Daily AI Newsletter: New AI21 model can handle more context than most models

In today’s email:

🖐️ Amazon’s palm-scanning service now lets you sign up from your phone
🤯 How We’ll Reach a 1 Trillion Transistor GPU
🔫 How Silicon Valley’s ‘Oppenheimer’ found lucrative trade in AI weapons
🧰 8 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

AI21 Labs Unveils Jamba: It can handle more context than most models

AI21 Labs has introduced Jamba, a new generative AI model that stands out for its large context window and efficient computation. Unlike traditional models that require significant computational resources for handling extensive contexts, Jamba efficiently processes up to 140,000 tokens on a single GPU. This model, supporting multiple languages, is designed to perform tasks comparable to those of OpenAI's ChatGPT and Google's Gemini but with greater efficiency due to its unique architecture.

Jamba utilizes a hybrid architecture combining transformers, known for their complex reasoning capabilities, and state space models (SSMs), which offer computational efficiency and are adept at processing long data sequences. This combination allows Jamba to deliver three times the throughput on long contexts compared to similar-sized transformer-based models. While transformers focus on the relevance of various input data to generate output, SSMs leverage attributes from older AI models to handle extensive sequences more effectively.

Despite Jamba's potential, it's released as a research tool under the Apache 2.0 license, indicating it's not yet ready for commercial use, particularly due to the absence of safeguards against generating harmful content. AI21 Labs emphasizes Jamba's role in showcasing the SSM architecture's capabilities and its future potential once further optimized. The company plans to release a more refined version of Jamba that addresses these limitations and enhances its applicability in various domains.

Google·org launches $20M generative AI accelerator program

Google·org is introducing the Google·org Accelerator: Generative AI program, investing $20 million in grants to assist 21 nonprofits, including Quill·org and the World Bank, in developing generative AI technology. These organizations will receive not just funding but also technical support, workshops, and mentoring over a six-month period. Additionally, Google staff will collaborate directly with selected nonprofits like Tarjimly, Benefits Data Trust, and mRelief to foster AI tools aimed at language translation for refugees, assistance for caseworkers, and streamlining the SNAP benefits application process in the U.S.

The initiative underscores the potential of generative AI to enhance productivity, creativity, and effectiveness in the social impact sector. According to a PwrdBy survey, a substantial majority of nonprofits align AI innovation with their missions, recognizing its capacity to simplify tasks and reduce costs. However, the adoption of AI in this sector is hampered by challenges related to cost, resources, and awareness, with many organizations not yet utilizing AI technology due to these barriers.

The landscape is gradually changing as more AI-focused startups emerge within the nonprofit sector. Fast Forward notes an increase in AI company applicants, reflecting a broader trend of nonprofit organizations globally embracing ethical AI development. Entities like AlgorithmWatch, JoyEducation, and Earth05 are contributing to this shift, focusing on ethical AI practices and leveraging technology to advance their respective causes.

Biden orders every US agency to appoint a chief AI officer

The White House has mandated every federal agency to appoint a chief AI officer to manage the risks and benefits of artificial intelligence, as part of a broader government-wide policy. These officers, possibly chosen from existing chief information, data, or technology officers, will oversee AI initiatives, ensuring they align with safety, security, and democratic values. They are tasked with rectifying any non-compliant AI uses by a set deadline, with a focus on minimizing impact on rights and safety.

Chief AI officers will be responsible for evaluating AI applications within their agencies, especially those affecting critical services and citizen rights. They will conduct risk assessments, ensure staff training, and maintain public transparency about AI's potential impacts. These officers will also lead a new Chief AI Officer Council to share best practices and innovations across the government, aiming to harness AI for public benefit and address significant societal challenges.

In their role, chief AI officers will engage with the public, seeking feedback and ensuring accountability for AI's impact on society. They have the authority to adjust opt-out options for AI-driven decisions to ensure fairness and avoid undue hardship on agencies. Their decisions will play a crucial role in shaping how the U.S. government employs AI to enhance service accessibility, tackle societal issues, and uphold democratic principles and human rights.

What's next for AI agentic workflows ft. Andrew Ng

Other stuff

How Silicon Valley’s ‘Oppenheimer’ found lucrative trade in AI weapons
Amazon’s palm-scanning service now lets you sign up from your phone
X’s Grok chatbot will soon get an upgraded model, Grok-1.5
Amazon Bets $150 Billion on Data Centers Required for AI Boom
AI hustlers stole women’s faces to put in ads. The law can’t help them.
How We’ll Reach a 1 Trillion Transistor GPU
Large language models use a surprisingly simple mechanism to retrieve some stored knowledge
A.I. Leaders Press Advantage With Congress as China Tensions Rise
How AI Reshapes Vocabulary: Unveiling the Most Used Terms Related to the Technology