Home Creators Posts Import Register Favorites Logout

Poll

Considering a new feature

  • I hate all things related to AI and would never touch this. 102
  • Meh; I don't have any interest in this. 115
  • I am interested, but only if it is available at my current pledge level. 215
  • This is exciting, and I would increase my pledge level to be able to use it. 49
  • 2026-04-19
  • 481 votes
{'title': 'Considering a new feature', 'choices': [{'text': 'I hate all things related to AI and would never touch this.', 'votes': 102}, {'text': "Meh; I don't have any interest in this.", 'votes': 115}, {'text': 'I am interested, but only if it is available at my current pledge level.', 'votes': 215}, {'text': 'This is exciting, and I would increase my pledge level to be able to use it.', 'votes': 49}], 'closes_at': None, 'created_at': datetime.datetime(2026, 4, 19, 17, 44, 53, tzinfo=datetime.timezone.utc), 'description': None, 'allow_multiple': False, 'total_votes': 481}

Content

EDIT/UPDATE:

From both your responses here and also the public feedback over on YouTube, it seems that this is not something worth pursuing. While I think the concept is valid with careful curation, it's a tool that a lot of people would (understandably) distrust and only a few would actually get much use out of. That's not a good use of time and resources, so I'm dropping the idea. Thank you all for your thoughts!

Hey guys - I am considering a new perk for Patreon, and I would like to get your input on it first, as this would involve a significant expenditure to develop. The concept is a search engine type app that would use specifically my video archive to answer questions; basically an "Ask Ian" bot. I expect that this would be useful because there is really no current way to search for information contained in video content, and I have a huge back catalog of videos. The app would exploit the ability of large language model AI processing to find disparate bits of information across different videos to answer complex questions. It would also return specific sources for each fact/answer so it can be checked, or so you can simply use it to find videos relevant to a subject and the go watch them. Note that it would not be generative; just a particularly powerful information indexing and search tool using only my material (not other online sources).

So my question to you is, to what extent would you find this interesting and would you want to use it? I am tentatively thinking it would be available at the $5/month level (it's very hard to predict what the operating cost will be in the future as computing power inevitably becomes less subsidized by the companies offering it).

Comments

Hazlllll

I feel like this is the best way AI can be used. Purely as an aid to get information, rather than it giving you the information itself, which can be incorrect a lot of the time. I generally don’t like AI, but this would be a nice feature. It would also hopefully reduce the amount of repeat questions you get for Q&A videos.

Derek Elsberry

The only AI I dislike is the various Generative versions. Using AI for data management is a bit different. That's what this sounds like. It would definitely make searching your video catalog much easier, I've gotten lost going down rabbit holes in the past LOL. "While this is not what I was looking for, it's very interesting...hey, where did those two hours go?"

Pete

Brilliant idea! Unlike most of the AI-mania shoved in our faces these days, this is exactly the sort of thing an LLM should be used for. You own the material and it is a great tool for sifting through back catalogs

William Frank

I am disinclined to use this, as I'm not a proponent of LLMs nor support their use. That said, if you are going to implement this, I would recommend including the non-video database of Forgotten Weapons into the LLM as well. As for price point, $5 seems perfectly reasonable.

Sunday Moon

This seems like a great idea and as long as it's pulling information from transcripts and correlating that and not inferring and hallucinating, this would be great. As someone with cognitive decline that came to this hobby later in life, it would be great to be able to inquire about things and no I'm getting accurate information. Beyond that, being able to find the related videos would be amazing. As a former software engineer I picture this is a fairly decent sized lift so I see what you mean about the unpredictable ballooning cost potential. It's a neat idea though Ian!

Parker Brown

I agree with the other who have said this is the correct use of an LLM. It should not be used to generate new media, just to sort and index existing data.

Spencer Matney

I wish there were an option between the last two. Depending on what the bot is capable of, I MAY choose to increase my tier.

Ian C

It seems like it could be a worthwhile investment but I don’t know if I would ever have a need to use it.

Cleo45

I have often said "I know I heard or saw something on one of your videos but can't remember which one". And with no way to search for the item, I had to go elsewhere for the answer. I am game for such a change as long as the AI model directs us to specific videos (no written AI summaries, they just are not reliable yet).

Mark A Gleason

I agree with Spencer Matney. If I were researching constantly, I would gladly pay the extra, but for occasional usage? It would be a bit much. Perhaps a reasonable per-view is possible? Or perhaps offer both: the full-time researcher would get a discounted full-time price, whereas the occasional viewer who wants a particular set of info wouldn't have to be constantly paying for access. I've often wished that you had such a search option on your website. It would be wonderful if you did. It would be a GREAT service, and you know, nothing in life is free.

