How to build and use an AI (4 Viewers)

[SamAndreas]

In creating this thread I intend to show all of you what I've discovered so far about using AI. The intention is not that I'm the expert sayings how it is.

This thread is a discussion thread open to all to add their opinions and examples of about how to best use this new tool.

I will explain what I'm doing to make best use of it. In this post, and the following post, I will show you how I ask the questions.

The point being the "garbage in garbage out" old principle is even more applicable when applied to questions, than it is statements, since it doubles when the "garbage out" part it added to the garbage output from the original question.

I think about it before starting to type my question. The question needs to be very specific insofar as what my goal is.

AI's have a bad tendency to go off into the weeds if this point isn't considered carefully. In the example I'm about to show you I specify Who, What, When, Where as elements I seek answers to.



I saw a tweet at Twitter and the AI I'm asking a question to Grok version 3. This is the tweet I saw.

To keep this opening post short since it is the opening post, and will be seen at the top of every page of this thread, I'm going to post it now and do the example it applies to in a following post. Please keep in mind this tweet is not the subject of this thread, it's an example used to illustrate this first example about asking questions to an AI.

 

[SamAndreas]

The follow up post to the opening post. The following is a cut and pasted from the dialog box at the Twitter site where Grok resides. My question is in bold with increased font size at the top:

Would you please look at John Hudson's media activity and see if there is more there which would fill in the who, what, when, where aspects of this statement by Hudson?


To provide a comprehensive analysis of John Hudson's media activity related to the specific post and fill in the "who, what, when, where" aspects, I’ve reviewed his recent activity on X, cross-referenced with available web data, and considered the context of the thread you provided. Since my knowledge is continuously updated and I can access real-time information up to 11:35 AM PDT on Sunday, June 29, 2025, I’ll base this on his posts and related sources available as of that time. Here’s the breakdown:
Original Post Context

• Post ID: 1939370205898997810
• Metadata: John Hudson
  @John_Hudson
  , posted at 17:07 UTC (10:07 AM PDT) on June 29, 2025.
• Text: "SCOOP: The United States obtained intercepted communication between senior Iranian officials discussing this month’s military strikes on Iran’s nuclear program and remarking that the attack was less devastating than they had expected, per four sources."

Filling in the "Who, What, When, Where" Aspects
Who

• John Hudson: The author is a seasoned journalist with The Washington Post, specializing in national security and State Department coverage. His bio (from washingtonpost.com, updated as of June 2025) notes his Pulitzer Prize finalist status for coverage of Jamal Khashoggi’s murder and extensive reporting from conflict zones like Ukraine and Afghanistan. This establishes his credibility in handling sensitive intelligence leaks.
• Sources: The post cites "four sources," unnamed but implied to be within U.S. intelligence or diplomatic circles, given Hudson’s access to such information. The follow-up post (1939370207081893980) suggests these intercepts involve "senior Iranian officials," likely from Iran’s military or nuclear oversight bodies (e.g., IRGC or Atomic Energy Organization of Iran), though specific names are not disclosed.
• Collaborator: The thread concludes with a link to a Washington Post article co-authored with Warren Strobel (
  @wstrobel
  ), indicating a team effort to verify and expand on the scoop.

What

• Event: The scoop details intercepted communications revealing Iranian officials’ assessment of recent U.S. military strikes on Iran’s nuclear program. The officials noted the strikes were "less devastating than expected," suggesting a discrepancy between U.S. claims and Iranian perceptions of damage.
• Context from Thread: Subsequent posts (e.g., 1939370517536072081, 1939373647581364620) elaborate that the Trump administration targeted nuclear facilities at Fordow, Natanz, and Isfahan. CIA Director John Ratcliffe’s briefing to Congress claimed destruction of key sites like Fordow’s metal conversion facility, critical for bomb core production, but Iranian officials downplayed this, hinting at possible pre-strike preparations (e.g., relocating enriched uranium).
• Dispute: The Trump administration counters that the Iranians misjudged the damage, asserting precision strikes achieved their intended effect, while analysts debate the extent of destruction due to Iran’s potential mitigation (per 1939373644779598071).

