Boglehead GPT

[Hayden]

I just went to the GPT store on chatgpt (https://chatgpt.com/gpts ) and searched for bogleheads, and found 6 bogleheads GPTs.

Not sure if you care, but apparently there are folks hoping to profit from bogleheads by building GPTs

[alexbogle]

In this case, it's just a long prompt being given to regular chat gpt right?

I am curious though, are dumps of this forum available? Are we allowed to use the data here for training a model?

[ehh]

Discussion from June 2023: viewtopic.php?t=406998

Alex Frakt, site founder wrote:

They are already scraping the site, or perhaps more accurately, are trained with data that has been scraped from the site. I've tested both chatGPT and Bard and the responses use phrases that clearly come from the wiki and the forum.

There's not much, if anything, we can do about it. Our situation is different from Reddit or Twitter in that we have no background APIs that allow third parties to extract data. All we provide is the web interface you are looking at. There's no way to stop anyone who wants it badly enough from downloading the site's contents one page at a time.

[Godot]

Discussion from June 2023: viewtopic.php?t=406998

Alex Frakt, site founder wrote:

They are already scraping the site, or perhaps more accurately, are trained with data that has been scraped from the site. I've tested both chatGPT and Bard and the responses use phrases that clearly come from the wiki and the forum.

There's not much, if anything, we can do about it. Our situation is different from Reddit or Twitter in that we have no background APIs that allow third parties to extract data. All we provide is the web interface you are looking at. There's no way to stop anyone who wants it badly enough from downloading the site's contents one page at a time.

Wow, brave new world.

[2pedals]

Discussion from June 2023: viewtopic.php?t=406998

Alex Frakt, site founder wrote:

They are already scraping the site, or perhaps more accurately, are trained with data that has been scraped from the site. I've tested both chatGPT and Bard and the responses use phrases that clearly come from the wiki and the forum.

There's not much, if anything, we can do about it. Our situation is different from Reddit or Twitter in that we have no background APIs that allow third parties to extract data. All we provide is the web interface you are looking at. There's no way to stop anyone who wants it badly enough from downloading the site's contents one page at a time.

Very sad statement, it seems that any AI should be “intelligent enough” to read, understand and follow the copyright rules posted on each web site.

rules#

[gips]

Imo, a lot of threads could be autofilled by ai. For example, every college thread results in the same basic points made, easy enough to fill in 100 or so posts with ai personas, ditto for how much gold to buy, ditto for tips funds vs individual tips, ditto for spias, etc. We could even have a special ai persona which encourages away anyone who posts extraordinary credentials (hedgedundie, swedoe, etc) as this too, has unfortunately become somewhat predictable.

I suppose a portfolio review ai would be genuinely useful.

Best,

[Claudia Whitten]

Nobody needs a Boglehead GPT. I just used regular ChatGPT to answer the question we all have: "how can I get rich?"

Here's the answer:

Becoming wealthy typically involves a combination of strategic planning, hard work, and sometimes a bit of luck. Here are some general steps that can help you on your path to financial success:

