Discover more from Artificial Ignorance
AI Roundup 031: Think of the children
September 8, 2023
Think of the children
In a letter sent Tuesday, attorneys general from all 50 states called on Congress to establish a commission to study how AI might be used to exploit children, and expand restrictions on CSAM (child sexual abuse materials) to specifically cover AI-generated images.
Why it matters:
The letter raises reasonable concerns about AI - it understands how deepfakes work and that new generative AI tools have made creating deepfakes drastically easier.
In regulating AI-generated imagery, we're also dealing with a thorny question around creating CSAM-adjacent material that isn't based on any specific person.
Ultimately, I'm pretty skeptical that Congress will move to regulate this space anytime soon, though CSAM is often an easy bipartisan topic.
The big picture:
We're used to seeing concerns over AI copyright and job displacement, including new efforts to ban Books3, a popular book-based training dataset.
In California, Governor Newsom is directing state agencies to examine AI threats while authorizing staff to experiment with AI tools.
And in Norway, the country's sovereign wealth fund has outlined its top three corporate AI concerns: board responsibility, transparency, and risk management.
Microsoft's defense fund
Microsoft committed to defending its AI Copilot customers from copyright infringement lawsuits. The Copilot Copyright Commitment will protect customers as long as they’ve "used the guardrails and content filters we have built into our products."
What to watch:
Who owns the copyright for AI-generated works is still a huge open question. The US Copyright Office is currently seeking public comment on specifically this issue.
Microsoft’s position is that "it does not claim any intellectual property rights in the outputs of its Copilot services", but it's hoping to assuage the fears of bigger enterprises.
I'm not sure this will change anyone's mind - small startups will use Copilot regardless, and conservative companies will probably wait a while before adopting (if ever).
Elsewhere in AI defense spending:
Microsoft researchers say Chinese hackers are using AI to generate content designed to go viral.
Palmer Luckey, former Oculus CEO turned defense contractor, is leveraging AI hype to sell autonomous military drones.
And interviews with AI "red team" leaders discuss the nuances of breaking AI models and the challenges of fixing them.
In a new report from The Information, sources say a majority of the original Llama researchers have quit Meta over feuds for compute resources and talent.
Between the lines:
Retaining top talent can be difficult with so much hype and investment pouring into AI. And with new incentives to launch and commercialize AI, organizations can be stretched to a breaking point.
All of the original Transformer paper authors have left Google for other research labs or to start billion-dollar companies. And the company bruised some egos when it forced Google Brain and Deepmind to merge in April.
Elsewhere in the FAANG free-for-all:
Apple reportedly has a budget of millions of dollars per day for training AI models.
In a positive move, Pinterest has tweaked its algorithms to surface diverse body types, after training an AI on billions of images of different-sized figures.
And Bloomberg looks at the AI chatbots offered by Chinese tech giants, which are now rolling out after government approval.
Anthropic launches a paid tier of Claude for $20/month. OpenAI DevDay is set for November 6th in San Francisco. The creator of the viral AI-Drake track is meeting with record labels and Grammy organizers. Slack's new AI can generate channel highlights and summarize threads. "I replaced 50 lines of code with a single LLM prompt." Google to require election ads to disclose AI-generated content. Anguilla, the owners of .ai domains, will make up to $30M in registration fees this year. The missing storage layer for AI. Sam Altman thinks Silicon Valley has lost its culture of innovation. Buzzy AI startup for generating 3D models used cheap human labor. Can LLMs learn from a single example? eBay's new AI tool can create entire product listings from a single photo. Meta's Yann LeCunn isn't buying the AI doomer narrative. The Falcon 180B model is now available. A Deepmind co-founder on how AI “threatens to reshape life as we know it.”
Artificial Ignorance is reader-supported. If you found this interesting or insightful, consider becoming a free or paid subscriber.