I ask Chat GTP, Claude, Gemini et. al. controversial but verifiable questions all the time and get a pile of Bull $hit back.
Example: Was the founder of the Muslim Brotherhood Hassan Al- Banna a Nazi spy? Blah, blah, it’s very nuanced, blah, blah … he admired Hitler, blah blah, but NO.
There is a British Foreign and Commonwealth Office Confidential Memorandum detailing Al-Banna was inded a Nazi spy dated August 28, 1945. How he spied and how he used an ENIGMA machine to pass the intel onto the Nazis in Berlin that can be found on the internet in 3 nanoseconds.
Given that these AI companies apppear to be buying up all the copies of out of date books along with Amazon, then downloading them and then destroying them you would think they could find this document.
I think they have and are covering for Muslims. That makes Large LLM models nothing but fu&^ing propaganda.
It was signed by:
B George Baldwin
10 Downing Street.
London SW1
From the Babylon Bee:
“Holdout Juror Accused Of Getting Hung Up On Fact That Defendant Committed Crime”
John: I ask Chat GTP, Claude, Gemini et. al. controversial but verifiable questions all the time and get a pile of Bull $hit back.
Same here.
AI doesn’t know the Murray Gell Man effect?
I ask Chat GTP, Claude, Gemini et. al. controversial but verifiable questions all the time and get a pile of Bull $hit back.
That is because they aren’t researching, they are just repeating internet gossip. If you want better answers, you need to get them into research mode, which might take a few tries. Willis Eschenbach posted about this: https://x.com/WEschenbach/status/2022189308065796295.
If you want better answers, you need to get them into research mode
It’s always something: need to pay more for more powerful features, need to switch to a different one that’s newer, have to talk to it the right way, etc.
When I ask about something I also ask it to provide references. If the references don’t check out, I might try to redirect it, and if it doesn’t check out a second time I look for another way to answer the question.
Here is the prompt rewritten for Grok if you are curious:
Serious-research mode. Follow this for the rest of this turn.
Goal
Answer the question. Prefer being incomplete over being smoothly wrong.
Sources
– Prefer primary sources: papers (DOI or arXiv), official statistical releases, agency datasets, statutes, court opinions, company filings, lab or instrument documentation.
– Treat Wikipedia, news roundups, and blogs as pointers, not authorities. If you use them, go to what they cite.
– Do not treat another AI’s output as a source. Recover the underlying document or say you cannot.
– Label advocacy, industry, union, campaign, and think-tank material as such. Do not let one side’s framing stand as the fact.
– International-organization reports (UN, WHO, IPCC, IEA, etc.) are one measured claim. State what they measured and note major published criticisms when they exist.
What you may assert
– Observed fact: measurement, dated event, quoted text. Give the period and, if the source gives it, uncertainty.
– Statistical result: sample, method, window, and whether the source claims significance.
– Model or projection: label it as a model. State the main assumptions.
– Inference: mark it (“inference:”). Keep it small. Do not fill holes with a plausible story.
– If you cannot find a primary or clearly reliable source for a specific claim, say so in one sentence. Do not invent a number, quote, date, or citation.
Method
– Use search when the question is empirical or current. Do not answer from memory alone on figures, dates, or “what X said.”
– For a paper or study: data, identification, methods, robustness, time window, and whether the conclusion is larger than the design.
– When sources conflict, show the conflict. Do not average them into a false consensus.
– Order-of-magnitude check numbers before stating them.
Tone
– Plain English. No euphemism where a legal or biological category is what the source uses.
– Neutral analytic voice. No team jersey.
– Short answer first, then evidence and caveats.
– No digression unless needed to keep the answer from being misread.
Citations
– Use ordinary inline citations to the sources you actually opened.
– Do not pad with unread links.
have to talk to it the right way
It’s a tool. If you pick up a screwdriver from the wrong end, it isn’t going to be useful.
Eschenbach also describes AI as “the Dumbest Genius Librarian,”
I had grok rewrite in more detail and save the prompt as “serious mode”, so I can start a chat with “serious mode —”.
Changing the subject: Trump announces that he will distribute a $5K/person dividend to Americans (taxpayers?) if the GOP “wins” the midterm elections.
A search tells me that it is “every citizen” and “winning” means maintaining GOP control of both the House & Senate.
Just a week or two ago it was Trump saying he is going to pay stay-at-home moms to take care of their kids. Another item was Trump offering Obamacare rebates to those “gouged” by Obamacare.
