AI Roundup 081: Creative differences

August 23, 2024.

Aug 23, 2024

∙ Paid

Ditching Discord

If you love Midjourney but hate Discord, you can now use the web version for free!

And if you want a crash course on Midjourney’s features, prompts, and secrets, check out the upcoming Midjourney Masterclass!

Join The Masterclass

Creative differences

Popular illustration app Procreate took a public stance against AI, pledging not to incorporate generative AI into the product.

Why it matters:

The announcement was praised by creatives, who have been fighting against the unauthorized use of their works to train models.
This week alone, two new lawsuits dropped: writers and journalists sued Anthropic and a YouTube creator filed suit against Nvidia.
Ultimately, there's a growing disconnect between techies excited about the prospect of AI and everyone else who's deeply skeptical - if not outright hostile - to this new technology.

Elsewhere in AI anxiety:

Arize AI reports a significant increase in Fortune 500 companies citing AI as a risk factor in their annual reports, with 86% of software firms expressing concerns.
Professors grapple with widespread AI cheating as many students submit ChatGPT-generated answers regardless of the assignment.
And South Korea's plan to introduce AI-enabled tablets in schools faces backlash from parents and academics worried about children's overexposure to devices.

SB 1047

California's major AI bill, SB 1074, advanced to the floor this week, drawing even more opposition from expected and surprising sources.

Between the lines:

In an unusual move, several Democratic members of Congress, including Nancy Pelosi, have urged Gavin Newsom to veto the bill.
Both OpenAI and Anthropic have opposed the bill, though the latter believes that with new amendments, the pros outweigh the cons.
But even with recent amendments, it seems that very few technologists actually support the bill, which doesn't bode well for its potential impacts.

Elsewhere in AI politics:

Lingo Telecom agreed to pay a $1M fine for making fake AI-generated robocalls imitating President Biden in New Hampshire.
Despite numerous AI product offerings, most US political campaigns remain wary and are largely not adopting these technologies.
And Donald Trump shared AI-generated images on Truth Social falsely depicting Taylor Swift and her fans supporting his campaign.

From Gods to gadgets

As we wait for the next generation of frontier models (GPT-5, anyone?), it's becoming harder to ignore the fact that there are more speculative rumors and fewer state-of-the-art models.

Why it matters:

The excellent
AI Snake Oil
put a finer point on this: AI labs are pivoting from "creating gods" to building consumer products, especially from Big Tech.
New releases (like Phi-3.5) have focused on speed, cost, and multimodality, which implies that the low-hanging fruit of "reasoning" has already been picked.
To be clear, I still think LLMs are incredibly useful, even if their capabilities plateau - but I do wish there was less of “GPT-5 will be incredible. Just trust us.”

Elsewhere in the FAANG free-for-all:

Nearly 200 workers at Google DeepMind, about 5% of its headcount, signed a letter calling to drop its military contracts.
Microsoft, Amazon, and others plan to turn old power stations and industrial sites into data centers, as land and power in many markets are heavily constrained.
And according to sources, Meta's Llama models have struggled to gain traction on AWS.