Searching Online: AI and Traditional

“How can I find ________?” Learning how to find what you’re looking for is a skill in itself. In ancient times, an understanding of card catalogs and Dewey Decimal Classification when using a library was essential, and is still useful today - for the gun people, see 363.x (Marksmanship), 623.x (Ordnance), 683.x (Firearms), and 799.x (Target Shooting) for starters. Knowing how to search online will help you find what you’re looking for by using the biggest source of information ever compiled in human history.

Thanks to J. C. Tate, CDR USN (Ret., Distinguished Rifleman), for his write-up on artificial intelligence. Download a copy:

Use more than one search engine. Google.com is the most popular and obvious but far from the only one. Others worth using are DuckDuckGo.com (which emphasizes user privacy), Luxxle.com (which emphasizes being unbiased), FreeSpoke.com (also emphasizes unbiased/uncensored results), SwissCows.com (user anonymity), Bing.com, Yahoo.com, AOL.com, Ask.com, Excite.com, and Lycos.com. While all search engines index publicly available websites, the approach each company takes in indexing and presenting search queries differs, which can lead to different results during a given search. If your first search comes up dry, try using different search engines.

Vary your search terms. Consider a search for specific parts. A good way to begin is by typing in the manufacturer name, model number, and any other part names or any particulars you know. For example, a search for Remington 700 parts for sale should get you a number of listings of parts sellers. You can then refine that by modifying keywords, such as replacing parts for more specific part names and variations. Look up the manufacturer’s schematic and/or owner’s manual to know the exact name the maker used to call their part.

Understand common search engine syntax. Surrounding a phrase in quotation marks prioritizes results having that exact word or phrase. Adding a “+” in front of a word indicates that it must be in the search result.

For example, a search for “Remington 700” specifies that exact phrase rather than just Remington 700 with Remington and 700 somewhere in the result but not necessarily together as an exact phrase.

“Remington 700” +parts is more specific still, with results having an exact match to the exact phrase “Remington 700” that also includes the word parts somewhere in the result. This is different from “Remington 700 parts” requiring that exact phrase together. Dropping the quotation marks indicates a search for results that have these words or numbers anywhere in the result and not necessarily together, which may or may not be what you want.

Exclude results by adding a “-” to a word or phrase. For example, Remington is also a brand name of electric razors and shavers. So, a search of Remington -shave would exclude any results having Remington that also have the word shave in it.

Searches can specify an OR operator. Separating search words with OR (it must be capitalized) or a pipe (typed as “|”) includes results that have any of the keywords. Consider the part that moves a shotshell from the level of the tube magazine up to be chambered. They can be referred to as lifters, carriers, or elevators depending on the make and manufacturer. Your search phrase could be shotgun lifter OR carrier OR elevator, which could also be typed as shotgun lifter | carrier | elevator.

Searches can also be confined to a specific site. Replacing https: or http: with site: as the protocol limits the search results to a specific website. If you’re only interested in results from GunBlabForum.com, try typing site:GunBlabForum.com Remington into the address bar of your browser to find pages, posts, or comments with the word Remington limited to that specific site.

Searches confined to main subjects. To search only for a topic that is the main subject of a page or result, use intitle. To search for news or web articles focused on Remington, try intitle:Remington. This will limit results to where Remington appeared in the title of the article or page and not just anywhere in the document. To limit this to a specific filetype, use filetype. A search of intitle:Remington filetype:pdf will only include results that are links to PDF files that have Remington in the title.

Ask smart questions. Don’t be afraid to ask for help in an online forum, but be smart about it. Before posting a question, search on your own and then include what you searched for, how you went about it, what the results were, and what you’re looking for but haven’t found. Before asking on a particular site or forum, do a site-specific search on that forum ( site:ForumName.com ) to double-check if your question or something similar hasn’t already been asked and answered. Include the fact that you did this to show that you were thorough in searching on your own before asking for help. This is part of learning how to ask questions the smart way.

It’s been said that cloud computing is just using someone else’s computer. Similarly, artificial intelligence is a way to automate synthesizing information that either uses someone else’s information already publicly available via search, or available information that’s been fed in. Training a Large Language Model (LLM) means using the provided data in a base model.

