Yeah, I think server overload. I get something every few hours, problems loading chat history, sometimes can’t reach. Probably 3 different times today. Seems to come back

Indeed

I think they are using a meta prompt on top of the original model. I don’t think they are using new data. This is from gpt 4o. Looks similar. What does this mean? With a similar meta prompt non OpenAI models should perform similarly. I would reflect on that. Have it write meta prompts for other models to get similar empowerment, ask questions about your own thoughts to a local model using those prompts if you want to protect your thoughts and data From gpt-4o after a meta prompt, none of which says use chain of thought. OpenAI prompt from their post: oyfjdnisdr rtqwainr acxz mynzbhhx -> Think step by step Use the example above to decode: oyekaijzdf aaptcg suaokybhai ouow aqht mynznvaatzacdfoulxxz . Let’s break down the problem step by step by analyzing the pattern in the initial example "oyfjdnisdr rtqwainr acxz mynzbhhx." Step-by-step analysis: 1. "oyfjdnisdr" → "think" The word "think" has 5 letters, and "oyfjdnisdr" has 10 letters. It seems like there are two characters for each letter of the decoded word. Each pair of letters corresponds to one letter in the decoded word. 2. Let’s check the first letter: "oy" becomes "t": "o" is the 15th letter in the alphabet, and "y" is the 25th. "t" is the 20th letter in the alphabet, which is halfway between 15 and 25. So, it seems the average of the positions of "o" and "y" gives us the position of "t." 3. Using this method, we will apply it to the rest of the word pairs: "f" (6) and "j" (10) → average = 8 → "h" "d" (4) and "n" (14) → average = 9 → "i" "i" (9) and "s" (19) → average = 14 → "n" "d" (4) and "r" (18) → average = 11 → "k" 4. Thus, "oyfjdnisdr" becomes "think." 5. "rtqwainr" → "step" We will apply the same method for this word. "r" (18) and "t" (20) → average = 19 → "s" "q" (17) and "w" (23) → average = 20 → "t" "a" (1) and "i" (9) → average = 5 → "e" "n" (14) and "r" (18) → average = 16 → "p" 6. So, "rtqwainr" becomes "step." 7. "acxz" → "by" "a" (1) and "c" (3) → average = 2 → "b" "x" (24) and "z" (26) → average = 25 → "y" 8. "acxz" becomes "by." 9. "mynzbhhx" → "step" "m" (13) and "y" (25) → average = 19 → "s" "n" (14) and "z" (26) → average = 20 → "t" "b" (2) and "h" (8) → average = 5 → "e" "h" (8) and "x" (24) → average = 16 → "p" 10. "mynzbhhx" becomes "step."

Chatgpt with the upgrades on 8/6 should be able to do it. Use the memory feature, then have it summarize itself so you can deploy to other models, then use the Json function calling = strict w 4o mini. Don't think about being fast. Be correct, go slow. Fast happens as a side effect. Literally ask it to produce a 1500 character summary of what it learned, drop that into a system prompt

@Ha My Tran my expertise is finance, if people want to see how you can integrate chatgpt with both basic and advanced topics check out https://www.skool.com/marketmatrix-ai I’m trying to educate people on topics that tend to be walled off, I believe in sharing knowledge. It’s pretty powerful once you can have the llm have domain knowledge. Still very early