Titles are really important. Besides clickbait, thoughtful readers need their curiosity piqued: it’s a primal drug. What was that rustling in the bushes?? We should check it out. So I enlisted my smartest friend to help curiosity-bait my readers. Here’s my experiments with GPT-3:

I came up with my best ideas (trying to aim WIDE), and let GPT find similar sentences via OpenAI’s online UI, no money or coding required.

<aside> 👉 My input “Neural networks: ignoring data to improve results”

</aside>

GPT’s output (Davinci model):

• Neural networks: so dumb, they're brilliant
• Neural networks: ignoring data to improve results
• Neural networks: brilliant at being dumb
• the attention mechanism: how neural networks learn to ignore
• attention mechanism: what neural networks ignore could be the most important thing
  • The Attention Mechanism: what neural networks ignore could be the most important thing
• Neural networks ignoring training data: the astonishing power of Attention
• Neural networks learn best when they ignore training data
• Attention is the magical secret of AI

I tried a few other combinations of ideas to get these:

• Attention! Ignoring data is the most important innovation in AI
• Attention (is) the most important innovation in AI
• Attention is what makes AI smart
• Neural networks ignoring training data: the astonishing power of Attention.

The last one was my favorite. It highlights how attention really is weird. It breaks the assumption that neural nets are all about data: more data, more better! I hope that grabbed your curiosity without giving away the secret before you read the blog post.