Hello, Niuralogists!‍

We hope you're having a great week! Once again, we're here to deliver the latest and most noteworthy developments from the dynamic realm of AI. The pace of innovation is truly remarkable, and we're dedicated to keeping you well-informed. In this week's update, explore Amazon's AI that transforms your hand into a magical tool, Baidu's democratization of ERNIE Bot generative AI, Google's unveiling of significant AI advancements, OpenAI's introduction of fine-tuning capabilities for GPT-3.5 Turbo and GPT-4, and Meta's reveal of the SeamlessM4T multimodal translation model.

‍

Amazon's AI Transforms Your Hand into a Magical Tool

Amazon's AI has worked wonders, turning your hand into a magical tool. Amazon One has accomplished this by training a neural network with millions of artificial palm images, resulting in a highly accurate and contactless payment and identity verification system, surpassing the precision of iris scanning. This innovative technology utilizes the distinctive patterns found in the lines, grooves, veins, and ridges of your palm to link a unique signature to your credit card or Amazon account. Currently rolling out in 500 Whole Foods Market stores and various other locations, Amazon One aims to render wallets and phones unnecessary. Its training involved an extensive collection of synthetic hand images, boasting an astounding 99.99% accuracy rate after over 3 million uses. Its applications range from touchless payments and age verification to venue access and loyalty rewards tracking.

‍

Baidu makes ERNIE Bot generative AI accessible to the public

Baidu, the Chinese tech giant, has officially made its generative AI offering, ERNIE Bot, accessible to the public through various app stores and its official website. ERNIE Bot possesses the remarkable capability to generate text, images, and videos based on natural language inputs, harnessing the power of the ERNIE (Enhanced Representation through Knowledge Integration) deep learning model. ERNIE, originally introduced and open-sourced in 2019 by researchers at Tsinghua University, exemplifies the potential of seamlessly combining textual data with knowledge graph information, demonstrating its impressive natural language understanding capabilities.

‍

‍

Google Unveils Significant Advancements in AI

During the commencement of Google's three-day Google Next conference in San Francisco, the tech giant introduced a plethora of new AI tools and capabilities. These notable developments encompass the addition of 20 new prebuilt AI models optimized for enterprises within Google's Cloud service, granting customers access to third-party systems such as LLaMa-2 (Meta) and Claude 2 (Anthropic). Furthermore, Google unveiled SynthID, a novel AI watermarking tool designed to discern AI-generated images while combatting disinformation and deepfakes. Cloud customers can now enjoy open access to the next-gen TPU v5e AI training chips, specially optimized for generative AI and LLMs. Google's Duet AI assistant is officially making its way to major Workspace apps, available at a price of $30 per month per user for enterprise users. Additionally, updates to Google's cloud-based Vertex AI platform include enhancements to PaLM 2, improved code generation, and the introduction of new search and conversational models. Notably, Google has also forged new AI partnerships with companies such as General Motors, FOX Sports, Estee Lauder, and Ginkgo Bioworks.

‍

OpenAI Introduces Fine-tuning Capabilities for GPT-3.5 Turbo and GPT-4

‍OpenAI has announced a significant advancement in its language models, which includes both GPT-3.5 Turbo and GPT-4. This development introduces the capability for fine-tuning, allowing developers to customize these models for specific applications and deploy them at scale. The goal is to bridge the gap between AI capabilities and real-world use, ushering in an era of highly specialized AI interactions. Early tests have shown promising results, with a fine-tuned version of GPT-3.5 Turbo surpassing the capabilities of the base GPT-4 for certain narrow tasks. Importantly, all data used in fine-tuning remains the property of the customer, ensuring data security and confidentiality. This advancement has garnered significant interest among developers and businesses, meeting the growing demand for customized models to create unique user experiences.

‍

‍

Meta Unveils SeamlessM4T Multimodal Translation Model

Meta researchers have unveiled SeamlessM4T, an advanced multilingual and multitask model that revolutionizes translation and transcription across both speech and text. In an age dominated by the internet, mobile devices, and global communication platforms, access to multilingual content has reached unprecedented levels. SeamlessM4T embodies the vision of effortless communication and comprehension across languages, offering automatic speech recognition for nearly 100 languages, speech-to-text translation for nearly 100 input and output languages, speech-to-speech translation for nearly 100 input languages and 35 output languages (including English), text-to-text translation for almost 100 languages, and text-to-speech translation for nearly 100 input languages and 35 output languages (including English).

‍

‍

‍