[arjvik]

As a quick hack, is this flaw fixed by using top-p or top-k or min-p or whatever the current hottest logit sampling algorithm is?

[FeepingCreature]

The "flaw" is also fixed by simply not recognizing the end-of-sampling token. Even the limited size of the context window can be worked around with tool calls.