I don't think AI is bringing anything that is not on the table in terms of the basic structure of propaganda. Seeing 10% of the specific propaganda you see on the internet in numbers of 10s of news, and seeing it on 100s but still at 10% doesn't really change something. The means of production is evolving fast. I would agree to that. But there is always one secret sauce that we usually don't produce under competition that gives propaganda an edge. Think about the feeling of sincerity between traditional and social media. Nobody was sure 360p videos, 140 character tweets, etc. would replace the traditional media. People found sincere feelings combined with quick dopamine in it. Same thing here, it's not the means to an end that brings the real change. Although it is catalyzing a substantial change for sure.
IMHO, this is less surprising discovery about him considering this probing is around 300 years ahead of his time, and Da Vinci already has proof for concepts 500 ahead of his time.
Very cool. I've been doing some "prompt design" myself, how did you land on the temperature 0.001?
The point is to use the least generative methods to parse the HTML.
It's kind of clunky to iterate on a prompt. If you change one thing about the prompt to get a better result for one aspect of the desired output, it can negatively impact the quality of another aspect.
There is an implementation for testing prompts internally when you set it up on local. This way you can get an idea about which parts a prompt and restrictive hyperparameters are failing you.
With that being said, you can also design a unique custom parser by changing the prompt for different subsets of data.
If you wanted to make the ratios more accurate. You might add something like "Make sure the ratios of ingredients are correct.
Which is better, but many times something else gets worse (e.g. the list of intructions to actually cook the thing). Curious what process you followed when designing the prompts.
This is really interesting. I would love to hear more about it if you would like to open an issue under the repository with something like: "Using additional instructions to improve parsing of preset parsers", I may add a preset parser on that issue to showcase.
I think someone could build a business around "prompt design". Small tweaks make such a huge difference that a tool that helped you improve prompts would be really valuable.