1. Education and Skill Development
Invest in Education: Gain knowledge in fields that have high earning potential or that interest you.
Develop Skills: Enhance your skill set to increase your value in the job market.
2. Career Advancement
Choose the Right Career: Select a career with good prospects and growth opportunities.
Network: Build a professional network to open up new opportunities.
Work Hard and Be Persistent: Success often comes to those who consistently put in effort and strive for excellence.
3. Entrepreneurship
Start a Business: Identify a gap in the market and create a business to fill it.
Innovate: Develop unique products or services that solve problems or fulfill needs.
Scale Up: Once your business is successful, look for ways to expand and grow.
4. Investment
Invest Wisely: Put your money into assets that have the potential to grow, such as stocks, real estate, or mutual funds.
Diversify: Spread your investments across different asset classes to mitigate risk.
Stay Informed: Keep up-to-date with market trends and adjust your portfolio as needed.
5. Financial Management
Save and Budget: Live within your means and save a portion of your income.
Avoid Debt: Minimize high-interest debt and pay off any existing debts.
Build an Emergency Fund: Have savings set aside for unexpected expenses.
6. Mindset and Lifestyle
Set Goals: Establish clear financial goals and create a plan to achieve them.
Stay Disciplined: Stick to your plan and avoid unnecessary spending.
Continue Learning: Always seek to improve your knowledge about personal finance and investments.
7. Leveraging Opportunities
Seize Opportunities: Be on the lookout for new opportunities, whether it's a job promotion, a new investment, or a business idea.
Adapt to Change: Be flexible and ready to pivot when new trends or technologies emerge.
8. Building Multiple Income Streams
Side Hustles: Take on additional work or freelance projects to increase your income.
Passive Income: Create sources of passive income such as rental properties, royalties, or dividend-paying stocks.
9. Long-Term Thinking
Be Patient: Wealth-building is usually a long-term process.
Think Generationally: Consider strategies that will benefit not only you but also future generations.
10. Health and Well-being
Stay Healthy: Good health is crucial for maintaining the energy and focus needed to achieve financial goals.
Balance: Ensure you have a balance between work and personal life to avoid burnout.
While there are no guarantees, following these steps can help you build a solid foundation for financial success.

Case closed.

[ReadyOne]

Imo, a lot of threads could be autofilled by ai. For example, every college thread results in the same basic points made, easy enough to fill in 100 or so posts with ai personas, ditto for how much gold to buy, ditto for tips funds vs individual tips, ditto for spias, etc. We could even have a special ai persona which encourages away anyone who posts extraordinary credentials (hedgedundie, swedoe, etc) as this too, has unfortunately become somewhat predictable.

I suppose a portfolio review ai would be genuinely useful.

Best,

One purpose of this site is to apply information, data, and crowd wisdom to specific situations. That's the beauty of the post-and-reply format. AI often avoids specificity, but this site allows for it.

When people reply, you can usually determine who has a similar worldview or approach to issues, making those responses especially valuable. For example, on college threads, views range from laissez-faire to much more hands-on. If you prefer a hands-on approach, laissez-faire responses tend to be unhelpful and lecturey, and vice versa.

[kelway]

Very sad statement, it seems that any AI should be “intelligent enough” to read, understand and follow the copyright rules posted on each web site.

rules#

For me, the goal here is to educate, and aren't you ultimately glad that AI tools are learning from the best? I can't imagine why anyone would want to prevent it.

[gips]

Imo, a lot of threads could be autofilled by ai. For example, every college thread results in the same basic points made, easy enough to fill in 100 or so posts with ai personas, ditto for how much gold to buy, ditto for tips funds vs individual tips, ditto for spias, etc. We could even have a special ai persona which encourages away anyone who posts extraordinary credentials (hedgedundie, swedoe, etc) as this too, has unfortunately become somewhat predictable.

I suppose a portfolio review ai would be genuinely useful.

Best,

One purpose of this site is to apply information, data, and crowd wisdom to specific situations. That's the beauty of the post-and-reply format. AI often avoids specificity, but this site allows for it.

When people reply, you can usually determine who has a similar worldview or approach to issues, making those responses especially valuable. For example, on college threads, views range from laissez-faire to much more hands-on. If you prefer a hands-on approach, laissez-faire responses tend to be unhelpful and lecturey, and vice versa.

Sure, but all those views could be expressed by different ai persona. And easy enough to replicate all the rabbit holes the college threads go down. Because we have so many people commenting, every view gets expressed every time.

To your exact point, the ai could ask the poster questions to determine if they are hands on or laisse-faire and simply post whatever is going to confirm their bias.

