Apparently the voices still can have variations in speed and inflection. But they barred singing, which was a key feature during the official demo :(
Apparently the voices still can have variations in speed and inflection. But they barred singing, which was a key feature during the official demo :(
It’s certainly going to be done on device. Existing pixels can already do this with Live Captions
TIL. This is weird as fuck.
Been there done that 💀
Reboot mid flight is a funny solution
I’ve spent over 2,000 hours on YouTube this year alone and am in the target demographic of this study. I watch a lot of videos in the background while I work, commute, or just chill, to keep myself stimulated.
Although not all of the content I watch is necessarily educational, a grand majority of it is. Whenever there’s a science video in my feed, I’ll probably click it. I’m subscribed to Veritasum, TED, Vox, No Boilerplate, etc.
Usually the plan is to go to China and then flee to a friendlier country. There are also people in China who work with North Koreans to help them defect.
I think it’s awesome that they paid the dev. More companies should do that for open source projects
They partnered with Anthropic and that seems to be going, fine I guess? But Anthropic’s models definitely need work.
It looks like it was made in MS paint lmao
It is a lot better now. Used to be absolutely horrible
As much as I’d love to see them back in OpenAI, I don’t think Emmett Shear will give up.
I have a soft spot for Greg since he was the one who introduced the world to GPT 4 on that developer livestream
What a disaster
Google Pixel has this but in the dark it just defaults to auto rotate mode so I can’t really use my phone in bed vertically… I wish they could let me choose whether it defaults to lock mode or auto mode.
You could start your own VPC data center with this lmao
It’s actually pretty difficult still if you’re using secure hashing functions like Argon2 and bcrypt because they’re hard on memory and computational power, meaning brute force attacks are pretty much infeasible, both due to hardware requirements and long hashing times.
This is from 2022 lmao
In Drake’s new album he says “android her messages is lime green”, and that came out days ago
Smaller LLMs have huge performance tradeoffs, most notably in their abilities to obey prompts. Bard has billions of parameters, so mobile chips wouldn’t be able to run it.
Nothing can stop 100% of bots. The goal with captchas like Turnstile is to use a significant portion of your resources to the point it’s expensive and slow to perform an attack.
Turnstile runs many background checks on your browser, so headless browsers automatically become futile.
JavaScript PoW challenges are performed that take up multiple seconds of execution time, memory and CPU. This alone is a deterrent because sequential attacks become extremely long to execute.
Concurrent attacks are still unfeasible because Turnstile ups the difficulty if it detects something is up, and receiving requests from thousands of botnet IPs is bound to trip an alarm.
Fears of copyright infringement. Maybe they trained using music 🤔