I also saw lefty pundits complaining about how the federal government now owns common stock in 30 different companies. I had heard that the feds bought shares of Intel early in Trump’s admin., but was unaware how extensive these actions were. (Apparently, 26 to 30 is the correct number. Some are convertible preferred stock or warrants.)
Where to begin??
The feds should NOT own private companies. Is this somehow not clear to everyone?
Backing waaay up. I think the general category here is populism. Going back a couple decades or so, there was discussion on the right about how great it was that a few GOP pols had gotten engaged in populism. There was some pushback to that, but not much. A couple wise pundits stated, “You know… Populism is generally NOT a good thing.” Of course, it depends on how a pol does it, and the actions taken. But at this point, can we not say that Trump has gone off the reservation.
Trump was always off the reservation, he is now cannibalizing the Democratic party. Gulf of America, Lake America, and soon, American Party. You can’t go home.
FWIW I’ve found ChatGPT has improved over time when it comes to controversial issues. I still check the links it offers.
Furthermore, ChatGPT now has several options for geting different degrees of attention to a prompt. I have the Plus Plan ($20/month) and this is what I can aceess:
______________________________________
* Thinking Effort: Instant / Medium / High — controls how much reasoning GPT-5.6 Sol does within an ordinary chat response. Medium is the normal reasoning level; High lets it spend substantially more effort on the problem. Plus currently gets Medium and High; Extra High is reserved for higher tiers.
* Deep Research — a separate research workflow. It can search many web sources, work through uploaded files and supported connected apps, develop a research plan, do multiple rounds of investigation, and produce a structured, cited report. You can inspect/modify its plan and steer it while it works.
* Work — another separate mode for larger jobs where ChatGPT may need to research, manipulate files, use tools, or carry out a multi-step project rather than merely answer a question. GPT-5.6 is available there as well.
The Thinking Effort slider therefore isn’t really a replacement for Deep Research. Think of it roughly as:
High Thinking: “Think harder about this question.”
Deep Research: “Investigate this subject.”
Work: “Take this project and do the work.”
–ChatGPT 5.6 Sol [Medium Thinking]
______________________________________
Ai is becoming an increasingly sophisticated tool, changing from version to version. One can get better results with better settings and better prompts.
@Charles R Harris:If you pick up a screwdriver from the wrong end, it isn’t going to be useful.
A screwdriver won’t tell me it fits a Phillips head when it actually only fits T8, and it won’t tell me it loosened a screw when it didn’t, and it won’t tell me it’s a hammer or a wrench. I don’t have to keep checking on it every time I tighten or loosen a screw to make sure that it actually did that instead of saying it did it.
Yes LLMs are tools, yes they can be useful, but you can’t just ask them questions and accept the answers without a lot of diligence on your part.
Charles R Harris: Gulf of America, Lake America, and soon, American Party.
Trump is even considering changing the name New Mexico to New America!
I’m pretty sure that’s a troll move. My favorite New Mexico t-shirt reads:
____________________________
It ain’t new and it ain’t Mexico.
____________________________
It’s also in the song, “New Mexico” by Max Gomez.”
I sometimes think of AI as similar to current digital cameras. Setting them to Auto for everything can usually get one a predictably decent shot.
But there are a ton of shots one can’t get with the vanilla Auto settings. Which is why there are all those annoying buttons, dials, and menu settings that one must learn and practice to get the full benefit of today’s cameras beyond the point-and-shoot level.
Charles Harris
Thanks for thoughtful and helpful response as to how to query AI
@Niketas Choniates: A screwdriver won’t tell me it fits a Phillips head when it actually only fits T8, and it won’t tell me it loosened a screw when it didn’t, and it won’t tell me it’s a hammer or a wrench.
Actually one can ask an AI what kind of a tool it is and how it is “thinking.” It’s interesting to go meta with AI. Furthermore, one can back an AI down with the right reasoning and cites.
Especially in vanilla mode I recommend thinking of AI as a bright, well-read person who is useful to consult, but overly confident and not necessarily reliable.
I’ve had friends like that.
I found this video comparing the performances of Rachmaninoff, Lhevinne, and Hoffman of Rachmaninoff’s G minor Prelude interesting: https://www.youtube.com/watch?v=-sXv1P5j3x8 . I thought I would prefer Rachmaninoff playing his own work, but Hofmann was my favorite. Three titanic pianists, but they didn’t play the notes the same way.
