
Machine Learning @GoogleAI formerly @UofT/@VectorInst
Joined June 2017
- Tweets2.7K
- Following932
- Followers1.5K
- Likes38.7K
Pinned Tweet
Excited to have a very weird paper accepted to NeurIPS! It’s half dataset paper, half meditation on what it would look like to have more systematic methods for evaluating model performance on unusual examples, contexts or subgroups. If that interests you: arxiv.org/abs/2110.13223
David Madras retweeted
A key point missing in a lot of post debate analysis is that Trump’s claim about immigrants eating pets almost perfectly syncs up to the piano in the Peanuts theme song.
David Madras retweeted
I don't even know what to say, except that it's funny :)
Photographer Disqualified From AI Image Contest After Winning With Real Photo petapixel.com/2024/06/12/pho…
Pretty sure we already live in a world where anyone can make music
We’re excited to announce that Suno has raised $125 million to build a future where anyone can make music, and where technology amplifies our most precious resource: human creativity. We will use this funding to accelerate product development and grow our world-class team of music makers, music lovers and technologists.
Our community of musicians deserves the very best tools, and building the very best tools requires the very best talent — not just technological expertise, but a genuine love of music. If you want to be part of this journey and help shape the future of music, join us here: suno.com/careers
Most importantly, thank you to our amazing community. We’ve been heads-down, hard at work on a number of major updates. We think you’re going to love them, and we look forward to sharing more with you soon!
To read our full announcement, visit our blog: suno.com/blog/fundraising-an…
worth thinking about how general artistic taste (ie what's "cool") is affected when certain types of content become increasingly simple to produce at high volume
C3PO x Childish Gambino 🤖 👑
100% AI (Official Music Video)
@openAI + @runwayml + @suno_ai_ + @resembleai + @fable_motion + @midjourney + @topazlabs
6
657
"there is nothing creative about typing a prompt and receiving an output" is a hill I will die on
2
452
My friend and collaborator @elliot_creager is looking for students! If you’re into ML and societal impacts, can’t recommend working with Elliot highly enough
Once more, in verbose mode: as of today I am an Assistant Professor @WaterlooENG (ECE dept), and faculty affiliate @VectorInst and @TorontoSRI.
Please help me get the word out - I will be hiring motivated students who want to study machine learning and its societal implications
Huge congrats to the authors on this, including my PhD supervisor @zemelgroup! Not an exaggeration to say this line of work totally changed my academic career, it’s great to see their research foresight recognized
Congratulations to Richard Zemel, @LedellWu, @kswersk, Toni Pitassi, Cynthia Dwork for winning the #ICML2023 Test-of-Time Award on their paper "Learning Fair Representations"!
Announcement: icml.cc/Conferences/2023/Tes…
Paper: proceedings.mlr.press/v28/ze…
Content storage seems quite different from content generation to me - the analogy is much more clearly a stock photo company or streaming service, both of which certainly pay artists
I might regret wading into this, but my hot take about AI and copyright is that the companies shipping generative models to people don't have any more obligation to pay royalties to artists than VCR manufacturers have to pay movie studios.
3
489
Any pointers to folks doing interesting, well-informed writing on the future impact of AI in music?
1
2
428
David Madras retweeted
A new study suggests that #MachineLearning models may not replicate human decisions when using solely fact-based data to make judgments that require an understanding of social norms. scim.ag/2BR
David Madras retweeted
💡 How can we train machines to make normative decisions? A new paper in @ScienceAdvances by Aparna Balagopalan, Gillian Hadfield, et al. describes how ML systems can be improved by providing data that explicitly reflects value judgments.
Learn more: uoft.me/9eU
David Madras retweeted
Very excited about this work out today in Science Advances with @dhadfieldmenell @AparnaBee @MarzyehGhassemi @david_madras & David Yang.
💡 How can we train machines to make normative decisions? A new paper in @ScienceAdvances by Aparna Balagopalan, Gillian Hadfield, et al. describes how ML systems can be improved by providing data that explicitly reflects value judgments.
Learn more: uoft.me/9eU
Really excited to finally have this published - @AparnaBee did a ton of great work spearheading this project over the past 3 years and we have some provocative conclusions about data labelling for normative decision-making. If that interests you, check it out!
Excited to share our Science Advances paper (science.org/doi/10.1126/scia…) w/ @david_madras , David H. Yang, @dhadfieldmenell , @ghadfield , and @MarzyehGhassemi !
Also presenting this at the CSAIL LIDS ML Symposium (May 12, 3:30pm EST) in-person!
Paper highlights in thread below:

ALT Machine learning systems trained with factual features as labels may not reproduce human rule violation judgments on the same data. The figure contains four columns: first column names the setting (meal), the second column shows an example meal image, third column shows how descriptive labels can be collected while the fourth how normative labels can be collected.
1
6
759























