French startup Mistral AI's Mistral 7B model outperforms competitors, offering powerful language capabilities and open-source availability under Apache 2.0 license. With small-scale models targeting text summarization, classification, and code completion, Mistral aims to provide high-performance AI for businesses, with plans for a larger model in 2024.

Understanding AI can lead to increased trust in the process. Apple and OpenAI maintain secrecy around their AI practices. Users show more confidence in AI systems when they don't fully grasp the workings. Trust in black box models is evident, emphasizing the importance of meeting customer needs over selling technological gimmicks.

Be My Eyes launches Be My AI feature, utilizing OpenAI's GPT-4V for detailed image descriptions to aid visually impaired users. Available for testing to iOS users, enhancing independence with real-time voice descriptions, follow-up questions, and written responses in 29 languages. AI accuracy improving with feedback from blind community. Android version in testing, aiming to make visual information more accessible while emphasizing human volunteer support.

AWS launched five generative AI innovations to transform businesses. Key features include Amazon Bedrock for scalable computing and Amazon Titan Embeddings for language models. Meta's Llama 2 and Amazon CodeWhisperer also offer advanced AI solutions. Additionally, Amazon QuickSight enables easy BI visualization.

Tracking and ensuring B2B payment integrity is challenging. AI startup Slope uses its technology and GPT-3.5 platform to streamline B2B payment processes. Slope validated with $30M funding led by Union Square Ventures. It aims to enhance B2B payment visibility and risk assessment for cleaner data and fraud prevention.

AlphaSense, an AI-driven market intelligence platform, raised $150M in Series E funding led by BOND. The funds will enhance AI search capabilities and expand business content library for over 4,000 clients worldwide.

Getty Images launches Generative AI tool, using AI to generate images from extensive stock content with commercial use exemptions. Contributors compensated for training. Adobe also addresses this issue with Firefly AI model.

Spotify introduces AI voice translation for podcasts, using podcaster's voice to translate into multiple languages. Initial partners include Dax Shepard and plans for French and German translations. OpenAI's Whisper facilitates speech transcription.

Amazon invests $4B in AI company Anthropic to enhance Alexa with generative AI. Anthropic to use Amazon's chips & AWS cloud. Competition intensifies with Google in AI chip & chatbot fields.

Computer vision systems have bias issues, especially with images of black or brown people and women. Sony and Meta propose new methods to measure bias and improve diversity in AI training data, expanding the skin color scale to two dimensions. Meta also introduces FACET for fairness evaluation in computer vision. Annotated data is now available for free online.

NVIDIA is projected to surpass $50 billion in revenue, aiming for a $1 trillion market value in 2023, dominating the AI field. Revenue growth is driven by high demand for NVIDIA processors, boosting Q3 estimated revenue to $16 billion. NVIDIA's market value soared, reaching $1 trillion alongside tech giants. Strategic partnerships with Infosys, Reliance, and Google highlight NVIDIA's commitment to innovation and growth amid challenges like trade restrictions.

German tech giant SAP unveiled new AI assistant "Joule" to enhance enterprise outcomes. Utilizing advanced technologies from various vendors, Joule offers multilingual support and solution recommendations. Integrated with SAP applications, Joule maintains contextual awareness and safeguards against biased responses. Initially rolling out in SAP SuccessFactors and expanding to other cloud solutions, Joule reflects SAP's commitment to fostering a strong enterprise AI ecosystem.

NExT-GPT, a multimodal large language model developed by NUS, supports text, images, videos, and audio in an open-source model. The system utilizes advanced training techniques to build AI agents capable of universal modalities, demonstrating promising possibilities for human-like AI research.

Researchers from the University of Oxford discuss the potential impact of generative AI in driving technology advancements. They highlight opportunities in silicon materials, infrastructure, models, software, and services, but also note challenges related to bias, inequality, and job disruption. The report emphasizes the need for regulatory frameworks to ensure sustainable benefits.

The issue of Deepfake, generated by AI, lacks solid legislation for protection. Current laws are weak. US states vary in penalizing non-consensual Deepfake. EU AI Act focuses on risk classification, not Deepfake. China and Singapore have regulations. Global consensus needed for effective regulation.

The Santa Fe Institute researchers propose testing tasks to evaluate true intelligence in AI models like GPT-4, finding gaps in visual abstract reasoning compared to humans. Developing specific evaluation datasets is crucial as AI algorithms advance.

Meta is set to launch a generative AI chatbot called "Gen AI Personas" targeting young users with engaging personalities. The company aims to introduce multiple roles, including characters inspired by popular culture. These chatbots will offer diverse behaviors to attract and retain young users, potentially leading to increased advertising opportunities.

Meta should accelerate LlaMa 3 release to compete with Falcon. Departures due to internal competition for resources within Meta. Falcon 180b emerges as a strong open-source language model contender. Expectations and predictions surround LLaMa 3's release date and capabilities. Small projects rely on Meta's LLaMa-based language models. Concerns over Falcon's license terms. Meta's delayed LLaMa 3 launch disappoints open-source community.

AI tools like Copilot and DeepCode assist developers with code completion, debugging, and testing. They excel at generating repetitive code, writing documentation, test cases, and analyzing data. AI empowers developers by enhancing their capabilities, but may not completely replace them.

Meta AI researchers apply LLMs to code optimization tasks, achieving a 3.0% reduction in code size with remarkable accuracy. Their tailored LLM handles compiler optimization well, outperforming current machine learning methods. They provide experiments and code examples to showcase LLM potential and limitations, expecting widespread interest.