Veteran Disability Claims, someone able to create one?

Im Looking to work with someone to Create a Disabled Veterans Chat GPt Plugin. to help with Claims, appeals, and Overall Resources

I understand how it could take, But Understanding CFR’s isnt easy for Most Veterans. Image a Resource were Someone Could Put in there Issues and Direct them to the Part of the CFR that Cover those Issues. As well as access to Nation Archives to Verify service Records. Including historical Incidents, with Possibly reference nexus Letters.

Its not impossible.

2 Likes

This isn’t for me it would be for Others im bless with 100% T&P, Also Retired. I Just Want to help Others using the Power of AI.

2 Likes

That sounds a very good idea to help disabled veterans. But, based on your single sentence, an average joe wouldn’t be able to understand what tasks would be involved, what the pain points are in the current process, and how ChatGPT could possibly solve them.

A friendly suggestion: you should publish a paper or something on your website to describe the details of the project. If it is clear enough, you may even find volunteers to help you.

2 Likes

Thank you Great Idea!! i will start on the Paper ASAP

Piggybacking off of @EricGT, while this is absolutely a good and wonderful use-case of the technology, it is absolutely a monumental undertaking.

At a minimum I would expect you would need at least 5–10 domain experts, 2–3 people adept at training large language models, a few hundred-thousand to a few million dollars, and six-months to a year to get to a serviceable alpha product.

There will come a time when there will exist a base model which you can feed all of the documents, rules and regulations, laws, etc for a particular domain and a fully-formed expert-AI will spring forth—but we’re not there yet.

1 Like

@elmstedt Thanks for your thoughtfulness. Please let the man @stevensanders352 write his paper first, will you?

As I said, there are a lot of very competent volunteers (domain experts, developers) around for a good cause or a good use case.

If that can be made technically functional, some politicians may have the wisdom to see its value.

I’d be happy to put some hours in to get initial ideas and milestones fleshed out.

Indeed, UK grown. Should not be an issue creating technical milestones and such, it seems like an interesting challenge.

1 Like

It would be an excellent application to demonstrate accessibility features like this, it’s a great use case for a range of services. I can also see this being a framework for a range of assistance and care systems in other sectors.

This is overkill. At least for the domain experts portion. Not sure what you are thinking the money would be used for and how much (training, developers, sme, etc). My brother is currently going through his veterans claims again. He met a guy at the VA that helps veterans get their maximum disability they are owed. Most of my family is in the military (definitely was never in my career sights). The guy is helping him, my step dad, and my grandpa get their full benefits. So I think you will only require 1-2 domain experts.

The hardest part is getting the VA to cooperate, but he pretty much has a standard play book he goes through. He only charges 100 bucks up front to investigate and see if there is anything there worth pursuing. If there is, once you start the paperwork, a clock starts ticking on how much you are owed by the VA. If the VA grants you extra benefits after everything is done, they have to pay you all the back pay since you started the paperwork. The guy who helps them takes 10% of the backpay only.

For the most part, it isn’t really about laws or intricate loopholes. It is mostly just finding a strategy to deal with the VA, percentages, and what counts as a disability. It seems that it’s the VAs job to stop you from getting your benefits, or they are just completely lazy and don’t give a shit. Either way, that and having someone there telling the vets it’s ok to admit they have PTSD or are disabled is the hurdle.

Well, you need domain experts in many domains beyond simply navigating the VA.

When you start doing anything which involves medical information then you need to worry about HIPAA which is it’s whole own thing. Then you’ll likely need some type of medical experts to ensure the model is correctly interpreting medical information.

Then there are questions as to whether or not there are any special regulatory hurdles involved when dealing with data from the federal government.

There’s a world of difference between a guy at the VA who charges $100 to help people out and building a service on top of OpenAI’s models.

To ensure you don’t get immediately shut down you’re going to want to guarantee all of the I’s are dotted and T’s are crossed—for that you’re going to need the appropriate domain experts.

The money is for paying experts, collecting and cleaning data, testing and validation of the model, ensuring compliance with all applicable regulations, paying developers and hosting costs, etc.

Because at some point the model will make a mistake and everyone associated with the project will get sued—that’s just the reality of doing business in the United States.

So, the domain experts and the money are so you can do everything right so you’re covered when something goes wrong.