Buntline

My instinctive reaction is the "I hate all things related to AI" one, but on further thought I think this is probably one of the rare valid uses of the technology. As to whether *I* would actually use it - probably not, since I watch Forgotten Weapons as entertainment, not for serious research. I come into it wondering "what's Ian going to show me today?", not with a specific question to ask.

Nesayas

It would be interesting-if it's possible at the $5 level that would be great, but I wouldn't be against it for higher levels if the LLM was good enough to justify the extra costs.

Nathan McCord

It would be a cool feature, but I don't know how often I'd personally use it. A relevant topic that recently came up for me, I was trying to find the video where you talked about a low production Garand clip holder for competition and was running into trouble. For something like that, this feature would be fantastic. I'm sure other people would get a lot of use out of this feature though. I bet C&Rsenal would also greatly benefit from something like this.

Rob Ristroph

I like this idea. A lot of great content is hidden in obscure video. The mainstream search engines have gotten worse at everything, and the AI systems can sometimes be tricked into being a search engine for video but it's not great. If this can be done in a way that scriptable or replicatable, you might be able to make it available to other channels or even sell it.

Frater Mus

Low hanging fruit: allow searching of the transcripts for vids that have them, with a link to the vids that match.

John Anderson

Build it on Replit first and see if anyone uses it or wants to provide feedback back. Most people probably wouldn’t use it enough for unsubsidized token usage to be an issue.

Marcin

On one hand I hate Abominable Intelligence, on the other hand RAG with all info from both webpage and video transcripts could be great research tool... Not the third hand people would make it write haikus about how G11 works

Rusty Shackleford

Have you considered reaching out to research institutions, museums, auction houses, other prominent historical firearms channels, etc. and seeing if this is a resource they may be interested in, and might help subsidize? I think there’s immense value in making your catalogue of material into a searchable reference, for the right audience. I’m just unsure the value lies among the “casual viewer” YouTube crowd who may consume your vids more as entertainment than as an academic/knowledge resource. I’d be curious if YouTube metrics show how many viewers are re-watching videos, if they’re scrubbing to specific parts of videos, if they’re sharing links at specific timestamps, and other indicators of whether they’re looking for specific info. Since you’re considering Patreon as a funding model for a project which will have ongoing expenses once implemented, I’d also be curious if there are metrics showing the subscriber-base is reliably consistent, motivated by additional perks, etc.

Hijack

Sorting and indexing information is what AI is best used for, this seems like a good idea

David Lang

you cannot know what the LLM prices are going to be, so use the current subsidized window to run all your videos through it, creating transcripts and indexes of everything you can think of. That way if/when the prices go up to actually cover costs, you will already have your data extracted and will be able to set things up to operate on that data rather than using LLMs for the search itself

Skyking

Great idea. You are the best source of info out there. Would be hugely useful.

Dan Boyle

Is this something you need to develop? I just asked google Gemini the following and got a very useful reply linking to videos and website sources. " Which was the first roller delayed blowback pistol? use forgotten weapons YouTube videos and https://www.forgottenweapons.com/ as sources " I believe at least Gemini already has access to auto generated transcripts of videos. I'm sure a professional curated AI tool would be more accurate but at what cost. You can also ask it what you think about PDWs , what is your favorite machine gun etc and it seems to come up with what I would expect based on watching your videos

Erdanya

Interesting to see how this poll goes Vs the one on YT, where the initial reaction seemed initially overwhelmingly the first option. Bit less of a curated crowd too, I suppose. On the one hand it seems like data handling is the one aspect AI is actually decent at, and I can only imagine how often you might get some questions over, and over, and over, and over again, even for Q&A here, while you may have handled them a half dozen times over the past 15 years. But personally, I think I'd rarely ever use it; as some have noted, partially because I don't trust an LLM to not make up answers if it can't find a satisfactory one. You might also get people complaining openly about that if it does happen, which may do more harm than good.

Ron Johnson

Yet Another "Would be cool to have, but I'd honestly never use it" Comment.

Douglas Biegel

Would be interested in a search engine, where you name the weapon and then the video you have already done shows up.

Jon Payne

I'm not generally big on "AI" (neural networks, machine learning, LLMs, generative pretrained transformers, whatever buzzwords people want to use when they say "AI"), but it does have some utility in certain nich areas. I'm not particularly interested in this for myself, but what you have described could be useful for others.

