News for hackers. Stuff that matters.

josecyc

Re:Understanding GPT Tokenizers on Thursday June 22, 2023 @10:30PM (#36442110)
Where are all the tokens for other languages though? How are these getting tokenized?
Re:OpenAI Employee: GPT-4 hasn't gotten worse since March on Thursday June 01, 2023 @02:54PM (#36155675)
I agree I've noticed I need to be quite specific now, it won't realize bugs in the code unless I tell it so. A head to head comparison of the different versions is needed to validate this.
Re:ChatGPT Plugins on Thursday March 23, 2023 @04:32PM (#35280940)
Yep
Re:Cheating Is All You Need on Thursday March 23, 2023 @04:31PM (#35280919)
It's only time till they have access to your filesystem and then that's it
Re:Cheating Is All You Need on Thursday March 23, 2023 @04:27PM (#35280864)
Compute and distribution is my guess
Re:More than you wanted to know about light scattering on Tuesday August 17, 2021 @04:30PM (#28214497)
Why is the sky blue, the moon red and mountains blue
Re:GPT-3: Language Models Are Few-Shot Learners on Friday May 29, 2020 @06:26PM (#23356290)
Haven't read the paper, but it's still unclear how the mechanism of one shot learning works. If the weights are not being updated, how is it "learning"?
Re:Waymo has raised $3B in the last two months on Saturday May 16, 2020 @01:54PM (#23205629)
Uber's drivers represent 40% of the cost per ride to Uber.

A more efficient service (algorithmic operation) which costs 40% less seems to be a no brainer.

A lot of the Tesla robotaxi fleet would be people that still own the cars but just deploy it to the fleet to make money for them.

Re:Waymo has raised $3B in the last two months on Saturday May 16, 2020 @01:47PM (#23205562)
Tesla has Full Self Driving as a clear objective and feature though, autopilot is just their way of releasing the useful features they acquire along the way to FSD. Tesla's difference in its approach (which I think most people tend to miss) is that it aims to be a robust general solution to L5, not a "hard coded", expensive, brittle, geofenced solution, which are basically what all the other L4 startups are aiming for. As famously Levandovski (the guy behind Waymo, Uber and Otto) and others have accepted that Lidar and HD maps are the wrong approach. https://medium.com/pronto-ai/pronto-means-ready-e885bc8ec9e9 It just is not a sufficiently general and economic solution to be scalable across time and space. It doesn't matter if Waymo's only focus has been FSD (which it completely hasn't) if they have a wrong approach they won't achieve it.
Re:Uber loses $2.9B, offloads bike and scooter business on Saturday May 09, 2020 @10:57AM (#23125180)
Not really, Uber has accumulated around 3 million drivers, Tesla has around 1 million cars with FSD capabilities and growing much faster than Uber. Tesla has the advantage that their fleet would be composed with their owned cars plus customer cars, my prediction would be that they would quickly becoming bigger than Uber's fleet of drivers once they have FSD and with a click of a button you could send your car to make money for you.
Re:OpenAI Releases Largest GPT-2 Text Generation Model on Tuesday November 05, 2019 @05:12PM (#21457313)
Thanks! any plans to make it faster? Like talk-to-transformer? Seems much more slower
Great job, ocean cleanup very interesting data right here. Tracking the problem in that way is more of a first principles approach. Happy that Mexico is so low on the amount.
Re:Amazon Pitches Facial Recognition to Monitor Immigrants on Tuesday October 23, 2018 @11:06PM (#18289390)
Amazon needs a manifesto
I don't know about this. Countless times researchers discover the next battery so much better than the current state-of-the-art being deployed, only to find out they just can't scale it. New battery technology is really hard.