• Mindstream
  • Posts
  • Anthropic's initiative for AI safety revealed!

Anthropic's initiative for AI safety revealed!

+ Meta updates its AI label

Mindstream Banner


Among the planets in our solar system, Earth is not named after a Greek or Roman deity. Its name comes from the Old English and Germanic word 'eorthe,' meaning ground or soil.

Love or hate this intro? Hit reply and tell us why!

What’s in store:

  • Anthropic sets out a game-changing plan for AI benchmarks.

  • Photoshop Metadata causes AI tagging chaos.

  • Around the web: Tesla's Q2 report shows a 14% drop in production and a 4.76% drop in deliveries from last year.

  • Draft legal documents quickly with ai.law’s advanced technology.

Read Time: 5 minutes


Anthropic has launched a program to fund the creation of benchmarks for evaluating AI models, including their own, Claude.

The aim is to upgrade AI safety and performance metrics.

The program will pay third-party organisations to develop these benchmarks, accepting applications on a rolling basis.

Anthropic states, "Our investment aims to elevate AI safety, offering valuable tools for the entire ecosystem. High-quality, safety-relevant evaluations are in short supply."

Current AI benchmarks often fail to reflect real-world usage and may be outdated.

Anthropic's solution is to create benchmarks focusing on AI security and societal impacts through new tools and methods.

They specifically seek tests for tasks like cyberattacks, misinformation, and bias mitigation, and plan to develop an early warning system for national security risks.

Here are the latest insights:

  • Current AI benchmarks often fail to reflect real-world usage and may be outdated.

  • Anthropic seeks benchmarks focusing on AI security and societal impacts, such as cyberattacks, misinformation, and bias mitigation.

  • The program will support research and development of new evaluation platforms and large-scale model trials.

Redefining AI safety

Anthropic envisions new platforms for experts to develop their own evaluations and conduct large-scale trials.

They’ve hired a full-time coordinator and may expand promising projects.

While Anthropic's effort is commendable, trust may be an issue given their commercial ambitions.

They want funded evaluations to align with their AI safety classifications, which may force applicants to accept definitions they might not agree with.

Anthropic hopes the program will "catalyse progress towards comprehensive AI evaluation."

It remains to be seen if open efforts will collaborate with a commercial AI vendor.

Feels like AI will be tackling world-ending scenarios before breakfast.

Do you think Anthropic’s new benchmarking program will truly elevate AI safety standards?

Login or Subscribe to participate in polls.

Vote for live results and see results + opinions from yesterday at the bottom of the email.


AI-Powered Lawsuits Have Arrived

A law firm recently announced that it had filed the first lawsuit completely drafted by AI.Law.

The lawsuit included 3 counts, spanned 9 pages, and took 3 minutes for AI.Law to draft.

The same lawsuit would have taken 8+ hours to draft by hand.

AI.Law makes litigation drafting (complaints, answers, discovery, contracts) less expensive by orders of magnitude.

With pricing starting at 49 bucks, don’t expect the billable hour to last much longer.

On This Day

July 3rd 1886 - The New York Tribune became the first newspaper to use a linotype machine, revolutionising the printing industry by increasing the speed at which newspapers could be printed.

Tool of the Week


This AI-powered tool streamlines the creation of step-by-step guides and interactive demos, enhancing user onboarding, reducing support queries, and boosting conversion rates. It allows users to easily create and customise guides with annotations, and images and it supports multiple languages, provides voiceovers, and offers SEO-friendly options for better discoverability.


Meta is changing the "Made with AI" label to "AI info" on its apps after users complained about incorrect tagging.

Former White House photographer Pete Souza noticed the tag on an old photo he uploaded, likely triggered by using Adobe's tools.

Meta's Kate McLaughlin stated, "We're consistently improving our AI products and working with industry partners on AI labelling."

The new label aims to indicate content that may be modified by AI, not entirely AI-generated.

Here’s what you should know:

  • The issue is caused by how tools like Adobe Photoshop apply metadata.

  • The new label "AI info" will replace "Made with AI."

  • The new labels aim to clarify content modified by AI, not fully generated by it.

AI label confusion?

The issue stems from how metadata is applied by tools like Adobe Photoshop and how platforms interpret it.

As Meta's policies evolved, real-life images on Instagram, Facebook, and Threads were mistakenly tagged "Made with AI."

The new labelling will start appearing on mobile apps first, followed by web view.

Clicking the tag still provides a detailed explanation, covering images fully generated by AI or edited with AI tools like Generative Fill.

Metadata tagging tech like C2PA was supposed to simplify identifying AI-generated from real images, but that technology isn't fully developed yet.

Simplifying the digital world, one tag at a time.

Mindstream Picks

Tesla's global second-quarter deliveries dropped 4% year-over-year but exceeded analyst expectations, causing TSLA shares to surge early Tuesday.

Cosmic Shielding Corp. supports Aethero's radiation-hardened edge computer launch on July 10 via SpaceX's Falcon-9, featuring Nvidia’s Jetson Orin NX GPU.

Microsoft's Xbox Game Pass for July 2024 features new titles, including the acclaimed puzzle adventure The Case of the Golden Idol and Neon White, a fast-paced demon-clearing game, both announced via Xbox Wire.

Germany’s Bitcoin mining company Northern Data is reportedly in talks for a U.S. IPO next year, considering an offering for its AI cloud computing and data centre businesses.

Don’t Miss - Tesla's latest quarterly report shows a drop in both production and deliveries compared to last year. Production fell 14% to 410,831 vehicles, while deliveries decreased 4.76% to 443,956 vehicles. The company faces challenges with demand, production issues, and increasing competition from other automakers.

AI Art

Our Image of the Day

Artwork submitted by Mindstream reader Jeff A: “A Roman gladiator training with a robot sparring partner”

Daily Image Prompt

A group of cavemen discovering a futuristic tablet in a cave

Yesterday’s Poll

“How do you feel about Apple's latest AI partnerships?”

Excited and optimistic - 68%

Skeptical and unimpressed - 32%

Your Views:

“Apple consistently delivers exceptional technologically advanced and user friendly products.” - jimyarb

“Wake me when it’s over?” - rgray

Submit your opinions in our daily poll to be featured!

Collage of Authors

Midweek check-in! How are we doing so far?

❤️ We need your feedback to make our newsletter better.

📣 Refer our newsletter to your friends and make us feel all warm and fuzzy.

🚀 Advertise in our newsletter to reach 150,000+ founders, engineers, and content creators.

So, How Did We Do?

Login or Subscribe to participate in polls.