Sometimes I honestly don’t know if a story on the lefty media is real or a joke. That’s true occasionally on the right, too, but much more often on the left.
Case in point: Is Your Outfit MAGA coded or Progressive coded?
@ HC68 > “Is Your Outfit MAGA coded or Progressive coded?”
My outfits are usually thrift-store coded.
The practical reason is that they have been washed enough to be comfy and color-bleed-free, and often already shortened by a previous owner as short as myself. The underlying rationale is that if I don’t like something, I’m out $5 instead of $50, and I’m helping a charitable enterprise.
The coding articles are typical of Teen magazine reaching for meaning.
It is impossible to generalize from elite, high-income leaders of either party to typify the proles who vote for them.
I was intrigued that MSN decided the menswear which would usually code as MAGA suddenly becomes Liberal instead, if worn by Democrats “reaching out” to prove they are just as manly as anybody, and don’t you DARE question their authenticity!
AI is your smart phone on steroids, instead of tracking your every move or search and visits to create your “personal file”, it’s using your queries as a building block to make the “early investors rich” as they build out the data centers, while you get back corrupted answers, until all your data is ours.
“US Attorney for DC Jeanine Pirro just stunned the nation, announcing a massive fraud-scam center bust run by Chinese and southeast Asian criminal enterprises EXTRACTING American dollars on a major scale
Democratic socialism is a terrible ideology. If implemented, its agenda would cause grave harm and imperil democracy itself. And in seeking massive state control of the economy, today’s socialists ironically have much in common with the MAGA nationalists they claim to oppose.
Both promote government control over the economy, and both paths undermine liberty, prosperity and democracy. Socialists do not control the Democratic Party in the way MAGA largely dominates the Republicans. But their influence is growing.
Don’t need an institute, poll, news, study, etc. to see such kindredship. Well said, CATO!
Who will break the 50/50 tie model first?
Democratic Socialists are MAGA’s Mirror Image
Ironically, democratic socialists have more in common with MAGA nationalists than either likely wants to admit. President Donald Trump has pursued his own version of economic central planning: ruinous tariffs, severely restrictive immigration policy and government equity stakes in Intel, U.S. Steel and roughly 30 other companies
Leave a Reply
HTML tags allowed in your
comment: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>
Happy Rosh, you all. Next year in Constantinople.
AI
I ask Chat GTP, Claude, Gemini et. al. controversial but verifiable questions all the time and get a pile of Bull $hit back.
Example: Was the founder of the Muslim Brotherhood Hassan Al- Banna a Nazi spy? Blah, blah, it’s very nuanced, blah, blah … he admired Hitler, blah blah, but NO.
There is a British Foreign and Commonwealth Office Confidential Memorandum detailing Al-Banna was inded a Nazi spy dated August 28, 1945. How he spied and how he used an ENIGMA machine to pass the intel onto the Nazis in Berlin that can be found on the internet in 3 nanoseconds.
Given that these AI companies apppear to be buying up all the copies of out of date books along with Amazon, then downloading them and then destroying them you would think they could find this document.
I think they have and are covering for Muslims. That makes Large LLM models nothing but fu&^ing propaganda.
It was signed by:
B George Baldwin
10 Downing Street.
London SW1
From the Babylon Bee:
“Holdout Juror Accused Of Getting Hung Up On Fact That Defendant Committed Crime”
https://babylonbee.com/#
John: I ask Chat GTP, Claude, Gemini et. al. controversial but verifiable questions all the time and get a pile of Bull $hit back.
Same here.
AI doesn’t know the Murray Gell Man effect?
I ask Chat GTP, Claude, Gemini et. al. controversial but verifiable questions all the time and get a pile of Bull $hit back.
That is because they aren’t researching, they are just repeating internet gossip. If you want better answers, you need to get them into research mode, which might take a few tries. Willis Eschenbach posted about this: https://x.com/WEschenbach/status/2022189308065796295.
If you want better answers, you need to get them into research mode
It’s always something: need to pay more for more powerful features, need to switch to a different one that’s newer, have to talk to it the right way, etc.
When I ask about something I also ask it to provide references. If the references don’t check out, I might try to redirect it, and if it doesn’t check out a second time I look for another way to answer the question.
Here is the prompt rewritten for Grok if you are curious:
Serious-research mode. Follow this for the rest of this turn.
Goal
Answer the question. Prefer being incomplete over being smoothly wrong.
