[ML News] GPT-3 learns to edit | Google Pathways | Make-A-Scene | CLIP meets GamePhysics | DouBlind

#mlnews #gpt3 #pathways

Your updates on the latest and greatest from the depths of Machine Learning!

Sponsor: Weights & Biases

OUTLINE:
0:00 - Intro
0:15 - Weights & Biases Report about Reports
2:45 - GPT-3 learns to edit
6:30 - Make-A-Scene: Text-to-Image with Human Priors
8:00 - Pathways: Google's new High-Performance ML scheduler
10:45 - DouBlind: Open Peer-Review
12:45 - CLIP meets GamePhysics
14:40 - Residual Quantization pushes Image Generation SOTA
16:15 - Helpful Things

References:
Weights & Biases Report about Reports

GPT-3 learns to edit

Make-A-Scene: Text-to-Image with Human Priors

Pathways: Google's new High-Performance ML scheduler

DouBlind: Open Peer-Review

CLIP meets GamePhysics

Residual Quantization pushes Image Generation SOTA

Helpful Things

Links:

If you want to support me, the best thing to do is to share out the content :)

If you want to support me financially (completely optional and voluntary, but a lot of people have asked for this):
Bitcoin (BTC): bc1q49lsw3q325tr58ygf8sudx2dqfguclvngvy2cq
Ethereum (ETH): 0x7ad3513E3B8f66799f507Aa7874b1B0eBC7F85e2
Litecoin (LTC): LQW2TRyKYetVC8WjFkhpPhtpbDM4Vw7r9m
Monero (XMR): 4ACL8AGrEo5hAir8A9CeVrW8pEauWvnp1WnSDZxW7tziCDLhZAGsgzhRQABDnFy8yuM9fWJDviJPHKRjV4FWt19CJZN9D4n

[YannicKilcher]

OUTLINE:
0:00 - Intro
0:15 - Weights & Biases Report about Reports
2:45 - GPT-3 learns to edit
6:30 - Make-A-Scene: Text-to-Image with Human Priors
8:00 - Pathways: Google's new High-Performance ML scheduler
10:45 - DouBlind: Open Peer-Review
12:45 - CLIP meets GamePhysics
14:40 - Residual Quantization pushes Image Generation SOTA
16:15 - Helpful Things

[ai_doyup]

14:46 It's a great honor to have our work mentioned on your YouTube channel !!

[florianhonicke]

Nice 👍 the title looks promising already

[andregn]

Really great quality! Keep the great work! 😊

[brandomiranda]

how often do you release the ML news segment?

[sfarmapietre]

If you change the "dfs" function name to something else, like "blabla", it will generate just this as a docstring: " This function returns the id of the node with the given value." So yeah, I think it might not be so smart.

[MrStarchild]

Thank you for the videos and the cool summarizes!

[mgostIH]

16:20
Seems like Stumpy uses Matrix Profiles, a very simple but effective technique for time series that's quite recent, I suggest anyone to check it out!

[Guytron]

The big question for the "GPT-3 Learns to Edit" subject is how long until that feature is implemented into Github Copilot or other VSCode plugin? Where I guarantee it will see a lot more use.

[serta]

Always looking forward to the ML news 📰 🤓

[herp_derpingson]

2:50 Technically, there are less pixels of you in the frame so there is less of you.
.
13:40 This would really be good for adult movie search. The current search engines in adult websites suck. It would be great, if the search engine actually watched the contents instead of just the title.
.
17:00 Good to see NVIDIA pick up where open AI GYM left off. I think rockstar games forbid researchers from using GTA5 in their experiments. Although, it would be a pretty scary world where an AI learns about the real world from GTA 5 ;)

[Robert-jsf]

Great video! Both humorous and informative at the same time

[DanFrederiksen]

Great videos as usual. How do you find new papers of interest, just go through all of them and look at the title and the pictures?

[memegazer]

Can you cover midjourny text to image...can't find much about it

[Kram]

Is there any info yet as to *how* that GPT-3 edit/insert stuff actually works? Like, it's a linear language model. How can it look at the future?

[jameshughes]

I'd love to see a version/remake of the imagenet music video, maybe a year after its first post, with the newer text to image generators.

[johnyjohnjohnson]

your song as a comon test for image from text generators : D

[serta]

I wonder what the large multilingual 172b parameter model that is now trained at Huggingface 🤗 will produce

[siruitao]

Haha, "there are models other than deep learning" really hits

[aihacker]

One subscriber = one pathway at the Google data center :D