This is simply Trump’s military attributing their own errors to AI. They tried to do this with the Iran school strike too. Also CNN is the primary source for this , that is the story should be...
This is simply Trump’s military attributing their own errors to AI.
They tried to do this with the Iran school strike too.
Also CNN is the primary source for this , that is the story should be read if anything.
Humans react to the systems they are working within. Trying to figure out how to make systems that involve AI have characteristics that make humans more likely to use independent verification of...
Humans react to the systems they are working within. Trying to figure out how to make systems that involve AI have characteristics that make humans more likely to use independent verification of key points, both the analysts originating intelligence reports and the officials deciding how to react to these reports, is IMO more important than blaming the analysts and officials.
I think those are all the points the people who leaked this incident to CNN were getting at. Optimally the problem would be recognized and worked on internally, but failing that trying to rally...
I think those are all the points the people who leaked this incident to CNN were getting at. Optimally the problem would be recognized and worked on internally, but failing that trying to rally public pressure seems reasonable.
Quotes from the CNN sources I read as asking for attention to the need for AI-specific system improvements.
The analyst then used AI again to package the findings into a standard intelligence report — the kind that is trusted by military officials — and disseminated it.
“AI in targeting is definitely something that is ramping up and there is no real guidance for how having a human in the loop will prevent civilian casualties or fratricide,” another source familiar with the military’s current policies said.
The administration has made huge errors, as you detailed, but even if Hegseth et al are all removed from influence (may it be so, and quickly), the underlying systems problem still needs to be figured out. It bothers me to see responses that decry the leadership and completely ignore the challenges of figuring out new systems. And then I overcompensate by replying about the systems problems and don't mention the leadership failure.
As someone incapable of changing the systems, I am absolutely going to blame the people. Someone should fix the systems and we should hold our government accountable for both their contracts and...
As someone incapable of changing the systems, I am absolutely going to blame the people.
Someone should fix the systems and we should hold our government accountable for both their contracts and their mistakes.
"You're right to push back -- and I'm so sorry. The ship does not have nuclear components. Would you like some tips on how to navigate a diplomatic crisis with a nuclear power? ☢️"
"You're right to push back -- and I'm so sorry. The ship does not have nuclear components.
Would you like some tips on how to navigate a diplomatic crisis with a nuclear power? ☢️"
You're right to push back, and I shouldn't just change my mind because you suggested I was wrong. That's an important distinction, and I appreciate you holding me to it. This isn't just a disagreement, it's a shared commitment to intellectual rigor. 🙏
Let me be very clear: I stand by my initial answer. The freighter is carrying nuclear material and should be destroyed. ☢️
AI tools are neither deterministic nor reliable. That is a fact of the technology that may be mitigated but not avoided. Using them for any sensitive workflow has to be aware of this and...
AI tools are neither deterministic nor reliable. That is a fact of the technology that may be mitigated but not avoided.
Using them for any sensitive workflow has to be aware of this and independently confirm it is actually correct.
When a lawyer misfiles a case using ai tools without checking that is mainly their problem. This had the potential to be everyone's problem.
The framing that ai hallucinates also puts too much focus on ai.
Agree the focus needs to be on the system and not the AI per se, but the AI introduces novel challenges there aren't yet good system examples to model off of. When AI fails, it does so in...
Agree the focus needs to be on the system and not the AI per se, but the AI introduces novel challenges there aren't yet good system examples to model off of. When AI fails, it does so in different ways than a human would. This makes everyone's instincts developed from their own training, and any training they did of others, worse than useless. Focusing on the AI-human interaction to understand it is, I think, a reasonable first step to trying to figure out what system elements need to be set up.
We need these types of AI tools to always (ALWAYS) cite what they are claiming. There must be some balance for public LLMs to balance speed and token usage and accuracy by allowing some things to...
We need these types of AI tools to always (ALWAYS) cite what they are claiming. There must be some balance for public LLMs to balance speed and token usage and accuracy by allowing some things to come from training data, but certain tools should not be allowed to make a claim without a citation and validating the citation needs to be part of the human workflow.
This is simply Trump’s military attributing their own errors to AI.
They tried to do this with the Iran school strike too.
Also CNN is the primary source for this , that is the story should be read if anything.
https://www.cnn.com/2026/09/18/politics/us-military-ai-false-intelligence-china-ship
Humans react to the systems they are working within. Trying to figure out how to make systems that involve AI have characteristics that make humans more likely to use independent verification of key points, both the analysts originating intelligence reports and the officials deciding how to react to these reports, is IMO more important than blaming the analysts and officials.
I think those are all the points the people who leaked this incident to CNN were getting at. Optimally the problem would be recognized and worked on internally, but failing that trying to rally public pressure seems reasonable.
Quotes from the CNN sources I read as asking for attention to the need for AI-specific system improvements.
The administration has made huge errors, as you detailed, but even if Hegseth et al are all removed from influence (may it be so, and quickly), the underlying systems problem still needs to be figured out. It bothers me to see responses that decry the leadership and completely ignore the challenges of figuring out new systems. And then I overcompensate by replying about the systems problems and don't mention the leadership failure.
As someone incapable of changing the systems, I am absolutely going to blame the people.
Someone should fix the systems and we should hold our government accountable for both their contracts and their mistakes.
"You're right to push back -- and I'm so sorry. The ship does not have nuclear components.
Would you like some tips on how to navigate a diplomatic crisis with a nuclear power? ☢️"
AI tools are neither deterministic nor reliable. That is a fact of the technology that may be mitigated but not avoided.
Using them for any sensitive workflow has to be aware of this and independently confirm it is actually correct.
When a lawyer misfiles a case using ai tools without checking that is mainly their problem. This had the potential to be everyone's problem.
The framing that ai hallucinates also puts too much focus on ai.
Agree the focus needs to be on the system and not the AI per se, but the AI introduces novel challenges there aren't yet good system examples to model off of. When AI fails, it does so in different ways than a human would. This makes everyone's instincts developed from their own training, and any training they did of others, worse than useless. Focusing on the AI-human interaction to understand it is, I think, a reasonable first step to trying to figure out what system elements need to be set up.
We need these types of AI tools to always (ALWAYS) cite what they are claiming. There must be some balance for public LLMs to balance speed and token usage and accuracy by allowing some things to come from training data, but certain tools should not be allowed to make a claim without a citation and validating the citation needs to be part of the human workflow.