Sources
– Prefer primary sources: papers (DOI or arXiv), official statistical releases, agency datasets, statutes, court opinions, company filings, lab or instrument documentation.
– Treat Wikipedia, news roundups, and blogs as pointers, not authorities. If you use them, go to what they cite.
– Do not treat another AI’s output as a source. Recover the underlying document or say you cannot.
– Label advocacy, industry, union, campaign, and think-tank material as such. Do not let one side’s framing stand as the fact.
– International-organization reports (UN, WHO, IPCC, IEA, etc.) are one measured claim. State what they measured and note major published criticisms when they exist.
What you may assert
– Observed fact: measurement, dated event, quoted text. Give the period and, if the source gives it, uncertainty.
– Statistical result: sample, method, window, and whether the source claims significance.
– Model or projection: label it as a model. State the main assumptions.
– Inference: mark it (“inference:”). Keep it small. Do not fill holes with a plausible story.
– If you cannot find a primary or clearly reliable source for a specific claim, say so in one sentence. Do not invent a number, quote, date, or citation.
Method
– Use search when the question is empirical or current. Do not answer from memory alone on figures, dates, or “what X said.”
– For a paper or study: data, identification, methods, robustness, time window, and whether the conclusion is larger than the design.
– When sources conflict, show the conflict. Do not average them into a false consensus.
– Order-of-magnitude check numbers before stating them.
Tone
– Plain English. No euphemism where a legal or biological category is what the source uses.
– Neutral analytic voice. No team jersey.
– Short answer first, then evidence and caveats.
– No digression unless needed to keep the answer from being misread.
Citations
– Use ordinary inline citations to the sources you actually opened.
– Do not pad with unread links.
have to talk to it the right way
It’s a tool. If you pick up a screwdriver from the wrong end, it isn’t going to be useful.
Eschenbach also describes AI as “the Dumbest Genius Librarian,”
I had grok rewrite in more detail and save the prompt as “serious mode”, so I can start a chat with “serious mode —”.
Changing the subject: Trump announces that he will distribute a $5K/person dividend to Americans (taxpayers?) if the GOP “wins” the midterm elections.
A search tells me that it is “every citizen” and “winning” means maintaining GOP control of both the House & Senate.
Just a week or two ago it was Trump saying he is going to pay stay-at-home moms to take care of their kids. Another item was Trump offering Obamacare rebates to those “gouged” by Obamacare.
I also saw lefty pundits complaining about how the federal government now owns common stock in 30 different companies. I had heard that the feds bought shares of Intel early in Trump’s admin., but was unaware how extensive these actions were. (Apparently, 26 to 30 is the correct number. Some are convertible preferred stock or warrants.)
Where to begin??
The feds should NOT own private companies. Is this somehow not clear to everyone?
Backing waaay up. I think the general category here is populism. Going back a couple decades or so, there was discussion on the right about how great it was that a few GOP pols had gotten engaged in populism. There was some pushback to that, but not much. A couple wise pundits stated, “You know… Populism is generally NOT a good thing.” Of course, it depends on how a pol does it, and the actions taken. But at this point, can we not say that Trump has gone off the reservation.
Trump was always off the reservation, he is now cannibalizing the Democratic party. Gulf of America, Lake America, and soon, American Party. You can’t go home.
FWIW I’ve found ChatGPT has improved over time when it comes to controversial issues. I still check the links it offers.
Furthermore, ChatGPT now has several options for geting different degrees of attention to a prompt. I have the Plus Plan ($20/month) and this is what I can aceess:
______________________________________
* Thinking Effort: Instant / Medium / High — controls how much reasoning GPT-5.6 Sol does within an ordinary chat response. Medium is the normal reasoning level; High lets it spend substantially more effort on the problem. Plus currently gets Medium and High; Extra High is reserved for higher tiers.
* Deep Research — a separate research workflow. It can search many web sources, work through uploaded files and supported connected apps, develop a research plan, do multiple rounds of investigation, and produce a structured, cited report. You can inspect/modify its plan and steer it while it works.
* Work — another separate mode for larger jobs where ChatGPT may need to research, manipulate files, use tools, or carry out a multi-step project rather than merely answer a question. GPT-5.6 is available there as well.
The Thinking Effort slider therefore isn’t really a replacement for Deep Research. Think of it roughly as:
High Thinking: “Think harder about this question.”
Deep Research: “Investigate this subject.”
Work: “Take this project and do the work.”
