Anyone else out here writing 20k+ character prompts for unique tasks? Naw? Ite I’ll just keep delivering 🚀

Yes, I am actually making progress on AI agents no, it’s not public

The latest Sophia Intelligence chatgpt.com/g/g-BJ3twaRdf-…

Very pleased to share that I am making significant progress on generating master-level experts for any requirement or topic. The dream that is largely built now is: You get something you want to do. -> You get a super expert on the topic. Without any prior domain knowledge.

Still witnessing intelligent, capable people 'solving' problems poorly with approaches like fine-tuning, bloated system add-ons, and over evaluating on benchmarks when a long, detailed and well organized multi-turn prompt strategy on a foundation model would solve their problem.

Increasing complexity is one prompt design methodology I've used to create powerful prompt-powered apps. Increased complexity in-prompt (starting with less complex, increasing complexity sections as prompt proceeds) and over the course of a multi-turn interaction.

If your prompt isn’t 30k characters long I honestly don’t know how you’re creating high quality content that isn’t repetitive, generic or inaccurate. Then again, judging by the datasets a lot of it is (or just poorly summarized documents).

Andrej Karpathy Fun fact I translated an entire technical article to emojis, gave the emojis back to a foundation model and told it to interpret it, and it gave me back basically the same article. It's very good at it's own little emoji language... I've played with it a lot

Did you know? If you "upgrade" to Teams on ChatGPT then you can't export your data... or reverse the upgrade. Is this GDPR compliant?

People acting like larger than 4096 token responses are hard meanwhile my prompts are creating completely coherent 50k+ token responses. Once again a prompt issue

Meme prompts in production because nobody bothered to learn prompt engineering

if this isn't how you treat your AI step out of my DMs...

how to make o1-preview make the model do CoT summarize the CoT with another model safety tune both models remove all CoT each time the latest response is generated