Ever since Hypertext Transfer Protocol supplanted Gopher protocol and other predecessors to become dominant in the 1990s, online search engines have used software to automatically gather (crawl or spider) large amounts of linked data online and present it when searched based on an estimated level of importance derived from that search engine’s particular algorithm; PageRank, SALSA, and HITS are prominent examples.

Traditional search simply presents available data as a response to a query, with an attempt to rank order it based on the algorithm used. The user has to read and interpret those results. AI adds to this by taking that found data and synthesizing it into a written narrative. This is both the good and bad news.

Think of AI as a very enthusiastic, fast, but dim-witted research assistant. AI can very quickly gather information, distill it, and present a response, but it doesn’t really understand, and the quality and accuracy of that presented response can vary based on a large host of factors.

I compared AI to a military friend as being similar to Public Affairs personnel or other media. They can visit your unit during your mission, gather info and quotes, take pictures and record video, and then create a news article or edited video about you efficiently. However, if Public Affairs doesn’t really understand you, your unit, or your mission, and if they remain unguided, their finished result will be filtered through their limited understanding and the directed narrative set by their boss/leadership. Their finished result may or may not be good, and someone reading/viewing it who also lacks knowledge and experience won’t know the difference. Here’s how that pans out with military marksmanship training:

AI can be very useful, but it can also automate Gell-Mann Amnesia. You can guide AI (and Public Affairs/media for that matter) by ensuring they have context and guidance from someone who does understand. John Tate’s excellent AI write-up (download link above) suggests Seven Prompt Disciplines.

The following seven practices were developed and validated through extensive real-world use, including legal research, historical analysis, medical queries, and technical troubleshooting. Each has a documented effect on output reliability.

Discipline 1: Require Authority Mapping, Not Content Generation. Instead of asking ‘What is the law on X?’ ask ‘Here are the specific propositions. Find authorities and citations for each.’ This shifts the model’s role from author to auditor. It forces retrieval-and-verification behavior rather than narrative synthesis.

Rule: Prompts that demand sources collapse the solution space and suppress invention.

Discipline 2: Pre-Commit the Factual Claims. Provide the claims yourself; ask the model to find support. When the model must guess what claims might apply, hallucination risk surges. When you supply the claims and ask for validation, the model operates under evidentiary constraints.

Rule: ‘Here is the claim — find support’ is safer than ‘What is the law on X?’

Discipline 3: Atomize Your Questions. “Atomic” here means a single, indivisible unit. Numbered, discrete questions prevent doctrinal bleed-through and compound hallucinations. Each proposition is independently testable. Cross-contamination between related but distinct rules is a major hallucination vector.

Rule: Atomized questions prevent compound hallucinations.

Discipline 4: Signal Your Level of Expertise. When a user signals subject-matter expertise, the model shifts from persuasion mode to precision mode. Use domain-specific vocabulary. Reference the expected standard (attorney-grade analysis, appellate brief standard). Ask for contested versus settled distinctions to be explicitly flagged.

Rule: When the user signals expertise, the model shifts from persuasion to precision.

Discipline 5: Use Neutral Framing. Adversarial prompts (’prove this is legal,’ ‘defend this position’) bias the model toward motivated reasoning and suppress acknowledgment of contrary authority. Neutral framing (’find authorities for these propositions’) produces more balanced and reliable output.

Rule: Neutral framing reduces motivated hallucinations.

Discipline 6: Allow Partial and Negative Answers. Binary framings (’is this legal or illegal?’) increase hallucination pressure because the model feels compelled to resolve ambiguity. Framings that explicitly permit ‘this is contested’ or ‘no reliable authority found’ dramatically reduce fabrication.

Rule: Prompts that allow partial validation reduce fabrication.

Discipline 7: Anchor Time and Jurisdiction. Unanchored questions invite reliance on outdated doctrine or confusion between jurisdictions. Always specify: the legal system, the temporal frame, and the applicable standard of authority.

Rule: Temporal and jurisdictional anchoring suppresses spurious sources.