zaptrem

7 ms·

61.
by zaptrem 1y ago
This sort of tech is also useful in that situation since it can better understand and deliver vocal nuances (e.g., emphasis/tone that delivers meaning)
62.
by zaptrem 1y ago
I'm around the age these guys were during this story. I feel the exact opposite way. I spent middle/high school feeling similarly, only pining for the 2000s ("wow, with smartphones and the internet the industry was wide open
63.
by zaptrem 1y ago
In addition to what Centigonal said, even if the autoencoder was trained on only speech data, an all zero vector is probably just out of distribution (decoder has never seen it before) and causes weird sounds. However, given the hallucinati
64.
by zaptrem 1y ago
I may know why that first one happens! They’re not correctly padding the latent in their decoder (by default torch pads with zeros, they should pad with whatever their latent’s representation of silence is). You can hear the same effect in
65.
by zaptrem 1y ago
> Additionally, rare hallucinations in Voice Mode persist with this update, resulting in unintended sounds resembling ads, gibberish, or background music. We are actively investigating these issues and working toward a solution. Would be
66.
by zaptrem 1y ago
Taping over the camera causes it to lock out (it can tell the difference between that and night time).
67.
by zaptrem 1y ago
Even on stuff it has no chance of doing on its own, I find it useful to basically git reset repeatedly and start with more and more specific instructions. At the very least it helps me think through my plan better.
68.
by zaptrem 1y ago
Maybe also 11L’s Scribe model?
69.
by zaptrem 1y ago
Does Google hold the rights to this content? They've been providing a less capable version of this service for >25 years.
70.
by zaptrem 1y ago
Transformers are what made ML infinitely scalable and caused a huge amount of progress in very few years since everyone could just go scale things. However, idk how many of those papers actually even cite the transformer paper?
71.
by zaptrem 1y ago
Good chance a future llama will output image tokens
72.
by zaptrem 1y ago
My favorite: https://www.youtube.com/watch?v=5pidokakU4I
73.
by zaptrem 1y ago
Fighting back against one set of AI opinions for everyone (which tended to be criticized as woke) to better reflect the “vibes” of the user is also part of what got the most recent 4o release to start enthusiastically agreeing with flat ear
74.
Is AI alignment repeating the mistakes of "New Coke"?(en.wikipedia.org)
2 points by zaptrem 1y ago|4 comments
75.
by zaptrem 1y ago
"New Coke" is one of the most notable failed product launches in the American food and beverage industry, and I feel like some of its core lessons are becoming increasingly relevant to modern AI developers. For those <40: In th
76.
by zaptrem 1y ago
Where does the B come from?
77.
by zaptrem 1y ago
Lol why do the info sheets include blood type?
78.
by zaptrem 1y ago
I think when they were figuring out RLHF they avoided this by interleaving RLHF and normal cross entropy on training set gradients.
79.
by zaptrem 1y ago
Which part of "General Intelligence" requires replacing white collar workers? A middle schooler has general intelligence (they know about and can do a lot of things across a lot of different areas) but they likely can't repla
80.
by zaptrem 1y ago
Mostly 1 I think. There are a few open source efforts doing what you mentioned https://github.com/EleutherAI/aria
81.
by zaptrem 1y ago
I do find it likes to subtly reformat every single line thereby nuking my diff and making its changes unusable since I can’t verify them that way, which Sonnet doesn’t do.
82.
by zaptrem 1y ago
Prompt on bottom is also easier for humans to read as I can have my actual question and the model’s answer on screen at the same time instead of scrolling through 70k tokens of context between them.
83.
by zaptrem 1y ago
So we’d replace income tax with a regressive tax on consumption
84.
by zaptrem 2y ago
I work on music models, and this is a very cool paper! There are no papers that go into depth on how token-based AR music models (that aren't absurdly inefficient like Yue) are trained. I'm particularly interested in your semantic
85.
by zaptrem 2y ago
Token by token diffusion was done by MAR https://arxiv.org/abs/2406.11838 and Fluid (scaled up MAR) https://arxiv.org/abs/2410.13863 When you randomly pick the locations they found it worked okay,
86.
by zaptrem 2y ago
> "Yahoo!" offered their index as a text file that would fit on a 1.44 floppy Are there still records of this somewhere? Can't find anything on Google. I don't remember this as I didn't exist but it sounds intere
87.
by zaptrem 2y ago
The title of the video says “Self Driving Car” which can and is misleading viewers into thinking it’s a test of Tesla’s “supervised Full Self Driving” product since they do not sell other products that use that term. FSD at one point had a
88.
by zaptrem 2y ago
To be clear, this is like buying a car that has traffic aware cruise control available as an option, but turning it down, then insisting the TACC is broken and dangerous because it doesn’t work on your car.
89.
by zaptrem 2y ago
Stripe issue should be fixed, second issue likely happens if you go to the api page sometime in your session before going to the profile page and then you try to edit your picture. We'll work on that. Thanks for reporting!
90.
by zaptrem 2y ago
Many people may not have the time, talent, and/or dedication, but I still think they should get to make music.

More ›