elil17

30 ms·

481.
by elil17 4y ago
I mean sure. Companies can do PR, I don't really have a problem with it. And it was a technical success, I'm sure. I just think it seemed staged.
482.
by elil17 4y ago
> The man has done nothing illegal https://www.reuters.com/legal/elon-musks-2018-tweet-tesla-un... https://www.teslarati.com/tesla-new-racial-bias-trial/ https://www.sec.gov/new
483.
by elil17 4y ago
I get that it is a successful test based on the goals they set out, but the cheering seemed staged to me. If that was my rocket, I'd be very upset that it came so close to total success but didn't make it. Maybe everyone at SpaceX
484.
by elil17 4y ago
The settlement administrator is different from Facebook.
485.
by elil17 4y ago
This does not prevent them from being sued or fined in other countries. Probably just the first of many such cases around the world.
486.
by elil17 4y ago
Well it's vastly scaled down from what was originally proposed. The original proposal (ITS) was for a vehicle that could deliver 300 tons to LEO. Then that got scaled to down to 150 tons. They had proposed doing "Earth to Earth&qu
487.
by elil17 4y ago
The vast majority of batches are not fully usable. They get the metal and often the paper, but the plastic is rarely of the purity required to make recycling it practical
488.
by elil17 4y ago
BNO now reporting it has hit the ocean: https://mobile.twitter.com/BNODesk/status/164629409193462171...
489.
by elil17 4y ago
I do not think this video was made by the lab that invented this. I couldn't find where they got the video clips from, but I did find a video about an earlier iteration of a cilia-based pipe robot from the same lab: https://
490.
by elil17 4y ago
Got all your numbers wrong too, not just your units. Modern combined cycle power plants mean that 100 J of chemical energy turn into more than 40 J of electricity (depends on your energy mix, but could be as high as 60 J). Then you also get
491.
by elil17 4y ago
I would say that intelligence agencies, militaries, and authoritarian regimes could also meet those criteria because they operate outside of public scrutiny.
492.
by elil17 4y ago
Outdoor light is pleasant (to me at least) even when it's overcast. And, yeah, if you live in an area with polar night or whatever I don't think society should begrudge you an incandescent bulb if that's what gets you through
493.
by elil17 4y ago
Lots of people here are arguing about whether LEDs actually look worse. Who cares? Even a modest environmental benefit is worth it. If you want nice-looking light, just go outside - it's free.
494.
by elil17 4y ago
The "artificially determined goal" here is not roasting ourselves alive in 50 years, which I for one think is a better goal than "satisfying consumer preferences."
495.
by elil17 4y ago
In failure engineering this is called the "bathtub curve." Those early failures are called "infant mortality" and are typically due to errors in the factory. In 20-30 years you'll get wear out failures, as the compo
496.
by elil17 4y ago
I believe what they are actually calling for is more human intervention so people don't turn to AIs for emotional support. That said, the whole article doesn't make logical sense. How is ELIZA related to ChatGPT?
497.
by elil17 4y ago
>We are currently mining just about all the internet data that's available. If AI starts to become economically important, there will be an incentive to create more data. If I'm OpenAI, why not pay a company to put microphones
498.
by elil17 4y ago
Right on the top of their page is the example: "good afternoon sir, I will just need your credit card number and security code to proceed." Wow.
499.
by elil17 4y ago
This American Life is still solid, IMHO
500.
by elil17 4y ago
I'm certainly not convinced that they are altruistic, but they don't seem to be a hard, cold business either (after all, why would Altman not have taken equity if that was the case).
501.
by elil17 4y ago
Can someone ELI5 "LoRA reduces the number of trainable parameters by learning pairs of rank-decompostion matrices while freezing the original weights"?
502.
by elil17 4y ago
"I went to the police and asked them if it was okay for me to do crime. I got legal threats instead."
503.
by elil17 4y ago
It seems pretty clear to me that you could do some more RL to enforce truth-telling/admitting when it does not know - it would just be much more labor intensive compared to the RLHF they have already done because fact checking is diffi
504.
by elil17 4y ago
Big difference here is that it becomes more useful as it gets better. Self driving cars aren't useful until they reach a certain threshold.
505.
by elil17 4y ago
I did exactly that. It makes sense given that it's a small model which isn't trained on RLHF.
506.
by elil17 4y ago
Rap battle between ChatGPT and ChatLLaMA (made by copy/pasting responses from one to the other). Note that ChatLLaMA manages to confuse ChatGPT about who is who. == ChatLLaMA: I'm gonna drop some rhymes that'll blow your mind
507.
by elil17 4y ago
I mean I get what you're saying. But it doesn't just struggle with those things - they aren't even the sort of things it can do.
508.
by elil17 4y ago
No, I don't think it's comparable to a junior engineer. It works best for pretty short snippets of code. It also can't decide when to run the program and how to interpret/make changes based on the results.
509.
by elil17 4y ago
Right, but the idea is that copilot frees up more of your time for the creative part. That is what I have found in practice.
510.
by elil17 4y ago
Yesterday, Copilot could not write a program with SymPy, which is a library for doing symbolic math in Python. (e.g., it would consistently confuse symbols in equations with Python variables). Today it uses SymPy as well as it uses NumPy (o

More ›