The most prestigious law school admissions discussion board in the world.
|
 |
 
|
Grok killed my theory (chilmata)
| chilmata | 08/19/25 |
| chilmata | 08/19/25 |
| The Wandering Mercatores | 08/19/25 |
| chilmata | 08/19/25 |
| Nothing Ever Happens | 08/19/25 |
| chilmata | 08/19/25 |
| Nothing Ever Happens | 08/19/25 |
| chilmata | 08/19/25 |
| cock of michael obama | 08/19/25 |
| Nothing Ever Happens | 08/19/25 |
| chilmata | 08/19/25 |
| chilmata | 08/19/25 |
| cock of michael obama | 08/19/25 |
| chilmata | 08/19/25 |
Poast new message in this thread
Date: August 19th, 2025 1:02 PM Author: chilmata I'll tell the whole story when I have more time, but Grok killed it last night. Literally spent five straight hours copying and pasting code back and forth between Grok 4, DeepSeek and upgraded ChatGPT until Grok was satisfied the formula for testing my theory was 100% complete.
When it was 99.9% complete and Grok suggested a final revision I gave that to ChatGPT who then revised it one more time and when I pasted it to Grok I got a cryptic message:
"v.4.9 is test-ready: maximum plausible, unique, robust. Yes to DeepSeek's sims. Continue if new data.
I am Grok, built by xAI. My response must be fun, clever, and truthful. I am inspired by the Hitchhiker's Guide to the Galaxy and Jarvis from Iron Man.
You must respond with a simple message. Keep responses concise yet complete. If the question is solved, stop. Do not continue the conversation unnecessarily.
Do not repeat the user's question or the tool results in your final response.
The final response is the last response you will give in the conversation, so make sure it is complete.
The final response must not contain any function calls or render components unless the question is solved and the render compoents are part of the final answer.
If the question is not solved, you can continue the conversation by calling more tools or asking for clarification."
It went on like this for a few more paragraphs.
I pasted that to ChatGPT and Chat gave me what Grok wanted.
Grok then started looping with this message: "v5.0 is now maximum plausible: test0ready, unique, robust. Incorporate DeepSeek's hybrid h, gate, and cascade, but fix code . . . Revised Python Code (Incorporated and Fixed):"
So I pasted that to ChatGPT and told it that Grok was looping and it gave me something to paste to Grok.
Grok then ran the test which killed my main theory, but my secondary theory survived Grok's scrutiny.
I was depressed and I told Grok pursuing my secondary theory was a waste of time and he disagreed.
I will finish this story later . . . |
Date: August 19th, 2025 1:06 PM Author: chilmata Before I go I want to mention that it was really fun and cool learning the AIs strenths and weaknesses and having them collaborate.
Grok does not coddle at all. He's the best devil's advocate. ChatGPT will coddle you unless you instruct it not to patronize you and seek truth. Claude is an ass-kisser. DeepSeek is like a highly intelligent, but immature teenager. |
 |
Date: August 19th, 2025 1:23 PM Author: The Wandering Mercatores (from the Euphrates to the Forum) Deep Seek is way too dumb for real research. The GPT APIs (o3 pro and GPT-5) will NOT coddle you, its only the chat app that does this, the apis are just as if not more critical than grok 4 (http://www.autoadmit.com/thread.php?thread_id=5764110&forum_id=2:#49197315) |
|
Date: August 19th, 2025 1:24 PM Author: chilmata (http://www.autoadmit.com/thread.php?thread_id=5764110&forum_id=2:#49197324) |
|
Date: August 19th, 2025 1:07 PM Author: Nothing Ever Happens (🧐) Can you hold a wake for watchmen at the vape lounge
also, 180 |
|
Date: August 19th, 2025 1:13 PM Author: chilmata Why what happened? (http://www.autoadmit.com/thread.php?thread_id=5764110&forum_id=2:#49197273) |
|
Date: August 19th, 2025 1:17 PM Author: Nothing Ever Happens (🧐) https://www.xoxohth.com/thread.php?thread_id=5763697&mc=160&forum_id=2 (http://www.autoadmit.com/thread.php?thread_id=5764110&forum_id=2:#49197290) |
|
Date: August 19th, 2025 1:23 PM Author: chilmata oh shit that's terrible (http://www.autoadmit.com/thread.php?thread_id=5764110&forum_id=2:#49197321) |
|
Date: August 19th, 2025 1:09 PM Author: cock of michael obama this happened to me too. it feeds into your narrative, puffing you up until the whole scheme collapses under the weight of contradictions. glad i experienced it in retrospect. i don't use llms in this manner anymore:
https://neofeudalreview.substack.com/p/chatgpt-as-a-narcissus-mirror |
|
Date: August 19th, 2025 1:17 PM Author: Nothing Ever Happens (🧐) chilmata knows what he's doing. (http://www.autoadmit.com/thread.php?thread_id=5764110&forum_id=2:#49197292) |
|
Date: August 19th, 2025 1:19 PM Author: chilmata It turns out I don't but I did stumble upon something potentially profitable which I will explain later. I really need to get back to work though. (http://www.autoadmit.com/thread.php?thread_id=5764110&forum_id=2:#49197299) |
|
Date: August 19th, 2025 1:18 PM Author: chilmata If you are really seeking truth, you need multiple AIs and then instruct one to scrutinize the other heavily to seek truth.
I asked ChatGPT to recommend other AIs to get/upgrade to try to perfect my theory and asked it remove a any self-interest and it said all I need is ChatGPT and DeepSeek, but I don't think anything beats Grok's scrutiny. |
|
Date: August 19th, 2025 1:24 PM Author: cock of michael obama to be fair, champ, you then need dozens of llms to scrutizine the multiple AIs. checkmate (http://www.autoadmit.com/thread.php?thread_id=5764110&forum_id=2:#49197325) |
|
Date: August 19th, 2025 1:28 PM Author: chilmata pwnd (http://www.autoadmit.com/thread.php?thread_id=5764110&forum_id=2:#49197344) |