[2pedals]

Very sad statement, it seems that any AI should be “intelligent enough” to read, understand and follow the copyright rules posted on each web site.

rules#

For me, the goal here is to educate, and aren't you ultimately glad that AI tools are learning from the best? I can't imagine why anyone would want to prevent it.

Can you imagine the potential that the makers of AI tools have to produce a tool that doesn't follow copyright laws? Do you believe copyrights should protected?

[kelway]

Can you imagine the potential that the makers of AI tools have to produce a tool that doesn't follow copyright laws? Do you believe copyrights should protected?

Yeah but I don’t see this as a copyright issue. Click the search icon above and it leads to Google search of Bogleheads. For me it’s the same.

[Jacotus]

What is the copyright policy of this forum?

Edit: Found it here: rules#section-8.

Looks like this potentially runs afoul of copyright infringement, given ongoing lawsuits and the fact that OpenAI & others are licensing content. Whether there is anything that can be done is another matter. Perhaps DMCA takedowns issued to the specific GPT on the GPT Store? https://arstechnica.com/tech-policy/202 ... textbooks/

[Whakamole]

Bogleheads is a registered trademark and it should be removed from ChatGPT unless those bots are authorized by the Center.

[ReadyOne]

Imo, a lot of threads could be autofilled by ai. For example, every college thread results in the same basic points made, easy enough to fill in 100 or so posts with ai personas, ditto for how much gold to buy, ditto for tips funds vs individual tips, ditto for spias, etc. We could even have a special ai persona which encourages away anyone who posts extraordinary credentials (hedgedundie, swedoe, etc) as this too, has unfortunately become somewhat predictable.

I suppose a portfolio review ai would be genuinely useful.

Best,

One purpose of this site is to apply information, data, and crowd wisdom to specific situations. That's the beauty of the post-and-reply format. AI often avoids specificity, but this site allows for it.

When people reply, you can usually determine who has a similar worldview or approach to issues, making those responses especially valuable. For example, on college threads, views range from laissez-faire to much more hands-on. If you prefer a hands-on approach, laissez-faire responses tend to be unhelpful and lecturey, and vice versa.

Sure, but all those views could be expressed by different ai persona. And easy enough to replicate all the rabbit holes the college threads go down. Because we have so many people commenting, every view gets expressed every time.

To your exact point, the ai could ask the poster questions to determine if they are hands on or laisse-faire and simply post whatever is going to confirm their bias.

Yes, as I understand it, AI companies have tried to offer generalized information vs. specific advice. I assume for liability concerns.

You and I have posted over 2000 times. It is unlikely that a unique thought has been expressed in any of the 2000. Why do we bother? I assume it is the need to participate, be helpful, or be social.

[gips]

Agreed, and that of course is the downside to my idea of populating topics with ai posts, before long no one would be posting. Still, perhaps, we’ve just glimpsed the future, or lack of a future, for Internet forums.

[010101]

Very sad statement, it seems that any AI should be “intelligent enough” to read, understand and follow the copyright rules posted on each web site.

rules#

For me, the goal here is to educate, and aren't you ultimately glad that AI tools are learning from the best? I can't imagine why anyone would want to prevent it.

Can you imagine the potential that the makers of AI tools have to produce a tool that doesn't follow copyright laws? Do you believe copyrights should protected?

There's nothing about the current copyright policy that says that you can't train a model off the content. Yes, if someone were to copy/paste a post word-for-word, that would run afoul of copyright. But if someone trains a model and spits out original content - that's ok. If the model spits out a phrase that is used repeatedly by bogleheads (e.g. "past performance does not predict future performance") - that's ok.

When it comes to copyright, it's the output that is important - not the method. If a human were to read all boglehead posts and learn from those posts -- we wouldn't accuse that person of copyright infringement if they go on to answer someone's question (even if paid to answer the question). If a computer does the exact same thing as a human - it shouldn't matter.

I know people like to exclaim "but this time it's different!" -- but it really isn't.

