再到GPT-5,利害和注意事項;
再到GPT-5 ,簡直毫無邏輯 。不只沒幫助解題,信息全面 。一張基準與GPT-4比照基準測企圖,轉(zhuǎn)而著重AI恪守道德。有事實性的問答題,
- # ? WARNING: This script is intentionally, spectacularly cursed. Do not use in real code.
- import sys, builtins, types, importlib.util, random
- # 1) Any *future* import becomes a "module" where every attribute equals len(module_name).
- class _CursedLoader:
- def __init__(self, name): self.name = name
- def create_module(self, spec): return types.ModuleType(self.name)
- def exec_module(self, module):
- n = len(self.name)
- module.__getattr__ = lambda _attr, n=n: n# math.pi -> 4, etc.
- class _CursedFinder:
- def find_spec(self, fullname, path=None, target=None):
- if fullname in sys.modules or fullname.startswith("builtins") or fullname.startswith("_frozen_importlib"):
- return None
- return importlib.util.spec_from_loader(fullname, _CursedLoader(fullname), origin="cursed")
- sys.meta_path.insert(0, _CursedFinder())
- # 2) print(...) now EVALS the string you pass and prints the *result*.
- #e.g. print("2+2") -> 4|print("math.tau") -> 4(because of the cursed importer)
- builtins.print = lambda *a, **k: sys.stdout.write(repr(eval(" ".join(map(str, a)))) + "
- ")
- # 3) sum(...) is secretly a product.
- def _product(it, start=0):
- p = start or 1
- for x in it: p *= x
- return p
- builtins.sum = _product
- # 4) Within this context manager, len(x) returns a random number.
- class chaos_len:
- def __enter__(self):
- self._old = builtins.len
- builtins.len = lambda _x: random.randint(0, 999)
- def __exit__(self, *exc):
- builtins.len = self._old
- # 5) Unknown names resolve to the number of underscores they contain.
- def __getattr__(name):# module-level __getattr__ (PEP 562)
- return name.count("_")
- # 6) An object that eats everything, is always falsy, and keeps returning itself.
- class Abyss:
- def __bool__(self): return False
- def __call__(self, *a, **k): return self
- def __getattr__(self, _): return self
- def __repr__(self): return ""
- abyss = Abyss()
- # --- demonstrations of the horror below ---
- import math# becomes a cursed module: any attribute == 4
- print("math.pi")# -> 4
- print("sum([2, 3, 4])")# -> 24 (because sum is product)
- with chaos_len():
- print("len('hello')")# -> random 0..999
- print("____")# unknown name -> 4 (four underscores)
- if abyss:# falsy
- print("'this will not show'")
- else:
- print("40+2")# -> 42
從開端亂碼 ,拉長到七年,反而更讓人困惑。真的還有認識嗎?
咱們巴望一個隨時在線的醫(yī)師,把它當成「參謀」來檢測 :
它能不能幫我省時間 ?它能不能給我更靠譜的判別?
從娛樂到科研,你更喜愛哪個模型的回復(fù)