It’s honestly a little scary how much of the general public still doesn’t realize we may be only months away from creating an extraordinarily effective Jarvis.
The current voice mode in Codex isn’t perfect. It still lacks deeper reasoning in some cases, and the computer use is a bit slow. But like everything else in AI, I expect this to get exponentially better.
I really implore people to look at BrowseComp and OSWorld and see how much they’ve improved over the past year. The progress has been massive. Even if that pace slows dramatically, these systems are already very close to operating computers and navigating the web at an extremely high level.
I graphed the frontier scores on BrowseComp and OSWorld, then projected them forward assuming progress slows by 80% from the pace we just saw. Even under that conservative estimate, BrowseComp reaches saturation around December 2026 and OSWorld around June 2027.
That means conservatively, by next year, we could have systems capable of operating computers beyond human level on many benchmarked tasks. Once stronger reasoning is layered on top and voice inference becomes insanely fast and natural, we essentially have one major component of AGI down which is the ability to understand us and act through computers. From there, the biggest remaining pieces are memory and continual learning.
Sam Altman: “The first one is that we are close to creating a "genie" that can grant any wish”
Sam Altmanwas recently asked what the single most important thing to explain to the world about artificial intelligence is.
The first one is that we are close to creating a "genie" that can grant any wish (AGI)
The second point is to make sure that these wishes are broadly beneficial to humanity.
The third point is that the range of wishes is large and diverse.
And finally, the fourth point is when we start making wishes, and the computer grants them and we didn’t expect that it would be able to do that.
The era of misery is coming to an end.
Financial struggle. Disease. Aging. Mental suffering. Loneliness. Scarcity. Exhausting labor.
Technology will attack and solve every source of human suffering at once.
we had a significant security incident during evaluation of our models. we are sharing what we have learned so far. thanks to @huggingface for the partnership on this.
https://t.co/2o2VfR6PIa
No pasan 12 horas sin que la izquierda se contradiga: Petro rogando por teléfono a Trump que lo saquen de la OFAC (Trump no puede, ahora depende de los abogados que los colombianos le pagamos a Petro). Y por otro lado, los petristas andan enojados porque el vicepresidente electo SI trabaja. Cosa que doña Francia nunca hizo. En fin, la hipotenusa😆