AD FREE - INSANE! China Attacked Their Own People with Drones - Episode #283
Added 2025-10-06 15:27:55 +0000 UTCINSANE! China Attacked Their Own People with Drones Episode #283
Comments
You are correct. However, ChatGPT has admitted it's wrong and corrected itself. Of course, this had to do with particular mathematical measurements and formulas that I was inquiring about, nothing "historical". To be particular, ChatGPT 5 uses CoT (Chain of Thought). Here's just a smidgen taken from the net: Chain-of-Thought (CoT): This method prompts the LLM to show its reasoning step-by-step. The model can then be prompted to review and correct its own reasoning. External feedback loop: The LLM's output is sent to a separate tool, such as a search engine or code interpreter, to verify the result. The verification output is then fed back to the LLM for correction. For example, a "wait" command can trigger a verification step that significantly improves results.
DNA Patriot
2025-10-23 03:32:45 +0000 UTCHello, will you be posting episodes 284 & 285? Hopefully this is just a small glitch in an automated process or something.
Justin
2025-10-23 02:30:05 +0000 UTCugh, Patreon is effing around, hopefully they have not been hijacked by CCP interests ... π€¨ An answer Winston, to your argument with grok: I'm a computer veteran of 5 decades, I know a bit about how these LLMs work. They are trained on a _snapshot_ of data, and then they are released to the public. After this * they cannot learn anything more! * Your arguments, while valid, are completely ignored! Your arguments are not integrated into the LLM. Who ever chose the training set caused the current model to validate CCP propaganda. (That is x.ai's choice, so we have to fight with x.ai and maybe Elon, not with the LLM -that is futile.) All the LLMs are currently vulnerable to the humans that choose the training set - it's garbage in = garbage out, but on a massive scale. Still, great expose! Hopefully enough viewers will rant about x.ai's pro CCP stance and get them to make better choices for grok 5.
dv00
2025-10-09 08:41:45 +0000 UTC
