Don’t use OpenAI’s outdated tools. Also, don’t rely on prompt engineering to force the output to conform. Instead, use a local LLM and something like jsonformer or parserllm which can provably output well-formed/parseable text.
I’ll be informal to boost your intuition. You know how a parser can reject invalid inputs? Parsers can be generated from grammars, so we can think of the grammars themselves as rejecting invalid inputs too. When we use a grammar for generation, every generated output will be a valid input when parsed, because the grammar can’t build any invalid sentences (by definition!)
For example, suppose we want to generate a JSON object. The grammar for JSON objects starts with an opening curly brace “{”. This means that every parser which accepts JSON objects (and rejects everything else) must start by accepting “{”. So, our generator must start by emitting a “{” as well. Since our language-modeling generators work over probability distributions, this can be accomplished by setting the probability of every token which doesn’t start with “{” to zero.
That reminds me of an old paper about how to create a compilable C program out of old game ROMs. Decompile to assembly. Implement a bunch of #define statements that implement all the ASM statements. Now compile it to a native binary on whatever platform.
Won’t likely be faster or more accurate than regular emulation methods, but it’s a neat idea considering that the source code on all this stuff was lost a long time ago.
It very well might be a real exploit. Lemmy was briefly taken down by an XS attack using the emoji library… so who knows, maybe a 3000% smiley face is all that is needed
I think there a lot of phone scammers that use font size to hide all the shit they’re doing. Like they make shit so small so that the old people can’t see anything
Yeah, I’m chalking that up to Python’s untypedness. I was going to write “integers”, but technically that function takes a “num”, whatever that is.
For all we know, it could be a string, asking ChatGPT to hack the government. Is that even? Probably no. Or None. Or T-Rex. Without reading the entire function, we don’t know that it’s not returning T-Rex.
Thankfully, it doesn’t matter. Just stick the result into an if-else, then False and None will land you in the else-branch. And both True and our Truthiness-Rex will land you in the if-branch. Just as Guido intended.
Well, in the sake of pointing things out, GPT-4 can actually correctly answer the prompt, because it arrives at it in the opposite direction. It can tell the integer is even or odd and knows that even or odd integers in binary end in 0 or 1 respectively.
programming.dev
Active