Daily Digest: Regulation and Copyright

PLUS: Will Apple steal the throne of on device performance?

October 31, 2023

Sign up | Advertise | Ben’s Bites News
Daily Digest #274

Hello folks, here’s what we have today;

PICKS

  1. The US gov released an executive order about regulation and reporting standards for AI. The order requires developers of powerful AI systems to share safety testing results with the government before public release. Also AI.gov, the official website of the US government is primarily focusing on AI talent.🍿Our Summary (also below)

  2. Apple’s entering into AI with hardware. Apple has released new Macs powered by M3 chips, showing they are embracing AI through custom hardware. The M3 Max chip targets AI/ML developers with its powerful GPU stack. 🍿Our Summary (also below)

    Additonal read: Fashionably late, per usual journey of apple of INto AI.

  3. Artists lose first round of copyright infringement case. A federal judge has dismissed most claims in a copyright infringement lawsuit brought by artists against AI art generators Stability AI, Midjourney, and DeviantArt. One claim against Stability AI for direct infringement was allowed to proceed.🍿Our Summary (also below)

from our sponsor

Did you know that you can get per-hour access to H100s RIGHT NOW? FLOP Auction is the first open market for compute. We have 23,000 H100 GPUs and are accepting bids for projects of any size. Pricing is simple and transparent. Current rate is $1.58 per GPU hour. Clusters available from 32 - 1016 GPUs. Bid now at FLOPAUCTION.com

TOP TOOLS

  • Conveyor* - automated security questionnaire answering that actually works.

  • GenVid - Generate SaaS demo videos using AI.

  • Embeds - Playground for comparing embedding models on Wikipedia+book retrieval.

  • Implai - Rethinking descriptive camera in the age of AI.

  • LlamaIndex Chat - Customizable open-source app to create and share LLM chatbots over your data.

  • Misgif - Add your face to popular gifs for free.

View more →
*sponsored

WHO’S HIRING IN AI

  • Fixie - The conversational AI app platform.

  • Elicit - Speeding up research paper analysis.

  • contextSDK - Real-time user context to improve engagement.

  • Rewind - Your AI meeting note taker that's not a bot

NEWS
QUICK BITES

The US gov released an executive order about regulation and reporting standards for AI. The order requires developers of powerful AI systems to share safety testing results with the government before public release.

What is going on here?

The US government is taking steps to regulate AI safety, but the executive order lacks concrete enforcement mechanisms.

What does this mean?

The rough calculations from @nearcyan put the limits around:

  • any model trained with ~28M H100 hours, which is around $50M USD

  • any cluster with 10^20 FLOPs, which is around 50,000 H100s

The order also directs different agencies to develop new standards for AI safety testing and monitoring, securing critical infrastructure with AI, preventing fraud and ensuring fairness in criminal justice AI.

Bilawal shared a tldr of all the things different government agencies are supposed to do according to the order:

  • NIST will set standards for AI safety testing and DHS will establish an AI Safety Board to review critical infrastructure.

  • DOJ & civil rights offices will give guidance on avoiding algorithmic discrimination.

  • HHS will support privacy-preserving techniques and evaluate commercial data procurement.

  • DOL will develop principles to protect workers from AI harms.

  • NSF will fund research on privacy-preserving technologies.

  • State Dept & Commerce Dept will collaborate internationally on AI safety.

AI.gov, the official website of the US government is also focusing on developing and bringing AI talent to the US including streamlining visa process.

Why should I care?

This signals the U.S. government is taking AI safety seriously amid rapid generative AI advances but wants to collaborate with industry versus take a heavy-handed regulatory approach.

Still, the language around safety and fairness is vague, focused on developing "best practices" and "guidelines" instead of firm regulations. While it codifies some safety protocols, the order depends on voluntary compliance and lacks real enforcement power without further legislation.

QUICK BITES

Apple has released new Macs powered by M3 chips, showing they are embracing AI through custom hardware. The M3 Max chip targets AI/ML developers with its powerful GPU stack.

What is going on here?

Apple’s news M3 chips have new GPUs optimized for AI, up to 2.5x faster than M1.

What does this mean?

The new M3 chips have a new GPU architecture with a feature Apple calls dynamic caching. Dynamic caching allocates GPU memory adaptively in real-time, increasing the utilization of the GPU. The new neural engine is 60% faster, speeding up ML.

The M3 Max has a 16-core CPU and 40-core GPU supporting 128GB of unified memory. Apple claims that’ll allow developers to work with models with billions of parameters.

Apple is keeping pace with AI-focused offerings from rivals like Qualcomm. Qualcomm made a similar claim this month with their Snapdragon X Elite of being able to run a 13B model on-device.

Why should I care?

If you use graphics/ML apps, the new Macs will bring meaningful speed improvements. If you’re a developer, Apple's platform is now more compelling for generative AI projects needing local compute.

And there’s the obvious answer to why should you care: because it’s APPLE.

QUICK BITES

A federal judge has dismissed most claims in a copyright infringement lawsuit brought by artists against AI art generators Stability AI, Midjourney, and DeviantArt. One claim against Stability AI for direct infringement was allowed to proceed.

What is going on here?

Artists lose a copyright infringement case against image generators.

via The Hollywood Reporter

What does this mean?

The lawsuit alleged these companies used copyrighted images without permission to train their AI systems. The judge ruled the artists must clarify their legal theory and provide facts showing copies of their work exist in the AI systems. To quote the ruling: “I am not convinced that copyright claims based on a derivative theory can survive absent ‘substantial similarity’ type allegations.”

The dismissed claims can be refiled but artists must clearly explain how protected images exist inside the AI systems, and not just as data used for training. According to this ruling, the artists likely can't rely on rough similarities between their work and AI outputs.

Why should I care?

This ruling isn’t a surprise for most of the people who are taking a nuanced look at the lawsuits. Most of them conflate different parts of the technology. This lack of clarity in the claims makes it harder for the judiciary to evaluate them. To point to this specific ruling, there seems to be a need for identical responses by the AI systems to justify the claims.

If you’re an artist, you should focus on getting a sound theory behind your claims and if you’re a researcher, you should aim to establish (or deny) the direct relation between actual art pieces and the generations.

Ben’s Bites Insights

We have 2 databases that are updated daily which you can access by sharing Ben’s Bites using the link below;

  • All 10k+ links we’ve covered, easily filterable (1 referral)

  • 6k+ AI company funding rounds from Jan 2022, including investors, amounts, stage etc (3 referrals)

Reply

or to participate.