Sorry elmstedt, I know you’re trying to help, its ok, but I don’t think you have enough experience in how the VA works to make any useful inferences of what an AI VA assistant looks like. Most of the time, you’re not taking a lawyer through the door with you and nobody else is filling out your paper work for you. Most of the time you’re asking for your own documents, and you are filling out forms. Any AI model that would be useful for 90% of the Vets who aren’t getting their proper benefits would be more like a website navigation assistant than a lawyer. (Not exactly, but pretty close.)

It’s literally just administrative paperwork to get you in front of the right doctors and staff to get the wheels turning. @stevensanders352 If you are serious about making a chatbot to help vets navigate the processes in the VA, I can get you in contact. I’m sure you will have to have “CYA” insurance at some point as elmstedt has pointed out if you decide to make a bot that fills out the paperwork. But the first and foremost thing that frustrates vets is they don’t know where to start and no one at the VA is willing to help. I suspect your first MVP will be something that just helps them through the process.

You’ll have to have some sort of programming experience or a team willing to help. The gentleman I mentioned is a 25 year vet himself and he pretty much only helps vets get through the process now days. So he wouldn’t have the experience to make a bot himself. But he is always looking to help vets in anyway possible and I’m sure he’d be happy to help.

You can get some actual real experience from him as opposed to random guesses. He’s helped tons of guys and can let you know what the low hanging fruit is, and what is going to require lawyer level expertise. Like I said, I’d say 90% of the vets don’t require legal advise, they just need help navigating it.

As someone with quite a bit of startup experience under my belt and 15 years of experience in industries ranging from space to AI, I’d say shoot for the larger market of aimless vets and then you can expand to harder more impactful things like legal claims when you have product market fit. The absolute last thing you want to do is spend a “few million dollars” and a year on an idea that ends up not being too great. (Believe me I’ve been there and it sucks.) So shoot for solving that low hanging fruit problem and I think you can get some traction and help a lot of people along the way.

Hit me up in a DM for his contact info. He is based here in Las Vegas.

@stevensanders352 Just saw this message from Eric as well. While I wholeheartedly disagree with the reasoning, I whole heartedly agree it would be useful for getting you in front of the vet community. And it actually isn’t as hard as it sounds. Here in Nevada we have a program called APEX Accelerator, it helps Nevada businesses who are getting into federal contracts. The company my brother and I own is a federal contracting company (do not go that route, its a pain in the ass and lots of due diligence and money), we go to the events they host quite regularly. Just 2 weeks ago at UNLV we were talking to the director of the APEX accelerator and officials from the Department of Energy, JPL, NASA, and the City of Las Vegas.

I’m not sure of other states, but Nevada tries to have government/business interfacing events quite often and they are very supportive. There might be something in your area. If there is, go to them as often as possible and pitch this around to other vets. I’ve noticed a lot of these government officials love helping out people who are trying to help the country in some way.

So while I disagree with Eric’s reasoning, they are definitely going to be big advocates for you if you show real dedication to the effort.

You’re correct insofar as I’ve never had direct deadlines with the VA. Though I’ve done enough work involving PHI and other state and federal agencies that I feel comfortable extrapolating that the intersection thereof is likely a regulatory nightmare.

The big problem comes when the AI gives improper advice as to what to enter on a form which could carry with it substantial penalties if done so improperly.

Perhaps I’m wrong (and I hope I am) but I’ve never seen anything involving federal bureaucracy end up simple or straightforward. But, perhaps it really is as simple as getting them the right forms.

But, if that’s the case I’m not even sure where AI needs to come into it at all—it seems like a relatively simple heuristic would work just as well or better.

I see great value in embedding “all the things” putting every document and post regarding VA rules and regulations into a vector DB and then using the TopK retrieval result of those as context for a GPT4 query.

Then, if some tree of thoughts processing were to be done on that to reduce the possibility of hallucinatory responses making their way to the end user, I could see a 90-95th percentile accurate result, which would be far superior to anything that exists today.

You are correct, however, that the 5%-10% for which there is an erroneous result and potentially a costly one will require recourse. Such is the nature of a litigious environment.

I 100% agree with all of this, and I absolutely think this is a good idea for a project for anyone interested to take on.

I just also think OpenAI would be wary of allowing anything which touches on so many sensitive areas to be used as a plugin on top of GPT-4.

Especially if it’s not developed by a reputable organization.

Again, I am not saying a functional and useful product couldn’t be developed by a group of talented and motivated individuals, just that the calculus changes a bit when whoever is developing it is essentially judgment-proof.

1 Like

Doing it. Contact me manwell.clark@gmail.com. There are several connection points to make it happen. we are almost complete with our platform. The CFR is 1 of 17 tasks.

1 Like