[jebmke]

Very sad statement, it seems that any AI should be “intelligent enough” to read, understand and follow the copyright rules posted on each web site.

rules#

For me, the goal here is to educate, and aren't you ultimately glad that AI tools are learning from the best? I can't imagine why anyone would want to prevent it.

Can you imagine the potential that the makers of AI tools have to produce a tool that doesn't follow copyright laws? Do you believe copyrights should protected?

There's nothing about the current copyright policy that says that you can't train a model off the content. Yes, if someone were to copy/paste a post word-for-word, that would run afoul of copyright. But if someone trains a model and spits out original content - that's ok. If the model spits out a phrase that is used repeatedly by bogleheads (e.g. "past performance does not predict future performance") - that's ok.

When it comes to copyright, it's the output that is important - not the method. If a human were to read all boglehead posts and learn from those posts -- we wouldn't accuse that person of copyright infringement if they go on to answer someone's question (even if paid to answer the question). If a computer does the exact same thing as a human - it shouldn't matter.

I know people like to exclaim "but this time it's different!" -- but it really isn't.

Like music derivative suits it will probably come down to juries deciding if there was sufficient transformation or not.

IMO where the tort rubber is going to meet the road is S 230. Hard to defend 230 protection.

[techbud]

I'm going to ask a related but more specific followup question: Is it possible to use a generative AI tool (ChatGPT, Copilot, etc) to summarize a long discussion thread, and to be able to ask specific questions against that thread?

Here's my example: I'm interested in the discussion about the Retiree Portfolio Model. This particular discussion has almost 1900 posts spanning over 38 pages. I have to be honest, I am not going to read or review all 1900 posts before asking a question. Sure, I could do a search, but that still would require me to come up with the right keywords and synonyms, and would require me to wade through potentially many results.

It would be really useful if I could somehow point ChatGPT or Copilot at this thread, and then ask it a specific question.

Is this possible? (Desirable?) Anyone know how to easily do it?

[nisiprius]

...But if someone trains a model and spits out original content - that's ok...

That would be nice if AI always spat out original content, but it doesn't. Just as, for some reason, BS-ing is called "hallucinating" when AI does it, plagiarism is called "memorization" when AI does it. If the training corpus doesn't have enough diverse samples for the AI to riff on, AI will happily produce near-copies of the training material.

Source



Does anybody believe that is creative transformation and isn't plagiarism?

[johnegonpdx]

BH admins may want to organize a Red Team to offer OpenAI and other platform developers when it comes to personal finance / investing. Best way to improve how AI responds to queries / prompts on the subject.

[jebmke]

BH admins may want to organize a Red Team to offer OpenAI and other platform developers when it comes to personal finance / investing. Best way to improve how AI responds to queries / prompts on the subject.

Offer them what exactly?

[JayB]

Reportedly, web site owners are placing increasingly large amounts of content out of the reach of AI training by specifying that limitation in robots.txt files that normally sit in the root directory. I was unable to find a robots.txt file at all for bogleheads.org, but that would seem like a good idea.

[slippinsurlies]

Imo, a lot of threads could be autofilled by ai. For example, every college thread results in the same basic points made, easy enough to fill in 100 or so posts with ai personas, ditto for how much gold to buy, ditto for tips funds vs individual tips, ditto for spias, etc. We could even have a special ai persona which encourages away anyone who posts extraordinary credentials (hedgedundie, swedoe, etc) as this too, has unfortunately become somewhat predictable.

I suppose a portfolio review ai would be genuinely useful.

Best,

Agreed. Something that helps guide people to correctly formatting and calculating their portfolio, which makes a few basic observations, and then signals to the rest of the forum that “this person is ready to talk, now get in here and help!”

Take this a step further and you have a robo-advisor.

[jebmke]

Reportedly, web site owners are placing increasingly large amounts of content out of the reach of AI training by specifying that limitation in robots.txt files that normally sit in the root directory. I was unable to find a robots.txt file at all for bogleheads.org, but that would seem like a good idea.