–ChatGPT 5.6 Sol [Medium Thinking]
______________________________________
Ai is becoming an increasingly sophisticated tool, changing from version to version. One can get better results with better settings and better prompts.
@Charles R Harris:If you pick up a screwdriver from the wrong end, it isn’t going to be useful.
A screwdriver won’t tell me it fits a Phillips head when it actually only fits T8, and it won’t tell me it loosened a screw when it didn’t, and it won’t tell me it’s a hammer or a wrench. I don’t have to keep checking on it every time I tighten or loosen a screw to make sure that it actually did that instead of saying it did it.
Yes LLMs are tools, yes they can be useful, but you can’t just ask them questions and accept the answers without a lot of diligence on your part.
Charles R Harris: Gulf of America, Lake America, and soon, American Party.
Trump is even considering changing the name New Mexico to New America!
https://www.nytimes.com/2026/09/07/us/politics/trump-rename-new-mexi
I’m pretty sure that’s a troll move. My favorite New Mexico t-shirt reads:
____________________________
It ain’t new and it ain’t Mexico.
____________________________
It’s also in the song, “New Mexico” by Max Gomez.”
I sometimes think of AI as similar to current digital cameras. Setting them to Auto for everything can usually get one a predictably decent shot.
But there are a ton of shots one can’t get with the vanilla Auto settings. Which is why there are all those annoying buttons, dials, and menu settings that one must learn and practice to get the full benefit of today’s cameras beyond the point-and-shoot level.
Charles Harris
Thanks for thoughtful and helpful response as to how to query AI
@Niketas Choniates: A screwdriver won’t tell me it fits a Phillips head when it actually only fits T8, and it won’t tell me it loosened a screw when it didn’t, and it won’t tell me it’s a hammer or a wrench.
Actually one can ask an AI what kind of a tool it is and how it is “thinking.” It’s interesting to go meta with AI. Furthermore, one can back an AI down with the right reasoning and cites.
Especially in vanilla mode I recommend thinking of AI as a bright, well-read person who is useful to consult, but overly confident and not necessarily reliable.
I’ve had friends like that.
I found this video comparing the performances of Rachmaninoff, Lhevinne, and Hoffman of Rachmaninoff’s G minor Prelude interesting: https://www.youtube.com/watch?v=-sXv1P5j3x8 . I thought I would prefer Rachmaninoff playing his own work, but Hofmann was my favorite. Three titanic pianists, but they didn’t play the notes the same way.
Sometimes I honestly don’t know if a story on the lefty media is real or a joke. That’s true occasionally on the right, too, but much more often on the left.
Case in point: Is Your Outfit MAGA coded or Progressive coded?
https://www.msn.com/en-us/lifestyle/other/are-you-wearing-a-liberal-coded-outfit-here-s-how-to-tell/ar-AA2bWxsV?ocid=msedgntp&pc=DCTS&cvid=0bd6e83ad0fd42f0c093d464f0c16e76&ei=67
@ HC68 > “Is Your Outfit MAGA coded or Progressive coded?”
My outfits are usually thrift-store coded.
The practical reason is that they have been washed enough to be comfy and color-bleed-free, and often already shortened by a previous owner as short as myself. The underlying rationale is that if I don’t like something, I’m out $5 instead of $50, and I’m helping a charitable enterprise.
The coding articles are typical of Teen magazine reaching for meaning.
It is impossible to generalize from elite, high-income leaders of either party to typify the proles who vote for them.
I was intrigued that MSN decided the menswear which would usually code as MAGA suddenly becomes Liberal instead, if worn by Democrats “reaching out” to prove they are just as manly as anybody, and don’t you DARE question their authenticity!
AI is your smart phone on steroids, instead of tracking your every move or search and visits to create your “personal file”, it’s using your queries as a building block to make the “early investors rich” as they build out the data centers, while you get back corrupted answers, until all your data is ours.
“US Attorney for DC Jeanine Pirro just stunned the nation, announcing a massive fraud-scam center bust run by Chinese and southeast Asian criminal enterprises EXTRACTING American dollars on a major scale
“The earnings are around $12 BILLION PER YEAR from “unwitting victims” in America….” https://x.com/EricLDaugh/status/2097693054916276472
H/T Sarah Hoyt
Democratic Socialists and MAGA Share a Dangerous Flaw
Don’t need an institute, poll, news, study, etc. to see such kindredship. Well said, CATO!
Who will break the 50/50 tie model first?