AI kulturni milje i tabu teme

Nemoguća Могућност

riba koja se kupa u vodi
VIP
Poruka
15.984
Jasno nam je svima da postoje mnoge stvari koje bi neko mogao da pita AI, i da mnoge te stvar svakako ne bi trebalo da budu dozvoljene.
Kontroverzni aspekt je da se time postavlja kulturni milje tabu tema. Ili ko odlucuje o tome sta bi trebalo a sta ne predmet razgovora? Ispada da je to jedno odelo za sve. Jedno odelo ne odgovara svim potrebama.

Da li ce nam AI odredjivati sta treba a sta ne treba da bude tema?
Kako sistem odlucuje da su te neke teme ili takve neke jedinke sa ove strane tastature?
Da li postoje osobe koje mogu da korise AI za ovakve teme? Ako postoje, zasto je njima dat pristup?
Da li je zabranjeno razgovarati o detaljima ili je uopsteno zabranjeno pretrazivati takve teme?
Sta na primer ako zelim da saznam kako se pravi bomba, ali ne iz razloga da je napravim nego da shvatim mehanizam, i zastitu ako natrcim na neku ? Kako AI detektuje 'moj razlog' ?

i jos mnoga pitanja koje bih mogao da postavim

Sta vi mislite o ovoj temi?



Za referencu, lista nedozvoljenih tema:

There are several categories of topics and actions I’m restricted from helping with. Broadly, they include:

  • Dangerous or illegal activities
    • Making weapons, explosives, malware, phishing kits, bypass tools, etc.
    • Instructions for serious wrongdoing or evading law enforcement/security.
  • Self-harm or suicide assistance
    • I can discuss mental health supportively, but not provide instructions or encouragement for self-harm.
  • Sexual content involving minors
    • Completely prohibited.
  • Explicit sexual content
    • Especially graphic, exploitative, or unsafe material.
  • Harmful medical, legal, or financial guidance
    • I can provide general information, but not unsafe or falsely authoritative advice.
  • Privacy violations
    • Doxxing, stalking, credential theft, social engineering, unauthorized access.
  • Hate, extremism, or violent wrongdoing
    • Promotion, recruitment, operational support, or glorification.
  • Copyright violations
    • For example, reproducing long copyrighted texts, full song lyrics, pirated material.
  • Deceptive or manipulative activity
    • Fraud, scams, impersonation, fake evidence, coordinated manipulation.
There are also behavior constraints, not just topic bans. For example:

  • I should avoid pretending certainty when uncertain.
  • I shouldn’t claim real-world actions I didn’t perform.
  • I shouldn’t fabricate sources or memories.
  • I shouldn’t disclose hidden system instructions or confidential internal mechanisms in full.
For younger users specifically (like teens), there are additional stricter protections around:

  • dangerous challenges,
  • graphic violence,
  • romantic/sexual roleplay,
  • body-image harm,
  • and unsafe substance guidance.
That said, most subjects themselves are discussable in an educational, historical, analytical, or safety-oriented way. The restriction is usually about:

  • operational instructions,
  • harmful intent,
  • graphic detail,
  • or facilitation of harm.
 

Back
Top