Good Morning! Google's rethinking its cookie strategy for Chrome, they want to retain their third-party support. Bitwise launches its Ethereum ETF with a twist, giving 10% of profits to the devs. Meta steps up its AI game with Llama 3.1, boosting capabilities that rival Claude and GPT.

— Forrest Knight & Meghanadh Vasireddy

Google Abandons Plans for Third-Party Cookies in Chrome

Google has decided not to remove third-party cookies from Chrome as originally planned. This shift comes after years of working on the Privacy Sandbox project, which aimed to create new ways to handle online ads while protecting user privacy.

Instead of getting rid of cookies, Google will:

• Keep supporting third-party cookies in Chrome

• Create a new system for users to choose their privacy settings

• Let Privacy Sandbox tools work alongside existing cookies

Why the change?: Google faced pressure from regulators and pushback from the ad industry. The company also found that moving away from cookies was more complex than expected.

Testing and Performance: Google's tests showed that Privacy Sandbox tools could still help ads work well without cookies. However, the transition would require a lot of work from many different parties in the online ad world.

What's Next:

• Google will keep improving Privacy Sandbox technologies

• New privacy features will be added, like hiding IP addresses in private browsing

• The company is working with regulators, including the UK's Competition and Markets Authority, to finalize this new approach

Read More Here

Bitwise will Donate 10% of Ethereum ETF Profits to Developers

Bitwise has just launched its spot Ethereum ETF (ETHW) on the NYSE, marking a significant development in the crypto investment landscape. This new offering allows direct investment in ETH, providing exposure to the world's second-largest blockchain ecosystem without the complexities of self-custody.

What sets this ETF apart is Bitwise's commitment to the Ethereum ecosystem. They're pledging 10% of ETHW profits to support open-source development, specifically:

• Protocol Guild, which backs over 170 core Ethereum contributors

• PBS Foundation, funding research on block relay technology

On the investment side, ETHW boasts a competitive 0.20% management fee, waived for the first six months on the initial $500 million in assets. Coinbase Custody Trust Company has been tapped as the digital asset custodian, ensuring robust security measures.

Bitwise CTO Hong Kim emphasizes the symbiotic relationship between investors and developers: "Every ETHW investor has a stake in Ethereum's advancement, and our donation program directly contributes to this goal."

This launch follows the success of Bitwise's Bitcoin ETF (BITB), which accumulated $2.7 billion AUM in just six months. The firm anticipates ETHW will further catalyze crypto investment, potentially driving ETH to new all-time highs in 2024.

Read More Here

Meta Unleashes Llama 3.1

Meta has released Llama 3.1, its latest open source large language model that claims to rival top closed-source competitors like GPT-4 and Claude 3.5 Sonnet. The flagship 405 billion parameter version was trained on over 15 trillion tokens using 16,000 NVIDIA H100 GPUs, pushing Meta's training infrastructure to new limits.

Key technical improvements include:

• 128K token context window (up from 8K)

• Multilingual support across 8 languages

• Quantization from 16-bit to 8-bit for efficient inference

• Improved synthetic data generation and model distillation capabilities

The model uses a standard decoder-only transformer architecture, eschewing more complex mixture-of-experts approaches. Post-training involves iterative rounds of supervised fine-tuning and direct preference optimization.

Meta is also introducing the "Llama Stack" - standardized interfaces for building AI components and applications around Llama models. This aims to foster easier interoperability across the open source AI ecosystem.

Read More Here

🔥 More Notes

• The internal source code of WinUI is now available

• “Software engineers are not (and should not be) technicians”

• NVIDIA Transitions Fully Towards Open-Source GPU Kernel Modules

Youtube Spotlight

Training AI to Play Fortnite (Reinforcement Learning + Computer Vision)

Was this forwarded to you? Sign Up Here