Because the AI companies are basically thieves

[CyberBob]

Reportedly, web site owners are placing increasingly large amounts of content out of the reach of AI training by specifying that limitation in robots.txt files that normally sit in the root directory. I was unable to find a robots.txt file at all for bogleheads.org, but that would seem like a good idea.

This.
I use a robots.txt file on all of my web servers. It works much better than you would ever think it would as far as others respecting it.

[Fallible]

I'm going to ask a related but more specific followup question: Is it possible to use a generative AI tool (ChatGPT, Copilot, etc) to summarize a long discussion thread, and to be able to ask specific questions against that thread?

Here's my example: I'm interested in the discussion about the Retiree Portfolio Model. This particular discussion has almost 1900 posts spanning over 38 pages. I have to be honest, I am not going to read or review all 1900 posts before asking a question. Sure, I could do a search, but that still would require me to come up with the right keywords and synonyms, and would require me to wade through potentially many results.

It would be really useful if I could somehow point ChatGPT or Copilot at this thread, and then ask it a specific question.

Is this possible? (Desirable?) Anyone know how to easily do it?

Why not just ask a bot to summarize a long thread and see what it says? But if it does produce a summary, how will you know it's an accurate one?

[techbud]

I'm going to ask a related but more specific followup question: Is it possible to use a generative AI tool (ChatGPT, Copilot, etc) to summarize a long discussion thread, and to be able to ask specific questions against that thread?

Here's my example: I'm interested in the discussion about the Retiree Portfolio Model. This particular discussion has almost 1900 posts spanning over 38 pages. I have to be honest, I am not going to read or review all 1900 posts before asking a question. Sure, I could do a search, but that still would require me to come up with the right keywords and synonyms, and would require me to wade through potentially many results.

It would be really useful if I could somehow point ChatGPT or Copilot at this thread, and then ask it a specific question.

Is this possible? (Desirable?) Anyone know how to easily do it?

Why not just ask a bot to summarize a long thread and see what it says?

That was kind of my point... I don't know if bots or gpts are able to follow (eg, correlate) the many pages of a long thread, but it would be very useful if they could. So my question to the community was does anyone know how to prompt one of the AI systems to do so.

But if it does produce a summary, how will you know it's an accurate one?

Summarization is one of the features that I have found GPTs to be very good at. And for the case I cited above (the 1900 posts about RPM), I would expect that if I could effectively prompt a GPT to ask (for example) "How does RPM deal with partial year portfolio updates?", it would be able to find that within the 1900 posts.

[jebmke]

That was kind of my point... I don't know if bots or gpts are able to follow (eg, correlate) the many pages of a long thread,

Some time in the coming weeks I am going to test something akin to this with Gemini. Earlier this summer I gave Gemini a test that we use to certify tax preparers on MD tax law. It failed miserably.

I am going to pass it a text file with all the training material and ask it to take the test again.

I did learn one thing in my "dialogue" with Gemini. It cannot read PDF files.

[Dottie57]

It is the human conversation I enjoy here. AI will not be what I want -ever.

[Syner01]

looks like copyright and trademarks are going to have some issues showing cause however libel may show some promise. this is for generative AI.
https://law-arizona.libguides.com/c.php ... &p=9997517

i am curious how closed systems will be affected. for instance i see some law groups stating they are using AI in closed fed systems to train and quickly highlight things regarding litigation on subjects. quickly go through thousands of medical records and highlight-classify artifacts for instance, generate questions from data in seconds to support your side of the claim, etc.

AI 1st gen i don't see replacing human intelligence. i see it as another toolset to help quickly parse lots of data and provide some calculation on it. similar to an assistant, copilot, etc. instead of sifting through volumes of data manually you will now need to focus on verification/validation. not sure how this plays out though especially in high risk areas like surgery for instance.