# @voooooogel — 2025-12-03

♥307 ↻30 · https://x.com/voooooogel/status/1996161200187732396

82k likes, and only two quote tweets and two replies noticed this was written by ai (it was gpt-5.x-thinking)

pretty soon, even those people are going to stop being able to tell. yet everyone is walking around with this cached belief that "ai slop" is and always will be obvious.

i wish base models had become more popular for many reasons, but one would've been to get people used to the reality of this much earlier. because openai sucked at post-training writing for ages, everyone got this idea in their heads  that ai writing is necessarily easy to recognize as such for model capabilities reasons. but in reality, base model output selected to sound human has been nearly indistinguishable from human writing for a long time! and detectors like Pangram (which is the best one available by far, but it's not magic) can't detect it either. the labs just weren't able to / didn't care to preserve that capability in their chat assistants until recently.

this is quickly reverting to not being true, but now instead of this realization (models can write indistinguishably from a human) hitting back when the models were otherwise weak, it's now going to hit concurrently with everything else that's happening...

i was at a talk recently where someone from {frontier_lab} said their (personal) opinion was that labs should "mete out" model capabilities over time to avoid sudden economic disruption. i think op is just one example of why that's a very bad strategy - there's too many false stories going around about models, so people rely on firsthand experience to understand what they're actually capable of. when you hold those capabilities back, gatekeep access to ground truth about the frontier, people start telling themselves stories about fundamental limitations, until - oops! artificial overhang, and we wasted two years of time that could've been used to adapt to this pretending that the capability didn't exist.

openai of course didn't deliberately make chatgpt-3.5 bad at writing like a human for the sake of holding back that capability, it was an accidental result of their other priorities. but the inadvertent masking of it from the general public did create a natural experiment of how public beliefs about models develop in the absence of hands-on experience of the frontier - and the result was not great. people are just now starting to realize what's been true since 2020-2023.

tags: author:voooooogel, kind:tweet, model:gpt-3-5, on:gpt-3-5, year:2025
cited on: _dossiers/gpt-3-5.md, gpt-3-5