|
Jack Clark
@
jackclarkSF
San Francisco, CA
|
|
Policy Director @openai & @indexingai Newsletter: jack-clark.net Past: @business Neural Net Reporter. @theregister Distributed Systems Reporter.
|
|
|
27.160
Tweetovi
|
3.412
Pratim
|
38.853
Osobe koje vas prate
|
| Tweetovi |
| Jack Clark proslijedio/la je tweet | ||
|
Twitter Safety
@TwitterSafety
|
16 h |
|
Our new rule: You may not deceptively share synthetic or manipulated media that are likely to cause harm. In addition, we may label Tweets containing synthetic and manipulated media to help people understand their authenticity and to provide context.
bit.ly/2ttN0LP
|
||
|
|
||
| Jack Clark proslijedio/la je tweet | ||
|
|
John Carmack
@ID_AA_Carmack
|
20 h |
|
If an AI is going to grow up watching a lot of television, there are likely better and worse choices for material. I sent a request to the partnerships email for @SesameWorkshop to check the availability of the full back catalog of Sesame Street for research purposes. \
|
||
|
|
||
| Jack Clark proslijedio/la je tweet | ||
|
Nick Lutsko
@Nick_Lutsko
|
3. velj |
|
Here's an unusual use of AI/ML in climate
"Using Simulated Data to Generate Images of Climate Change"
arxiv.org/abs/2001.09531
-> besides improving climate models, weather forecasts, etc use ML to create images of potential climate impacts, to motivate action
h/t @jackclarkSF
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
2. velj |
|
What happens if sentience appears like blackmold in an apartment? A secret singularity among connected appliances. An actual IoT revolution.
"The Fire Alarm That Woke up" from Import AI 182.
Read: jack-clark.net/2020/01/27/imp…
Subscribe: us13.campaign-archive.com/home/?u=67bd06… pic.twitter.com/fCB3I0IGYx
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
2. velj |
|
Yeah, I figured this was the reason.
Question: if there was a way to magically 5X GPUs at all academic institutions do you think it would help or hinder AI research over the long run?
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
2. velj |
|
did a writeup of this paper for Import AI this week. A couple of things:
1) Clear writing that focused on the precise tweaks you did with helpful explanations for why they improve training
2) Was wondering whether there is utility on testing this in other GVGAI games also?
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
2. velj |
|
What are good examples of such problems?
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
30. sij |
|
The best part was when Yahoo! dropped the exclamation point they tried to get us to stop doing it. Didn't work
|
||
|
|
||
| Jack Clark proslijedio/la je tweet | ||
|
Jack Clark
@jackclarkSF
|
30. sij |
|
Everyone has a mutual interest in a basic level of stability, because when advanced technologies like AI go wrong, they can proliferate rapidly and chaotically, without regard for borders. If we just have safe powerful tech in democracies, we aren't actually much safer.
|
||
|
|
||
| Jack Clark proslijedio/la je tweet | ||
|
Joshua Achiam
@jachiam0
|
30. sij |
|
Check it out: Spinning Up is now a sort of "Rosetta Stone" for deep RL, with simple implementations of VPG, PPO, DDPG, TD3, and SAC in both PyTorch and TFv1. spinningup.openai.com/en/latest/spin…
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
30. sij |
|
Thank you for consistently advocating for dialogue on safety/stability across countries with different ideologies. So important!
|
||
|
|
||
| Jack Clark proslijedio/la je tweet | ||
|
Elsa B. Kania
@EBKania
|
30. sij |
|
In my opinion, and by my assessment, challenging conversations with our competitors is not a ‘softer’ approach in any sense. Of course, our policy brief and commentary also highlight the importance of coordination and collaboration with our allies and partners. twitter.com/ArthurLHerman/…
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
30. sij |
|
Everyone has a mutual interest in a basic level of stability, because when advanced technologies like AI go wrong, they can proliferate rapidly and chaotically, without regard for borders. If we just have safe powerful tech in democracies, we aren't actually much safer.
|
||
|
|
||
| Jack Clark proslijedio/la je tweet | ||
|
OpenAI
@OpenAI
|
30. sij |
|
We're standardizing OpenAI's deep learning framework on PyTorch to increase our research productivity at scale on GPUs (and have just released a PyTorch version of Spinning Up in Deep RL): openai.com/blog/openai-py… pic.twitter.com/lgvqDdWDoB
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
29. sij |
|
RL! What is it good for?
Actually... Real world robotics! twitter.com/CovariantAI/st…
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
29. sij |
|
They show tons of samples in the paper, which you might want to read: arxiv.org/pdf/2001.09977…
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
29. sij |
|
"tackling safety and bias in the models is a key focus area for us, and given the challenges related to this, we are not currently releasing an external research demo."
(Extremely cool work that also highlights the policy challenges of synthetic text. Lots of work to do!) twitter.com/GoogleAI/statu…
|
||
|
|
||
| Jack Clark proslijedio/la je tweet | ||
|
Aneel
@aneel
|
27. sij |
|
“AI models are going to magnify whichever culture they've been trained on.” —@jackclarkSF
|
||
|
|
||
|
Jack Clark
@jackclarkSF
|
26. sij |
|
Blelky Mungo!!!!!
|
||
|
|
||
| Jack Clark proslijedio/la je tweet | ||
|
Irene Solaiman
@IreneSolaiman
|
26. sij |
|
I'm at FAT* in Barcelona ready to *fatten* up my FAT+ research!
Come get a cafe con leche with me!☕️
|
||
|
|
||