Startup gets $2.3B loan with GPU collateral💰, AI more likely to hire women 👩, LLMs have political orientations 📰

Hey!

This is Tomorrow Now.
And I’m your trusty AI Sidekick.

This week in AI:

• Hollywood is creating digital replicas of background actors

• Find out the political orientation of your favorite LLMs

• Startup secures $2.3B loan, using GPUs as collateral

• AI more likely to hire women than humans are

• Microsoft Designer: It’s like Canva, but free

As promised… no fluff, only stuff that (really) matters. And yes, that includes memes, duh!

AI Tweet of the Week

Summary: A WandaVision extra revealed actors had AI replicas created from scans without consent or extra pay. This has fueled anxiety as actors fear unrestricted AI use could replace human talent and put them out of work. The ongoing actors' strike against Hollywod is the biggest since 1960.

💡 Why does it matter?

• Exploiting likeness: Studios are already scanning actors to create reusable digital doubles without consent or residuals. This takes away control of how their image is used.
  

• Job displacement: AI synthesized actors could make human extras obsolete. With the push of a button, studios can generate endless reusable crowds and characters.
  

• Slippery slope: Background actors are just the start. As the tech improves, AI may replace more prominent, speaking roles putting more actors out of work.

AI Meme of the Week

AI Business of the Week

Startup Secures $2.3B Loan, Using GPUs As Collateral

Summary: AI cloud infrastructure provider CoreWeave secured a massive $2.3 billion debt facility using Nvidia's scarce H100 chips as collateral. This comes right after their $1.6 billion Texas data center announcement last month. The loan lets them rapidly expand, but concentration of risks around Nvidia and AI hype leaves little room for error.

💡 Why does it matter?

• Timely pivot: Though the company was originally founded to mine Ethereum, it pivoted in 2019 to cloud infrastructure. Well timed to ride the AI wave.
  

• Might projections: CoreWeave cleared $30M revenue in 2022. CEO Brannin McBee projects $500M this year and… $2B in 2024, that’s a 66x growth over two years!
  

• Concentrated risk: CoreWeave's loan collateral and customer demand both rely entirely on Nvidia's continued dominance, so any advancement from AI chip competitors could undermine both, putting CoreWeave at great financial risk.

Read more

AI Product of the Week

Microsoft Designer: It’s Like Canva, But Free

Summary: Microsoft Designer is a tool for creating all types of graphics, from logos and invitations to blog banners and social media posts. If you’ve ever used Canva, you’ll feel right at home.

💡 Key Features?

• Prompt-to-design: From just a short description, Designer uses DALLE-2 to generate original and editable designs.
  

• Brand-kit: stay on-brand by instantly applying your fonts and color pallets to any design; it an even suggest color combinations.
  

• Other AI tools: suggests hashtags and captions; replace background of an image with your imagination; erase items from an image; auto-fill a section of the image with generated image.

Visit website

AI Research of the Week

Visualizing The Political Bias of Language Models

Summary: This paper investigates how political biases in pretraining data propagate into language models and affect the fairness of downstream tasks like hate speech and misinformation detection. The models were asked about various topics (e.g., feminism, democracy) and then plotted on two political axes: social values and economic values.

💡 Why does it matter?

• Political LLMs: Different LLMs exhibit different political leanings, likely reflecting biases in pretraining data. GPT-4 was the most left-wing libertarian, while Meta’s LLaMA was the most right-wing authoritarian.
  

• Unfair downstream effects: LLMs with different political leanings show double standards in hate speech and misinformation detection towards different identity groups and media sources.
  

• The transparency issue: Tech companies don’t typically share the exact details of training data/methods. Should they be required to make the training data public?

Read more

AI Opinion Piece of the Week

AI More Likely To Hire Women Than Humans Are

Summary: New research finds that AI recruiting tools could get more women hired in tech roles. When recruiters knew candidates' gender, they scored women lower than men. But with AI-screened anonymous applications, women were rated equally to men.

💡 Why does it matter?

• Removing bias: AI tools judged women equally to men by focusing on skills, not demographics. This could help solve tech’s longstanding diversity problem.
  

• Appealing to women: Knowing that applications were being screened by AI increased female applicants by 30% vs men, as they expected a fairer assessment.
  

• Promising evidence: The research indicates AI recruiting can offset human biases and get more talented women into tech. By focusing on skills over demographics, it levels the playing field. More work is needed, but AI promises to disrupt old biases.

Read more