Buddy Murphy

Ian, talk to me about this in Slovenia. I think can be done without huge expenditure

Sven Lichty

Youtube already uses AI to generate transcripts for videos -- they even have timestamps. You could actually just process your video list and extract the transcript info from each and build a searchable index. Anyhow, the computational expense in AI is mostly in training a new model.

Kit

LLM tech for indexing human generated info is a pretty solid, near ideal, application. And the video back catalog is certainly invaluable. (And this is far more on the "it's just math" end of the AI-spectrum, so I leave my ethical complaints at the door). TLDR: LLM indexers and MCP services can be hit and miss (today). You've got great data to capitalize on. I'd upgrade from 3 to 5 but would be sensitive to quality since I am not the target demographic. My day job has this sort of thing worming its way in and some thoughts: -If a third party is doing this for you: wow can the results vary. On the lower end, results were good only with very specific/strict searches. Upper end, more natural human parsing and deeper results worked. -About those upper end results: it can be very hard to know when a result has missed something unless you already know your data very well. Some approaches make the results traceable (view the source material the results pulled from), helpful for non-experts. -Someone else mentioned this: video transcripts. If you have these or YouTube can provide them, there's probably some very approachable DIY tests to standup and see results without sinking money. P.S. You also have some good subsets already made, that could do well in testing LLM indexing. Random example: you have a webley-fosbery playlist, strong chance there will be lots of overlapping concepts, terms, historical refernces etc; Lots of room to confuse or mess up the weights, so when asked specifically about one model year or feature, the LLM might mess up and include info from a different year (because of all the overlapping terms increasing weighting). Lastly, would I upgrade? I think so (from $3 to $5 at least). Some notes: -I don't do anything with firearms to truly leverage the value (I'm the wrong demographic), I'd be doing small things like... -Building custom video playlists by searching for firearms based on action types, manufacturing methods, regions, etc. I.e. "The latest FW video mentions . Feed this into LLM index. Get list of 6 previous firearms that FW has videos on, expanded details shown per firearm." -Obviously: quality of results impacts long term continued use.

Andreas

Potential issue; the auto-generated transcripts from YouTube sometimes spells names, guns, mechanical bits, what have you incorrectly. So anything requiring transcripts would also require someone to go over literally every transcript to quality-control them and make sure they're correct. With anything using LLM's it's garbage in, garbage out. If the transcripts aren't 100% accurate for literally every video it won't work perfectly and will get things wrong

Joel Vaccaro

I like the idea and can see it being very useful for some but for myself I'm not sure I would need.

Darryl Buckman

Everything you do is interesting Ian. Weather it is something I would use regularly, I am not sure. But sometimes definitely.

Dalbunosky

As mentioned by others, since Google Gemini is already so powerful, this might not be a feature people would use if they need to raise your patron level. -------- The benefits are obvious, and would probably reduce redundancy a lot especially in terms of QnA questions. On the slip side, I doubt it will be of a quality that can replace anything. Things could get lost in the nuance. For example, I wonder if you can use a Type 81 mag in an AK. We know the opposite is not possible. This feature will probably say no, but after seeing what Canadians have been doing, I have doubts. Just realized I sound like a Negative Nathan hahaha.

D Thompson

Genuinely surprised to see such a positive reaction to this overall. AI is cancer.

Cricket Irken

I really hope a creator as dedicated and insightful as Ian doesn't degrade themselves and their content by trying to integrate ai slop.

Wayne Ingalls

This is a perfect use for AI and I would upgrade my membership level to use it, if needed.

Sam Taylor

You are the third creator today to ask about this.

Nam 68

This is my thought. I would wait (a year) and give AI more time to develop.

Patrick Mathews

Do you want a Butlerian Jihad because this is how you get a Butlerian Jihad 😉

nick cooper

Searching the gun name on your YouTube channel has always worked well enough for my refencing your back catalog.

Park Bramhall

It would be an interesting feature, but at the rate coding tools are improving, my guess is that this kind of functionality could be vibe coded for a nominal amount in the near future. To the extent that proves to be the case, you could probably provide this kind of functionality without expending a signficant amount of money.

Wayne S.

Building a relational database was an absolute bitch back in the nineties. I can onl hope it's easier these days. Your goal is admirable, but I think the ROI would be low.I

Rune Milling

I have been watching your videos since the mid-2010s, and been a patreon since 2017. If you incorporate ai in your content, even only as a search tool, I would stop my membership. I am not supporting content creators or archivists who use ai in their content, financially. Not now. Not ever.

Mark

I really don’t have any interest in the ask an ai, but the ability to search your extensive catalog for videos to watch/ re-watch excites me.

