The intentional restriction closely matched the forecast's specific parameters.
I am uncomfortable with the number of @DKokotajlo's predictions coming true
Another quite successful prediction by @DKokotajlo : Fable is intentionally nerfed for frontier ML research. This is within ~3 months of Daniel's prediction of Q1 2026 (made in 2023).
Although I don't think Mythos is automating ML research to the same extent as his prediction.
Some users respect DKokotajlo's track record after Fable's intentional nerf validated his predictions, while others accuse Anthropic of treating those forecasts as self-fulfilling prophecies.
@a_karvonen Thanks! Note that this particular comment wasn't technically a prediction by me, so I shouldn't get that much credit. Iirc it was a hypothetical scenario I took seriously but wasn't my median timeline. At least that's what my fuzzy memory says.
@a_karvonen @wrongplace_eth @DKokotajlo DKokotajlo has a good track record with predictions and Fable nerfing is a big signal. His Q1 2026 call is looking more real by the day
@theojaffee @a_karvonen @DKokotajlo https://www.lesswrong.com/posts/K2D45BNxnZjdpSX2j/ai-timelines
@a_karvonen @DKokotajlo I have an alternate theory: Mythos is nerfed for frontier ML research because it helped too much Anthropic in the past months (even in its own training?). Helping others on this could reveal Anthropic own trade secrets
@a_karvonen @DKokotajlo Yes, "basically doing all the work itself" is incorrect, Anthropic just said explicitly it doesn't come close to replacing its researchers.
@a_karvonen @DKokotajlo source for the Kokotajlo comment?
@a_karvonen @DKokotajlo The truth is that Anthropic uses Daniel’s predictions as self-fulfilling prophecies. They were all preloaded in the Gantt charts years ago.
@a_karvonen @DKokotajlo I might be the only one to really have an issue here but have noticed an insane amount of slow and Steady death of Claude ability to help with my computer vision training / methods etc for about 6 months
@a_karvonen @DKokotajlo It'd be crazy if Q3 is when all the frontier labs declare that we officially enter the era of ASI
@a_karvonen @DKokotajlo all according to keikaku
@a_karvonen @DKokotajlo Somebody should put this idea in a book!
Oh wait.
@allTheYud @So8res
@a_karvonen @DKokotajlo Yes, Anthropic really do not like China companies distilling their models
@a_karvonen @DKokotajlo ml tasks have not really improved since opus 4.6 iirc so they have probably been stealth nerfing since then, which was q1
@a_karvonen @DKokotajlo Real respect for him!
@a_karvonen @DKokotajlo There is some distorting effect in that many of the core decision makers in Anthropic have read Daniel's writing.
@a_karvonen @DKokotajlo what? the prediction doesn't seem to imply _intentional_ nerfing
@a_karvonen @DKokotajlo i guesss…. we just gotta do homework 😔
@a_karvonen @DKokotajlo So I am building my own interference engine running Deepseek 4 flash 284b on Nvidia 3090 with custom cuda kernel and nvme streaming at 3.5 tokens / sec. This is ML research in interference, will I be blocked?
@Atenov_D @a_karvonen @wrongplace_eth @DKokotajlo You’re gonna freak out when you find out what year it is
@eric23332 @a_karvonen @DKokotajlo of course they’d say so
The intentional restriction closely matched the forecast's specific parameters.
I am uncomfortable with the number of @DKokotajlo's predictions coming true
Another quite successful prediction by @DKokotajlo : Fable is intentionally nerfed for frontier ML research. This is within ~3 months of Daniel's prediction of Q1 2026 (made in 2023).
Although I don't think Mythos is automating ML research to the same extent as his prediction.