When

• Post Time: The X post was made at 17:07 UTC (10:07 AM PDT) on June 29, 2025.
• Event Timing: The strikes occurred "this month," i.e., June 2025, with the intercepts likely obtained shortly after. The thread references classified congressional briefings "last week" (around June 22-26, 2025), suggesting the intelligence was fresh and analyzed rapidly.
• Related Activity: Hudson’s activity on X shows a concentrated thread starting at 17:07 UTC, with follow-ups through 17:33 UTC, indicating real-time reporting as the story broke. The linked Washington Post article (per 1939375712789491799) likely provides a timestamped publication, though X data alone doesn’t specify the exact article release time.

Where

• Strike Locations: The targets are identified as Fordow, Natanz, and Isfahan, all key Iranian nuclear sites. Fordow is a deeply buried facility (80m underground, per Wikipedia’s June 25, 2025, update), Natanz is a major enrichment site, and Isfahan hosts uranium conversion and storage. The thread (1939373647581364620) notes Ratcliffe’s assessment that much of Iran’s enriched uranium is "likely buried at Isfahan and Fordow," reinforcing these as focal points.
• Interception Source: The communications were intercepted by the U.S., likely via NSA signals intelligence operations, though the exact method or location of interception isn’t specified. Given the context, this could involve satellite or cyber surveillance over Iranian territory.
• Reporting Origin: Hudson’s post originates from his X account, tied to his Washington Post role, suggesting the story was developed from Washington, D.C., with input from U.S. officials briefed there.

Additional Media Activity by John Hudson

• Thread Expansion: Hudson’s subsequent posts (e.g., 1939370207081893980 to 1939375440776634705) provide a detailed narrative, including White House responses (via press secretary Karoline Leavitt), CIA briefings, and congressional reactions (e.g., Senators Chris Murphy and Lindsey Graham). This shows active follow-up within 20 minutes of the initial scoop.
• Collaborative Reporting: The mention of
  @wstrobel
  in post 1939375712789491799 links to a Washington Post article, which likely expands on the scoop with additional sourcing or analysis. As of 11:35 AM PDT, the article isn’t fully detailed on X, but Hudson’s history suggests it includes interviews with officials and possibly declassified data.
• No Contradictory Posts: A quick scan of Hudson’s X timeline (up to 11:35 AM PDT) shows no retraction or conflicting reports, indicating the story holds as his primary focus today. Earlier posts (e.g., from June 28) focused on State Department briefings, showing a shift to this breaking news.

Gaps and Inferences

• Unnamed Sources: The "four sources" remain anonymous, a common practice in national security reporting. Hudson’s track record suggests these could be intelligence analysts or officials with clearance, possibly from the DIA or CIA, given Ratcliffe’s involvement.
• Strike Details: The thread lacks specifics on the strike date (only "this month") or weapon types beyond the B-2 bombers and GBU-57A/B bombs mentioned in related BBC reporting (June 25, 2025). Hudson might elaborate in the Washington Post piece.
• Iranian Perspective: The intercepted communications are one-sided; no direct Iranian response is cited, though their downplaying aligns with state media narratives (e.g., Press TV’s June 26, 2025, claim of minimal damage).

Conclusion
John Hudson’s media activity on X, centered around the 17:07 UTC post and its thread, effectively outlines the "who" (himself, U.S. sources, Iranian officials), "what" (intercepted assessments of U.S. strikes), "when" (June 2025 strikes, reported June 29), and "where" (Fordow, Natanz, Isfahan). His rapid follow-ups and collaboration with Strobel suggest a coordinated effort to flesh out the story, likely detailed further in the linked Washington Post article. For the full picture, I recommend checking the article (https://t.co/UtpJe1XhpF) and monitoring Hudson’s X for updates past 11:35 AM PDT today, as the story may evolve with official reactions.

 

[LA 2 L.A.]

AI getting somethings right does not mean it's okay to rely on it. I see you're trying to shift the garbage in garbage out problem to the user's prompts. That's factually innaccrate. I will repost what I posted in the other thread.

You are factually wrong about who and what publicly accessible AI's are learning from and what they are learning. It's very much Garbage In-Garbage Out (GIGO), but don't take my word for it. Listen to what the folks at MIT have to say about it:

AI models work by training on huge swaths of data from the internet. But as AI is increasingly being used to pump out web pages filled with junk content, that process is in danger of being undermined.

New research published in Nature shows that the quality of the model’s output gradually degrades when AI trains on AI-generated data. As subsequent models produce output that is then used as training data for future models, the effect gets worse.

