - Anthropic]</ref><ref>[https://the-learning-agency.com/the-cutting-ed/article/hallucination-techniques/ Improving AI-Generated Responses: Techniques for Reducing Hallucinations - The Learning Agency]</ref><ref>[https://www.godofprompt.ai/blog/9-prompt-engineering-methods-to-reduce-hallucinations-proven-tips 9 Prompt Engineering Methods to Reduce Hallucinations (Proven Tips) - Workflows] "Step-Back Prompting is a technique where you ask the AI to review its previous response and make sure it is accurate." </ref>

			== How to Resolve: AI Tools for Story-to-Visual Content Creation ==
			📝 Question:
			I have a story written in text form and would like to transform it into multiple cohesive images (not comic-style illustrations). What AI tools can help me accomplish this?
			Similarly, which AI platforms would be best suited for converting the same narrative into a series of short video clips?

			💬 Processing Methods:
			The method I use is more indirect, because image generation is very time-consuming, but generating text descriptions for simple images is relatively faster.
			So I use Claude or ChatGPT to generate text descriptions for images with continuous plot development.
			After making modifications, I then use AI models with character consistency to generate the images.

	== Can AI Self-Verify Its Reasoning Errors? ==		== Can AI Self-Verify Its Reasoning Errors? ==

Planetoid: /* How to Solve AI Models Generating Incorrect Website Links? */

2026-02-09T03:22:24Z

How to Solve AI Models Generating Incorrect Website Links?

← Older revision		Revision as of 11:22, 9 February 2026
Line 203:		Line 203:
	If the information is too new or niche to be obtained through real-time web search, it is recommended to use the second method of uploading knowledge files. Sometimes AI will claim to understand AABB while actually not understanding (pretending to know). In such cases, you can propose basic conceptual questions for verification.		If the information is too new or niche to be obtained through real-time web search, it is recommended to use the second method of uploading knowledge files. Sometimes AI will claim to understand AABB while actually not understanding (pretending to know). In such cases, you can propose basic conceptual questions for verification.

	== How to ~~Solve~~ AI Models Generating Incorrect Website Links? ==		== How to Resolve AI Models Generating Incorrect Website Links? ==

	📝 Question: Currently using AI tools like Grok, ChatGPT, and Perplexity to collect financial data and capital status of the top 10 global companies in specific industries, organizing them into table format. Additionally, I need to attach original website links as reference sources for the data. However, all three AI tools generate completely incorrect website links. Does anyone have solutions to prevent these language models from continuously producing incorrect reference links?		📝 Question: Currently using AI tools like Grok, ChatGPT, and Perplexity to collect financial data and capital status of the top 10 global companies in specific industries, organizing them into table format. Additionally, I need to attach original website links as reference sources for the data. However, all three AI tools generate completely incorrect website links. Does anyone have solutions to prevent these language models from continuously producing incorrect reference links?

Planetoid: /* How to Solve AI Forgetting Training Content */

2026-02-09T03:22:07Z

How to Solve AI Forgetting Training Content

← Older revision		Revision as of 11:22, 9 February 2026
Line 161:		Line 161:
	3. Switch the chat mode from "Ask" to "Agent" - this enables automatic project structure analysis. You can then ask any questions about the codebase		3. Switch the chat mode from "Ask" to "Agent" - this enables automatic project structure analysis. You can then ask any questions about the codebase

	== How to ~~Solve~~ AI Forgetting Training Content ==		== How to Resolve AI Forgetting Training Content ==

	📝 Inquiry:		📝 Inquiry: