This is mindblowing. Like this could be real, and I'm learning stuff from it: There's an Indian epic that's 10 times as long as ...
There's some audio distortion (sounds like clips cut together, little "notches" in the soundscape) but apart from that, and some weirdness in sensing "the spatial location" where this audio was recorded...the concepts and the dialog are amazing.
Some parts are weird...but people can be weird. It you tidied this up, and added the right sounds affects and audio processing to this, without the cue that this is AI generated...holy fuck, I think people would believe it. Particularly if you cut it together as a "highlights reel". Jobs does sound a bit off tho, a bit thin...there should be enough data on him to do a sparse reconstruction of his voice to a level of accuracy beyond human discernment tho.
The thing this got wrong about Job's voice cadence, tho is: Jobs speaks a lot more slowly and deliberately, and with a lot more pauses, than here. I suspect the cadence / timing is not so emphatically modelled by this AI.
I think also they're missing some emotional trajectory coherence in both their voices. Like the emotional register of the voice does not sound or transition as naturally, and is less diverse.
Incredible PoC. AI folks are the new dark wizards. WTF can they not do? That list is shorter
I agree it's a fantastic PoC. Jobs' voice sounds very heavily derived from keynote speeches. People speak with different intonation and cadence in different situations and the uncanny valley effect here is due to having him speak casually on a podcast but in keynote voice. It'd be like training a Paul McCartney AI voice solely on Beatles songs - it wouldn't sound anything like McCartney's real life speech.
Joe Roegan I can guarantee you just had a MUCH better amount of background data to feed this thing - I doubt we have nearly enough high quality audio of Steve Jobs speaking casually to make a convincing facsimile.
A future Steve Jobs though where everyone has a smartphone and more is recorded.....that is kind of scary.
Stuff like this kind of makes me want to secretly record myself in very high quality in tons of situations and then put it all in a safe with a clause in my will to have it all donated to some AI research firm.
> The thing this got wrong about Job's voice cadence, tho is: Jobs speaks a lot more slowly and deliberately, and with a lot more pauses, than here. I suspect the cadence / timing is not so emphatically modelled by this AI.
That's because AI Broe Jogan got AI Jeve Stobs high before the show started.
>> That's because AI Broe Jogan got AI Jeve Stobs high before the show started.
He does this for every interviewee! I was shocked, but pleased when the Substack vp of comms wrote about her experience whilst colloquially escorting the Substack CEO for the interview. [1] [2]
>> At a long desk in the big main room sits an attractive nurse. She offers us an enhancer of B12 or NAD+, through a shot or an IV.
>> I get a shot of NAD+, which is supposed to be good for energy and metabolism. NAD stands for Nicotinamide Adenine Dinucleotide, and I don’t know what the + is. (Actually, I don’t know what any of it is, but the nurse said she takes it, and if you saw this woman, you too would ask for a shot of whatever she’s on.)
Ultimately, this is pretty crazy to me. Those along with the interviewees get “high” too and since Steve Jobs, this is the first time it’s been written about it.
I wish people knew more about this! It’s been a while since Steve Jobs has been interviewed. 1000+ shots later, Rogan has grown and is stronger than ever.
Neither of those substances are used to get what is
colloquially known as "high" (one is a vitamin), you don't take either as a recreational drug.
I think what GP is referring to is Joe Rogan smoking weed with some guests before, during and after the show, rather than some vitamin and pro-metabolism shot.
There are IV clinics popping up all over the place, so there has to be some sort of streamlined process. No way they are full of MDs overseeing everything.
It was really hard to find good quality of Steve Jobs' voice, most of his speeches are keynotes on stage, I honestly was surprised that we managed to get to that quality from such poor quality recordings.
You can hear the "space" when Steve Jobs is talking that a lot of the training material are from keynotes on a stage in a big space. That's why it sounds a bit "thin" compared to Joe Rogans voice which is probably mostly based on recordings in a studio with Joe being close to a microphone (and a proper microphone for recording)
I work in audio post production and would be happy to take a crack at cleaning up your source audio. There's a lot we can do to isolate the dialogue from the space and restore presence.
AI Jobs' comments on the emergence of spirituality from the Indian subcontinent and his belief that intuitive knowledge is as real as scientific knowledge sounds like something he absolutely would have said.
That might just be lack of inputs for Jobs compared to Rogan. Like we have tons of Rogan speaking in a podcast setting. We have very few Jobs interviews in comparison and the ones we do have take place over a much longer stretch of his life (so speech patterns will have changed).
I might also guess that a decent amount of Jobs' audio might be from non-conversational contexts like keynotes and lectures.
There's some audio distortion (sounds like clips cut together, little "notches" in the soundscape) but apart from that, and some weirdness in sensing "the spatial location" where this audio was recorded...the concepts and the dialog are amazing.
Some parts are weird...but people can be weird. It you tidied this up, and added the right sounds affects and audio processing to this, without the cue that this is AI generated...holy fuck, I think people would believe it. Particularly if you cut it together as a "highlights reel". Jobs does sound a bit off tho, a bit thin...there should be enough data on him to do a sparse reconstruction of his voice to a level of accuracy beyond human discernment tho.
The thing this got wrong about Job's voice cadence, tho is: Jobs speaks a lot more slowly and deliberately, and with a lot more pauses, than here. I suspect the cadence / timing is not so emphatically modelled by this AI.
I think also they're missing some emotional trajectory coherence in both their voices. Like the emotional register of the voice does not sound or transition as naturally, and is less diverse.
Incredible PoC. AI folks are the new dark wizards. WTF can they not do? That list is shorter