GPT-2: Too Dangerous To Release (2019)
naokishibuya.github.io
GPT-2: Too Dangerous To Release (2019)
1–10 of 139 posts
Re: GPT-2: Too Dangerous To Release (2019)
#2Feels like a hundred years ago.
Re: GPT-2: Too Dangerous To Release (2019)
#3Fable/Mythos: Hello World!
Re: GPT-2: Too Dangerous To Release (2019)
#4Seven years of this insufferable brand of "Oh it's so dangerous, I sure hope no one gives us a ton of money and takes us seriously" marketing and people are still falling for it at scale.
Re: GPT-2: Too Dangerous To Release (2019)
#5> Due to our concerns about malicious applications of the technology, we are not releasing the trained model.
They were not wrong, indeed whole industries are running on this technology maliciously now, because of which RAM, disk prices increased a lot.
- RAM, GPU, Disk prices are up
- Slop became the norm
- people are writing documents with AI, reading with AI, responding with AI
- students are doing homeworks with AI
- interviewees are using AI to cheat
- people are mass emailing with AI
- tiktok, instagram, youtube got even more non-sense videos
- and many more...Re: GPT-2: Too Dangerous To Release (2019)
#6In hindsight, they were entirely correct.
The social damage caused by low cost content generation that’s hard to distinguish from human authorship is astronomical. You don’t need to entertain the more ridiculous doomsday scenarios to wish that this technology had never been created.
Re: GPT-2: Too Dangerous To Release (2019)
#7The concern I heard was that releasing it would start an arms race for AGI, which I think it clearly did
Re: GPT-2: Too Dangerous To Release (2019)
#8Back in 2019, it was more fair to have caution around the larger GPT-2 models since robust text generation (by 2019 standards) was a complete unknown. For something like Mythos in 2026, where now the social implications of better LLMs are more understood, it's more fair to call it (EDIT: specifically, the declaration of its danger) a marketing gimmick.
Re: GPT-2: Too Dangerous To Release (2019)
#9Fable/Mythos: Hello World!
Different company. And doing security reviews with Fable is pretty annoying, it loves to downgrade to Opus
GPT-5.5 seems more dangerous in those regards
Re: GPT-2: Too Dangerous To Release (2019)
#10Say hypothetically that they were concerned that GPT models would see widespread abuse, for example by students cheating on homework assignments, in a way that could cause likely-irreversible societal changes some of which are harmful. Can we confidently say they were wrong?