AI trained on AI garbage spits out AI garbage​

As junk web pages written by AI proliferate, the models that rely on that data will suffer.
www.technologyreview.com


And this Forbes article points out that even restricted access AI has a garbage in garbage out issues:

What’s going on?

Industry observers agree that data poses a problematic issue for AI initiatives of all kinds. Executives understand “how subpar their data collection, cleansing, and curation process that is fed into AI to create a decision engine," said industry analyst Andy Thurai. "They know it is garbage in, garbage out. So, knowing they are feeding garbage in, would you feel comfortable using the decision?"

Complicating the situation is the fact that many enterprises now use synthetic data to train their AI models when not enough data is available, or to maintain security. The problem is that "executives’ confidence that the models are trained on real-world data is not there,” Thurai added.

Garbage In, Garbage Out? Trust In The Data Behind AI Is Vanishing​

Trust in the data needed for data-driven decisions is falling — precipitously.

www.forbes.com

 

[LA 2 L.A.]

The follow up post to the opening post. The following is a cut and pasted from the dialog box at the Twitter site where Grok resides. My question is in bold with increased font size at the top:

Would you please look at John Hudson's media activity and see if there is more there which would fill in the who, what, when, where aspects of this statement by Hudson?


To provide a comprehensive analysis of John Hudson's media activity related to the specific post and fill in the "who, what, when, where" aspects, I’ve reviewed his recent activity on X, cross-referenced with available web data, and considered the context of the thread you provided. Since my knowledge is continuously updated and I can access real-time information up to 11:35 AM PDT on Sunday, June 29, 2025, I’ll base this on his posts and related sources available as of that time. Here’s the breakdown:
Original Post Context

• Post ID: 1939370205898997810
• Metadata: John Hudson
  @John_Hudson
  , posted at 17:07 UTC (10:07 AM PDT) on June 29, 2025.
• Text: "SCOOP: The United States obtained intercepted communication between senior Iranian officials discussing this month’s military strikes on Iran’s nuclear program and remarking that the attack was less devastating than they had expected, per four sources."

Filling in the "Who, What, When, Where" Aspects
Who

• John Hudson: The author is a seasoned journalist with The Washington Post, specializing in national security and State Department coverage. His bio (from washingtonpost.com, updated as of June 2025) notes his Pulitzer Prize finalist status for coverage of Jamal Khashoggi’s murder and extensive reporting from conflict zones like Ukraine and Afghanistan. This establishes his credibility in handling sensitive intelligence leaks.
• Sources: The post cites "four sources," unnamed but implied to be within U.S. intelligence or diplomatic circles, given Hudson’s access to such information. The follow-up post (1939370207081893980) suggests these intercepts involve "senior Iranian officials," likely from Iran’s military or nuclear oversight bodies (e.g., IRGC or Atomic Energy Organization of Iran), though specific names are not disclosed.
• Collaborator: The thread concludes with a link to a Washington Post article co-authored with Warren Strobel (
  @wstrobel
  ), indicating a team effort to verify and expand on the scoop.

What

• Event: The scoop details intercepted communications revealing Iranian officials’ assessment of recent U.S. military strikes on Iran’s nuclear program. The officials noted the strikes were "less devastating than expected," suggesting a discrepancy between U.S. claims and Iranian perceptions of damage.
• Context from Thread: Subsequent posts (e.g., 1939370517536072081, 1939373647581364620) elaborate that the Trump administration targeted nuclear facilities at Fordow, Natanz, and Isfahan. CIA Director John Ratcliffe’s briefing to Congress claimed destruction of key sites like Fordow’s metal conversion facility, critical for bomb core production, but Iranian officials downplayed this, hinting at possible pre-strike preparations (e.g., relocating enriched uranium).
• Dispute: The Trump administration counters that the Iranians misjudged the damage, asserting precision strikes achieved their intended effect, while analysts debate the extent of destruction due to Iran’s potential mitigation (per 1939373644779598071).

When

• Post Time: The X post was made at 17:07 UTC (10:07 AM PDT) on June 29, 2025.
• Event Timing: The strikes occurred "this month," i.e., June 2025, with the intercepts likely obtained shortly after. The thread references classified congressional briefings "last week" (around June 22-26, 2025), suggesting the intelligence was fresh and analyzed rapidly.
• Related Activity: Hudson’s activity on X shows a concentrated thread starting at 17:07 UTC, with follow-ups through 17:33 UTC, indicating real-time reporting as the story broke. The linked Washington Post article (per 1939375712789491799) likely provides a timestamped publication, though X data alone doesn’t specify the exact article release time.

