AI guidelines

[PrimeNumber7]

Suddenly he is a know it all about bitcoin. merits will flow.

It is not the case, as long as AIs cannot count properly. If you ask it about the details, for example to compute N rounds of some hash function, then it will quickly fail.

Some example:

print first five rounds of SHA-1 of the empty message

GPT-4o mini

...

Below is the calculation of the first five rounds formatted descriptively:
Round   h0   h1   h2   h3   h4
0   0x67452301   0xEFCDAB89   0x98BADCFE   0x10325476   0xC3D2E1F0
1   Iterate...   Iterate...   Iterate...   Iterate...   Iterate...
2   Iterate...   Iterate...   Iterate...   Iterate...   Iterate...
3   Iterate...   Iterate...   Iterate...   Iterate...   Iterate...
4   Iterate...   Iterate...   Iterate...   Iterate...   Iterate...

...

See? You won't get anything beyond "Iterate...", or some hallucinated results, which would be totally incorrect. And the same is true for elliptic curves, even if they are quite simple, and if all private and public keys have numbers below 100. AI is still not there, and it won't get there, at least for some time. Which is also why it is much easier to spot AIs in technical board, because they are simply worse at math.

Challenge for AI enthusiasts: create a prompt, which will do that correctly.

Some example of AI failing at math in technical board: https://bitcointalk.org/index.php?topic=5574325.0

Most web interfaces of LLMs made by the hyperscalers contain some kind of agent workflow that will allow it to generate code in helping it provide a response. So it won't be able to recite that kind of answer, but it can generate code to provide an answer, and the better agent workflows will even check the code to ensure it alligns with the prompt before providing an answer.

Here is what I got when I asked gemini a similar question - https://gemini.google.com/share/9a81a9658277

[mrust_mobile]

https://old.reddit.com/r/LocalLLaMA/comments/1r8ectu/i_plugged_a_30_radio_into_my_mac_mini_and_told_my/

Obviously it won't be enough to finish an OS level app but it will be enough to create more basic apps which you would spend maybe 2-3 hours using an AI agent, openclaw will do it much quicker...

To me using AI in designing websites or even app(I don't think it can for now) is very wrong cause those websites lacks proper management, since you didn't write those codes by yourself, debugging, upgrading, and proper maintenance will be very hectic, you won't be as fluid as you should while designing the app, I know you can prompt it to add whatsoever features and style you want, but most times it won't even give you exactly what you want

This is the angle of the illiterate people which don't have a place in this new AI era.

Let me paint you a picture of what you can do with a LLM (preferably local) + openclaw combination.

You can create a basic instruction and tell openclaw:

1- scan the web for the price of this product, when it is lower than this price, send me a telegram message.
2- if the price is considerably lower than the 50 day average, create an account on the website using this email, put the item in the shopping cart, inform me immediately. if i don't respond, call me using whatsapp.
3- figure it out. create a full stack app or do it yourself i don't care.

Then you go out and roll on the grass.   That's just one example that came to my mind.