Nadia: It's not so much the origins, but the algorithims they use. For example, lobbyists would send our marketing BS to 50 media outlets. When asking a related question, 50 BS answers would flood the Internet. The primary AI's would treat those 50 as 50 different assessments. AlterAI treats them as one. BIG difference!
The distressing thing is Elon Musk pushes "truth" as a critical goal of AI but Grok doesn't appear to be trained for "truth" very well. I asked Grok about this and got the usual pro-Grok propaganda. But it does offer to explain it's behavior:
"If you share the exact questions and the specific outputs from Grok (or want me to address any of Droz's five questions directly with evidence-based reasoning), I can demonstrate Grok's approach in practice. Truth-seeking requires evaluating claims against data, not just noting which system agrees with a preferred conclusion."
Dave: Grok phrased its answer dishonestly. It has NOTHING to do with GROK not giving a "preferred conclusion" answer. Zero. NADA. Facts are facts. As I wrote above, I asked all of the AIs the exact same questions. Yes I'm sure that Grok says it has "evidence based reasoning" - which amounts to the consensus view.. I the examples I gave, the consensus view is WRONG.
This is an example of the opposite of critical thinking. You start with a predefined conclusion, and then you decide all the sources are wrong if they disagreed with you. That's CONFIRMATION BIAS, and it has no place in critical thinking.
instead of:
Gathering evidence,
Evaluating sources,
Reaching a conclusion,
You:
Start with a conclusion,
Evaluates evidence based on whether it supports that conclusion,
Rejects contrary evidence because it conflicts with the desired result.
Related terms include:
Confirmation Bias: selectively accepting supporting evidence and discounting contradictory evidence.
Motivated Reasoning: using reasoning to defend a preferred outcome rather than discover the truth.
Circular Reasoning: if disagreement itself is taken as proof that a source is wrong.
Cherry Picking: keeping favorable sources while dismissing unfavorable ones.
The problem arises when the criterion becomes:
"Any source that agrees with me is credible; any source that disagrees with me is not."
At that point the conclusion becomes effectively unfalsifiable, because no contrary evidence is allowed to count against it.
Wind energy can be useful for powering ships if you’re not in a hurry, but you still need backup power if you’re traveling through low wind areas. Also you can’t head directly into the wind
Ray: Sorry you are wrong on all points. I started off with no conclusion. I then did extensive research. I then published my research (e.g., see samples above) to see if anyone could expose an error (none). I THEN tested the main AI sources, and the results are as posted.
Do we know the origins of Alter AI that makes it so much more accurate than the rest?
Nadia: It's not so much the origins, but the algorithims they use. For example, lobbyists would send our marketing BS to 50 media outlets. When asking a related question, 50 BS answers would flood the Internet. The primary AI's would treat those 50 as 50 different assessments. AlterAI treats them as one. BIG difference!
Hugh Kendrick
Thanks, John, very interesting.
I have been very leery of AI right from the start, basically because I see great opportunities for mischief whether unintended or not.
Cheers, Hugh
Hugh: You're welcome. Continue to be very cautious dealing with AI.
The distressing thing is Elon Musk pushes "truth" as a critical goal of AI but Grok doesn't appear to be trained for "truth" very well. I asked Grok about this and got the usual pro-Grok propaganda. But it does offer to explain it's behavior:
"If you share the exact questions and the specific outputs from Grok (or want me to address any of Droz's five questions directly with evidence-based reasoning), I can demonstrate Grok's approach in practice. Truth-seeking requires evaluating claims against data, not just noting which system agrees with a preferred conclusion."
Dave: Grok phrased its answer dishonestly. It has NOTHING to do with GROK not giving a "preferred conclusion" answer. Zero. NADA. Facts are facts. As I wrote above, I asked all of the AIs the exact same questions. Yes I'm sure that Grok says it has "evidence based reasoning" - which amounts to the consensus view.. I the examples I gave, the consensus view is WRONG.
This is an example of the opposite of critical thinking. You start with a predefined conclusion, and then you decide all the sources are wrong if they disagreed with you. That's CONFIRMATION BIAS, and it has no place in critical thinking.
instead of:
Gathering evidence,
Evaluating sources,
Reaching a conclusion,
You:
Start with a conclusion,
Evaluates evidence based on whether it supports that conclusion,
Rejects contrary evidence because it conflicts with the desired result.
Related terms include:
Confirmation Bias: selectively accepting supporting evidence and discounting contradictory evidence.
Motivated Reasoning: using reasoning to defend a preferred outcome rather than discover the truth.
Circular Reasoning: if disagreement itself is taken as proof that a source is wrong.
Cherry Picking: keeping favorable sources while dismissing unfavorable ones.
The problem arises when the criterion becomes:
"Any source that agrees with me is credible; any source that disagrees with me is not."
At that point the conclusion becomes effectively unfalsifiable, because no contrary evidence is allowed to count against it.
Wind energy can be useful for powering ships if you’re not in a hurry, but you still need backup power if you’re traveling through low wind areas. Also you can’t head directly into the wind
Bill: Yes, to get any place efficiently, you need a fossil fuel motor. How many merchant ships are sailboats?
Ray: Sorry you are wrong on all points. I started off with no conclusion. I then did extensive research. I then published my research (e.g., see samples above) to see if anyone could expose an error (none). I THEN tested the main AI sources, and the results are as posted.