Where

• Strike Locations: The targets are identified as Fordow, Natanz, and Isfahan, all key Iranian nuclear sites. Fordow is a deeply buried facility (80m underground, per Wikipedia’s June 25, 2025, update), Natanz is a major enrichment site, and Isfahan hosts uranium conversion and storage. The thread (1939373647581364620) notes Ratcliffe’s assessment that much of Iran’s enriched uranium is "likely buried at Isfahan and Fordow," reinforcing these as focal points.
• Interception Source: The communications were intercepted by the U.S., likely via NSA signals intelligence operations, though the exact method or location of interception isn’t specified. Given the context, this could involve satellite or cyber surveillance over Iranian territory.
• Reporting Origin: Hudson’s post originates from his X account, tied to his Washington Post role, suggesting the story was developed from Washington, D.C., with input from U.S. officials briefed there.

Additional Media Activity by John Hudson

• Thread Expansion: Hudson’s subsequent posts (e.g., 1939370207081893980 to 1939375440776634705) provide a detailed narrative, including White House responses (via press secretary Karoline Leavitt), CIA briefings, and congressional reactions (e.g., Senators Chris Murphy and Lindsey Graham). This shows active follow-up within 20 minutes of the initial scoop.
• Collaborative Reporting: The mention of
  @wstrobel
  in post 1939375712789491799 links to a Washington Post article, which likely expands on the scoop with additional sourcing or analysis. As of 11:35 AM PDT, the article isn’t fully detailed on X, but Hudson’s history suggests it includes interviews with officials and possibly declassified data.
• No Contradictory Posts: A quick scan of Hudson’s X timeline (up to 11:35 AM PDT) shows no retraction or conflicting reports, indicating the story holds as his primary focus today. Earlier posts (e.g., from June 28) focused on State Department briefings, showing a shift to this breaking news.

Gaps and Inferences

• Unnamed Sources: The "four sources" remain anonymous, a common practice in national security reporting. Hudson’s track record suggests these could be intelligence analysts or officials with clearance, possibly from the DIA or CIA, given Ratcliffe’s involvement.
• Strike Details: The thread lacks specifics on the strike date (only "this month") or weapon types beyond the B-2 bombers and GBU-57A/B bombs mentioned in related BBC reporting (June 25, 2025). Hudson might elaborate in the Washington Post piece.
• Iranian Perspective: The intercepted communications are one-sided; no direct Iranian response is cited, though their downplaying aligns with state media narratives (e.g., Press TV’s June 26, 2025, claim of minimal damage).

Conclusion
John Hudson’s media activity on X, centered around the 17:07 UTC post and its thread, effectively outlines the "who" (himself, U.S. sources, Iranian officials), "what" (intercepted assessments of U.S. strikes), "when" (June 2025 strikes, reported June 29), and "where" (Fordow, Natanz, Isfahan). His rapid follow-ups and collaboration with Strobel suggest a coordinated effort to flesh out the story, likely detailed further in the linked Washington Post article. For the full picture, I recommend checking the article (https://t.co/UtpJe1XhpF) and monitoring Hudson’s X for updates past 11:35 AM PDT today, as the story may evolve with official reactions.

Your prompts/questions to the AI do not guarantee that it did not provide false information. If you took the time to verify every bit of information, why not just link to those same sources, so that everyone would be able to read the full text of what they source said?

In my opinion it's both irresponsible and lazy to quote a response from AI. It's okay as starting point to find citations for information, but the original sources are what should be quoted. When's the last time anyone every quoted their Google search results page as a source citation?

 

[MT15]

The AI response included links to the source material, LA. I don’t have a problem with quoting AI, it’s not irresponsible as long as you disclose that it is an AI response.

IMO it’s not ethical to quote an AI source without disclosing that fact, though.

 

[LA 2 L.A.]

The AI response included links to the source material, LA. I don’t have a problem with quoting AI, it’s not irresponsible as long as you disclose that it is an AI response.

IMO it’s not ethical to quote an AI source without disclosing that fact, though.

I agree with that. I would add that a person should also quote the AI used (the integrity varies widely) and the prompts/questions that they gave it.