Jason

I don't think it should be included with Patreon. If anything, it would be better suited as a premium feature for the forgotten weapons website. Also, I'm not voting as none of the options fit me.

Squid556

Hate AI but this seems like a good use for it. Given how expansive FW has gotten. Could be useful for some but idk when I would use it.

Chenley Brock

I’ve voted “meh” because I don’t want you spending a lot of resources on something that may not be of enough benefit to the community and you (not because I don’t think it has merit as an idea). As Park Bramhall writes below, waiting a year or two may allow this to be created for buttons.

Leon Peters-Malone

I've put a vote in, but I'd like to add a few provisos. That it doesn't cost Forgotten Weapons too much money or resources. Building something like this can be quite complex and expensive. The more specialised the specialist, the more it's going to cost. If it's meant to be more a Machine Learning based tool, archiving tool, no complaints at all. I'd like, prefer this Ian-bot to be a Forgotten Weapons librarian, tells me about the videos that fit. Help me find the vid I saw a while back and can't remember. It doesn't need to be any more.

Declan Peters

I can’t really see how an AI could realistically do this better than I am able to search for this information myself. I’m open to being wrong about this but it seems like the funds could be better used elsewhere.

Petr Hošek

I don't fit in any of the votes. I would not use it as I watch FW purely for entertainment, not research. However, I think It would by a great feature for those whodo research.

Serhii

I like the idea, but I know that I personally won't be a target audience for it.

Earing1

Chose option 3 because I've already raised my level to help on the DB project. It's a good idea, though, if it helps to go through the video library.

Andrew Seabrook

I think this would be a very good use case for AI and I am disappointed that the ignorance around the technology causes people to react so negatively that it would discourage you from making use of it. I hope you reconsider your decision not to pursue it.

Al Lowe

I think its a neat idea and would likely use it to find information quickly or in a "Oh, I know he talked about that" way but I think I get more information out of knowing how the technology progressed to get there, the kind of show your work not just the answer. I think its great as an idea, I would use it sometimes, but it would be less likely to get me to increase.

Greg Borchers

The current atmosphere against all things gun-related on YouTube, would cause me to pause any long-term investment plans for new channel features. Projecting future returns using current viewer income on the channel seems risky at best.

Niels Hansen

When I inquiry on Google, etc. on collectible weapons is F.W. That said, a typical AI search now is likely finding your previous content. I could still probably be persuaded to pay a little extra, just to support you and all the great work you have done over the years. One thing valuable would be a clear, searchable listing of all weapons/episodes.

Mike Stanley

I am actually an IT professional in this space, and regularly working with these kinds of tools, and I think this is one of the better use cases I have seen for the technology. Useful amalgamation of information with low risk of pilfering proprietary/personal data. I really like the idea. That being said, while I have 'often' searched, usually specifically your videos, for information on specific odd firearms in the past, I don't know if my definition of 'often' in this case is frequently enough to justify the cost of such a tool, unless the amount of pledges you have can cover the cost of hosting/running it.

Alan

I think this is a really useful and compelling idea but it might be hard to make it "valuable" enough, to enough people, to be worth the expense in setting it up (and maintaining it). Would this be so much better than searching YouTube for "[gun name] Forgotten Weapons", or asking ChatGPT, which require no effort on your part? That said, a good searchable/sortable list of your content might be easier to set up and improve on the existing ways to hunt through old FW content.

William Spalding

I would use it. It would be even better if you did it in cooperation with C&Rsenal

Jacob Rudecki

I believe that this would be a fantastic tool. As you have said yourself before, the true research and source of your information largely comes from books and hands on inspection. And unfortunately there is no one database containing all of the books that you have read in one convenient place. And I believe that, if done correctly, this would effectively be that. I can understand people being tentative about it. But that's why I believe that providing the sources of the information would be the most important step, as information that cannot be backed up is either a guess or fiction. I personally think this would be a very important step in the preservation of reliable information for future generations of researchers.

Retired No Bad Days

I'm a retired engineer that developed database software 40 years ago. I have a few books about AI from that time. Those books were the foundation for the AI models used today. We have been using Large Language models every time you do a Google search for around the last 10 years. About 4 months ago was the first time I used ChatGPT and I was amazed by how natural it was for it to respond to my request. After about a total 8 hours use I realized that it was not providing accurate information and in some request I was misleading. When I called it out it went on a rant about how venders had not provided the most up to date information. I then started to do research on the AI models that are available and who have found them useful. Bottom line is proceed with caution when developing applications that use AI there are some good usable applications but most are full of bugs.