Ah, it’s ok to steal IP to build models, but not to use those models to steal IP. Got it.
What is permissible for Jupiter is not permissible for a cow.
Quod licet Jovi, non licet bovi.
Neither training models nor reverse engineering are necessarily IP theft. https://www.eff.org/issues/coders/reverse-engineering-faq
Going by the article you linked, game cloning runs into problems, including the EULA and potentially for copying information. Copyright cases also consider whether the potentially infringing text replaces the original in the market, which is another problem for cloning games.
Training models, as a concept, isn’t necessarily IP theft, but in practice has involved quite a bit of IP theft. Anthropic itself has come under fire for pirating books and scraping Reddit after agreeing not to scrape Reddit.
Game cloning and reverse engineering are probably completely different things. Reverse engineering games is normally legal. Distributing the assets of the original game is normally illegal. Making a new game clone that is nearly identical to an existing game is not something Claude can determine the legality of.
So one of the only interesting use-cases is getting blocked, aight
There is still every model made in china.
I know, I only use open-source models ran locally anyway and didn’t have a subscribtion to an American one in over a year, but Claude Opus 5.idk arguably was what made these projects possible. Let me word my comment differently: So one of the only reasons to pay Anthropic instead of using open models is getting blocked, good move
Models in china are only a couple of months behind. Even the latest Qwen 27b is about as good as fable 5 (the frontier model 4 months ago)
Models in china, and probably ChatGPT as well are most likely already capable of doing this.
Nah, that’s exaggerated. I have a 32GB Intel Arc Pro B70 and I would say Qwen3.8-27B is barely behind; Qwen3.8-Flash-Next is barely ahead of Opus 4.6, neither of them come remotely close to Fable 5. Some of the Chinese multi-trillion parameter models do compete with the frontier models tho, but nothing you could even remotely hope to run on local hardware.
Interesting, I guess the benchmarks aren’t always accurate.
Which benchmarks? Are you sure you didn’t misread Qwen3.8-Max for Qwen3.8-27B? That one’s up there, but if a benchmark ever compares 27B to Fable then that benchmark is probably just trying to sell you their vibecoded harness or something lol
Edit: Don’t get me wrong, I’ve been especially impressed with Qwen3.8-Flash-Next on my PC and to say it’s barely better than Opus 4.6 is a huge achievement in my eyes. Opus 4.6 was the point where people finally felt like that AI actually spits out mostly useful stuff and having models that can be ran on a realistic PC finally catching up is fantastic, but it’s still far from the frontier stuff. No way in hell am I going to get anywhere with this for decompiling/reverse-engineering stuff unfortunately. I really like coding the important stuff by hand and having it spit out bash scripts in parallel or reviewing my code in the end for oversights, for that I really can’t complain anymore and like Opus 4.6 any model better than that is just luxury. Would I let it vibecode the whole thing then it would end up being unmaintainable slop, but as an actual tool, not a crutch, it’s fantastic. But yeah, Fable 5, not in a dream lol
Hugging face code arena | webdev gives qwen3.8 27b a 1590 and fable 5 high a 1624 which seams very close.
Glm pls
只许州官放火，不许百姓点灯
Welcome to capitalist democracy.