 

[SamAndreas]

Your prompts/questions to the AI do not guarantee that it did not provide false information. If you took the time to verify every bit of information, why not just link to those same sources, so that everyone would be able to read the full text of what they source said?

In my opinion it's both irresponsible and lazy to quote a response from AI. It's okay as starting point to find citations for information, but the original sources are what should be quoted. When's the last time anyone every quoted their Google search results page as a source citation?

I think the AI tool if used with care can assist one in the search and evaluation of the subject under discussion.

It's tool and like all tools it can be misused. But that possibility of misuse doesn't mean we throw away all tools.

This new tool is not going away LA, you'll get used to it being in existence over time.

If I use it with care it will serve me well.

Imagine how folks who thought "if god wanted people to fly he would have given them wings" felt after the Wright Brothers showed the world that people can fly. I'm sure they eventually got used to other people flying. Some of them I imagine changed their minds, and they even flew in an airplane for themselves.

Flying is fun, once it got started there was no stopping it. Not even if God had failed to give them wings.

 

[MT15]

One of the best uses I have found is to conduct huge searches quickly. We are going through all of our older, unused possessions that have built up over the past decade as our parents have passed away and my husband’s aunt who had no children. We ended up with a fair amount of really old things and just don’t have room for everything.

ChatGPT is good for loading a picture of an item and getting back an idea of how old it is, what it was used for and how much these items are fetching in 3 different ways of selling them.

I feel like it gives me valuable info in a fraction of the time it would take me to find it on my own.

 

[SamAndreas]

AI getting somethings right does not mean it's okay to rely on it. I see you're trying to shift the garbage in garbage out problem to the user's prompts. That's factually innaccrate. I will repost what I posted in the other thread.

You are factually wrong about who and what publicly accessible AI's are learning from and what they are learning. It's very much Garbage In-Garbage Out (GIGO), but don't take my word for it. Listen to what the folks at MIT have to say about it:

AI trained on AI garbage spits out AI garbage​

As junk web pages written by AI proliferate, the models that rely on that data will suffer.
www.technologyreview.com


And this Forbes article points out that even restricted access AI has a garbage in garbage out issues:

Garbage In, Garbage Out? Trust In The Data Behind AI Is Vanishing​

Trust in the data needed for data-driven decisions is falling — precipitously.

www.forbes.com

If I give to the AI what is in effect garbage because of careless in the way I ask the question, the AI will spit that Garbage I gave it back at me, and then add a boat load of garbage to my original garbage and that will make a big mess.

The mess it makes of it will have been caused by my carelessness. It will be my fault.

For an issue like this magazine articles are not that helpful because they seem to be mostly made of speculation based upon fear the author may feel. Fear of the unknown.

 

[LA 2 L.A.]

I think the AI tool if used with care can assist one in the search and evaluation of the subject under discussion.

I agree and never said otherwise.

It's tool and like all tools it can be misused. But that possibility of misuse doesn't mean we throw away all tools.

I agree and never said we should throw it away or not use it. Don't rely does not equal don't use.

This new tool is not going away LA, you'll get used to it being in existence over time.

I know it's not going away. Your ignorant arrogance is really showing. I've been regularly using several different AI apps for several different types of tasks since they became available. I use them to speed up processes I already know how to do, but I never rely on them as the final step for anything.

If I use it with care it will serve me well.

Using AI apps well can serve us well. Relying on them for factual accuracy is not using them well. Quoting their results without citing the specific app used and the prompts/questions given to them is not using them well.

Imagine how folks who thought "if god wanted people to fly he would have given them wings" felt after the Wright Brothers showed the world that people can fly.

You're straying off the core issues, again. I don't see anyone in this conversation that actually thinks like that, so it's an irreverent therefore meaningless point for this discussion. Seems like you just making stuff up to be condescending and patronizing.

I know I preach, but I also listen as much as I preach. Maybe give an equal effort to listening as you do to preaching.

 

[SamAndreas]

I agree with that. I would add that a person should also quote the AI used (the integrity varies widely) and the prompts/questions that they gave it.

I stated all that yesterday. I think the important things to do is to say it is from an AI, and give the reader the information needed so that a reader could go to that same AI. Above in my second post I said it was from the dialog box of Twitters AI, Grok 3.

