Amazon's NEW Chips & more from AWS re:Invent 2023

πŸ“° No-Noise News πŸ“°

🎫 Interesting from AWS re:Invent 2023 Day 1:

AWS introduced Graviton4 and Trainium2 chips, promising significant improvements in processing power, energy efficiency, and faster AI model training at reduced costs.

Amazon launched Q, an AI chatbot starting at $20 per user annually, intended to help AWS customers by providing AWS-related solutions and automating business tasks.

In 2024, Amazon will release new EC2 instances with the latest NVIDIA H200 GPUs, boosting performance in AI, graphics, and computing.

Amazon is expanding its palm-scanning tech to offices, allowing employees to use their palm for entry and data access. Some companies are already using it despite privacy concerns

AWS improved its transcription service by adding AI to support 100 languages, enhancing speech-to-text accuracy, and introducing new features for users.

βš™οΈ Abu Dhabi has launched AI company A171, which will compete with OpenAI and specialize in marketing and supporting the Falcon large language model (LLM). The company kicks off with three platforms catering to specific domains.

πŸ’° Dataminr, a leading big data company valued at $4.1 billion, is laying off 150 employees, about 20% of its workforce. The New York-based company, known for using AI and big data to predict global events, attributes this decision to economic pressures, a push for operational efficiency, and advancements in AI technology. Despite the layoffs, Dataminr aims to strengthen its financial position and advance its AI capabilities

🎞️ RunwayML has bolstered its AI video editing suite with features aimed at making it more user-friendly and expressive. Enhancements like Motion Brush allow users to create motion in images, while style presets simplify artistic direction. Director Mode provides advanced camera controls for more precise storytelling, and image model upgrades offer improved resolution and consistency. 

πŸ•΅πŸΏβ€β™€οΈ California introduces draft AI rules aimed at empowering state residents with opt-out rights and clear information on data usage, potentially affecting tech giants' advertising practices and privacy approaches.

πŸ—£οΈ AntiFake, designed to combat voice deepfakes, proactively distorts audio signals slightly so that while humans can understand, AI finds it hard to replicate for misuse.

πŸ“š Nerd section πŸ“š

πŸ–₯️ LMQL - SQL for Language Models: LMQL, created by ETH Zurich researchers, simplifies language model programming and can cut interaction costs by up to 85%

βš’οΈ NVIDIA released HelpSteer, an open-source dataset aimed at creating more helpful LLMs by refining their responses for accuracy, coherence, and varying complexity, benefiting industries like customer service and education.

🎞️ ChatGPT 3.5 Turbo Fine Tuning For Specific Tasks - Tutorial with Synthetic Data

🌐 Web section 🌐 

πŸ•ΈοΈ What's this new section about?

On days when there's a lot of news, we aim to keep our newsletter short and sweet.

So here (but not in the email version) we're adding AI-related stories that may be slightly less attention-grabbing, but still could move the needle in specific sectors and fields.

βš™οΈ Server shipments are expected to decline by up to 20% in 2023, yet Microsoft and Meta are outpacing other hyperscalers in their deployment, with both companies set to have received 150,000 of Nvidia's H100 accelerators.

🎭 The OpenAI saga continues as fired CEO Sam Altman returns along with company president Greg Brockman, and the board puts a new governance structure in place. However, an independent investigation will be conducted into Altman's firing, leaving the story with potential twists to come.

πŸ‘Ž With AI set to make major life decisions, inadequate training could result in unjust rejections for loans, housing, and jobs, and could lead to greater biases against minority groups.

πŸ§’ The GAIA benchmark evaluates whether AI chatbots can demonstrate human-like reasoning and competence on everyday tasks, with humans scoring 92% and GPT-4 with plugins scoring only 15%. The benchmark focuses on human-like competence, not expertise, and aims to push the field beyond narrow AI benchmarks.

πŸ’ Curtis Priem, co-founder of Nvidia, could have been worth $70 billion but chose philanthropy over wealth, donates to education, and writes about repairing the earth from a semi-off-the-grid lifestyle.

πŸ… Zapata AI has earned a second DARPA award for the Quantum Benchmarking program. The award will support the development of open-source tools to estimate the resources needed for industrial applications of quantum computing. Zapata AI is working with academic and technology partners to identify high-value applications and quantify their utility. The goal is to unlock the full potential of quantum computing for solving complex industrial problems.

🀑 And a funny moment - Google Cloud ran an advertisement on Vegas Sphere during the AWS Re:Invent

Thank you for reading today’s edition!

We love to hear back from you!

Feel free to reply to our emails with questions, suggestions, or topics you'd like to see covered, or drop us a message on Twitter or Facebook.

Until tomorrow,
- Ts (Bits and Neurons)