weekend ai reads for 2024-05-10

She pointed out that the likes of deepfakes and AI technology can be "highly valuable tools both artistically and commercially when under the control of the artist."

“I had to laugh when I watched this Runway one. There’s a bit more photorealism, but the people are walking forwards and then backwards, so it’s certainly not a believable scene.

“As an industry professional, my expectation is perfection. I am looking for realistic-quality video, and AI is probably never going to quite get there.

“At the end of the Sora video, the couple is having a conversation in a coffee shop, looking like they’re enjoying themselves. That would be a shot that we’d use to sell a commercial property space as an amenity nearby.

“It's a little bit of thinking about developing a private garden, of looking at our own datasets that are exclusive to Nike – so performance data from an athlete, from our laboratories, et cetera. And then kind of commingling that with some things from the public garden, but making sure that that's all contained within what we're training the model on.”

We present an automated text animation scheme, termed “Dynamic Typography,” which combines two challenging tasks. It deforms letters to convey semantic meaning and infuses them with vibrant movements based on user prompts.

Make the marketers and ad companies earn their money. Be original and real and you’ll strengthen whatever community you love.

12. Knowing when to stop is a form of talent.

But I’m increasingly of the opinion that sharing unreviewed content that has been artificially generated with other people is rude.

Slop is the ideal name for this anti-pattern.

Within the next two years, frontier models will cross capability thresholds that even many open source advocates will agree are dangerous to open source ex ante.

The assumption that most people make is that these models can answer questions or chat with you, but in reality all they can do is take some text you provide as input and guess what the next word (or more accurately, the next token) is going to be.

As such, we hypothesize that the reward modelling process may have leaked information about the correct reasoning chains for GSM8k even if the problems themselves did not ever appear in the dataset. Finally, we observe that the Llema [sic] models have both high log-likelihoods and minimal overfit. These models are open-sourced alongside their training data, and the authors report finding a very small number of GSM8k examples in the training corpus. Nevertheless, they also find (and our study supports) that these few instances do not lead to overfitting. The existence of these outliers suggests that overfitting on GSM8k is not purely due to data contamination, but rather may be through other indirect means, such as model builders collecting data similar in nature to benchmarks as training data or selecting final model checkpoints based on performance on benchmarks, even if the model itself may have not seen the GSM8k dataset at any point via training. Conversely, the reverse is also true: small amounts of data contamination do not necessarily lead to overfitting

35% of all K-12 educators experienced burnout daily

66% report working beyond contractual hours

92% of teachers using AI found it helpful in addressing pain points at work

Machine unlearning can be broadly described as removing the influences of training data from a trained model. At its core, unlearning on a target model seeks to produce an unlearned model that is equivalent to—or at least “behaves like”—a retrained model that is trained on the same data of target model, minus the information to be unlearned.

In this video, we'll discuss how to create AI agents that interact with the YouTube Data API to extract comments from any given video and generate actionable insights. Based on user feedback, these agents can help you understand and create better content.