I almost typed Grog, and this isn't the first time I have almost typed Grog instead of Grok. I know why, it's cognitive association thing:

 

[Dragon]

I typically run every source— through at least one or two bias and fact-checking tools. If the source comes from an official organization, I also look into who’s behind it, as that often reveals a lot about its credibility and potential bias. On top of that, I always try to find independent verification from other, unrelated sources before considering the information reliable.

To be honest, these days I mostly use AI when there’s a word that keeps slipping my mind. I know exactly what I want to say, but I just can’t find the precise word—probably due to my illness—and it’s getting really frustrating. But AI is surprisingly good at figuring out which word I mean when I try to describe it. Honestly, it's better at that than Google or regular synonym searches.

 

[LA 2 L.A.]

One of the best uses I have found is to conduct huge searches quickly. We are going through all of our older, unused possessions that have built up over the past decade as our parents have passed away and my husband’s aunt who had no children. We ended up with a fair amount of really old things and just don’t have room for everything.

ChatGPT is good for loading a picture of an item and getting back an idea of how old it is, what it was used for and how much these items are fetching in 3 different ways of selling them.

I feel like it gives me valuable info in a fraction of the time it would take me to find it on my own.

I didn't know that people were using ChatGPT to do photo searches. I haven't had a need for that, but now I know I can use it that way if I need. Thanks. I don't use ChatGPT for written content. I use other AI apps that I find to be much better. I'm about to start working with one that supposedly can learn my creative writing style and voice. I'm hoping it can help me speed up the story breaking and first draft process. I'm much faster and better at polish rewrites than cranking out the outline and first draft.

 

[LA 2 L.A.]

If I give to the AI what is in effect garbage because of careless in the way I ask the question, the AI will spit that Garbage I gave it back at me, and then add a boat load of garbage to my original garbage and that will make a big mess.

It's also highly likely to spit out garbage content even if your prompt/question is good. Good prompts/questions are not a safeguard from garbage content from an AI.

For an issue like this magazine articles are not that helpful because they seem to be mostly made of speculation based upon fear the author may feel. Fear of the unknown.

Is it your claim that the authors for MIT's technology review are just speculating based on their fear? If you are, then you've really gone off deep end or are just really hellbent on being "right." Possibly both.

 

[Dragon]

You basically have to treat AI as you would any other source - Evaluate, Verify and Analyze, On top of that many search engines available today is already backed by AI - even though not all are publicly advertizing it, so doing regular searches are no guarantie of better results than using ChatGPT or other open AI search engines.

 

[LA 2 L.A.]

I stated all that yesterday. I think the important things to do is to say it is from an AI, and give the reader the information needed so that a reader could go to that same AI. Above in my second post I said it was from the dialog box of Twitters AI, Grok 3.

And you also included your prompts/questions. I think that is the responsible way to go.

 

[SamAndreas]

One of the best uses I have found is to conduct huge searches quickly. We are going through all of our older, unused possessions that have built up over the past decade as our parents have passed away and my husband’s aunt who had no children. We ended up with a fair amount of really old things and just don’t have room for everything.

ChatGPT is good for loading a picture of an item and getting back an idea of how old it is, what it was used for and how much these items are fetching in 3 different ways of selling them.

I feel like it gives me valuable info in a fraction of the time it would take me to find it on my own.

I have also found that if I pose my question carefully I can get the AI to give me a shortened pass link to sources which will get me in to see the article which If I tried to use a normal link to that same story, I would be shut out due to a fire wall because I don't have a subscription to that source. That's a big plus.

What I do is ask for a shorted links in its report, asking it place them in line in the report to sources it used to answer my question.

With Grok I have written questions using multiple sentences and run on clauses separated with commas. I recall a question which grew to half a page before I got all the details Grok will need to fully understand my intentions for asking that question.

What Grok can do that humans can't do very well is deal with is layers upon layers of comma separated clauses in a runon sentence from hell. And if one run on sentence is not enough, use two run on sentences, or three.

 

[SamAndreas]

I agree and never said otherwise.

And I agree and never said otherwise.

Your first post in this tread strayed off of this core issue, you said:

Your prompts/questions to the AI do not guarantee that it did not provide false information. If you took the time to verify every bit of information, why not just link to those same sources, so that everyone would be able to read the full text of what they source said?

In my opinion it's both irresponsible and lazy to quote a response from AI. It's okay as starting point to find citations for information, but the original sources are what should be quoted. When's the last time anyone every quoted their Google search results page as a source citation?