🗞 OpenAI Likely to Favor Incremental GPT-4.5 Over Ambitious GPT-5 This Year

🔎 The Latest on the AI Frontier:

• OpenAI Likely to Release GPT-4.5 This Year Instead of GPT-5 🤖

• Adobe Enhances Photoshop with Generative AI Tools Powered by Firefly 🎨

• Meta Integrates AI Chatbots Across Facebook, Instagram, and WhatsApp ⭕️

• Microsoft Unveils Lightweight AI Model Phi-3 for Efficient Performance 🧐

• AWS Aims to Host and Customize Companies' Generative AI Models 🌪

• Microsoft AI Chief Views AI as a New Digital Species 🧬

• Study Shows AI Can Predict Political Orientation from Facial Features 🚨

• Vietnam's FPT to Invest $200M in AI Factory with Nvidia 💰

🤖 OpenAI is more likely to release GPT-4.5 this year instead of GPT-5. Read More

• According to experts, GPT-5 would require around 100x the computational resources of GPT-4, equivalent to running 1 million H100 chips for 3 months.

• Analysts suggest GPT-4.5 aligns with a $1 billion training cost, a 10x increase from GPT-4, as noted by Anthropic's CEO.

• While competitors like Meta's Llama 3 are catching up, GPT-5 may be delayed due to the need for a revised training curriculum beyond ingesting human conversations.

🎨 Adobe introduces new generative AI tools in Photoshop, powered by its Firefly Image 3 model, making it easier to create and enhance designs. Read More

• The Reference Image tool allows users to guide AI generations by providing reference images, reducing the need for complex text prompts.

• Generate Background and Enhance Detail enable quick background replacement and detail enhancement for product photography and images.

• Adobe claims its Firefly model offers photorealistic quality and better comprehension of descriptive prompts, suitable for commercial use.

⭕️ Meta is integrating AI chatbots into its popular apps, allowing users to interact with AI within Facebook, Instagram, WhatsApp, and Messenger. Read More

• Meta's AI can comment on posts, join group messages, and provide suggestions in social feeds, aiming to keep users engaged within Meta's ecosystem.

• The AI can generate images, write messages, and answer queries by summarizing information from the web, potentially reducing the need to leave Meta's apps.

• Critics argue that this move could diminish the open web, making data harder to access and creators less incentivized to produce or share content.

🧐 Microsoft has launched Phi-3, its smallest AI model yet, designed for efficient performance on personal devices and targeted applications. Read More

• Phi-3 Mini has 3.8 billion parameters, trained on a curated dataset inspired by children's bedtime stories to build general knowledge.

• With performance close to models 10 times its size, Phi-3 Mini is touted to be as capable as GPT-3.5 in a smaller form factor.

• Lightweight models like Phi-3 are cheaper to run and better suited for custom applications, addressing the need for more affordable and efficient AI solutions.

🌪 AWS aims to become the go-to platform for hosting and fine-tuning custom generative AI models with its new Custom Model Import feature. Read More

• Custom Model Import allows companies to import their proprietary generative AI models into AWS Bedrock and access tools for expanding knowledge, fine-tuning, and implementing safeguards.

• AWS claims Bedrock offers unmatched breadth and depth of model customization options, including Guardrails for filtering problematic outputs and Model Evaluation for testing performance.

• AWS also announced the general availability of its Titan Image Generator, with improved creativity, and an upgraded Titan Text Embeddings V2 model for efficient text embedding.

🧬 Microsoft's AI chief views AI as a new digital species, representing the fastest and most consequential wave of creation ever. Read More

• Mustafa Suleyman emphasized the need to find appropriate analogies for AI to prioritize safety and ensure it serves humanity.

• He envisions a future where everything, from people to businesses, will have an interactive "personal AI" persona that is infinitely knowledgeable and reliable.

• While acknowledging the potential benefits of AI, Suleyman urged confronting potential dark scenarios to mitigate unintended consequences in the next 5-10 years.

🚨 A study found that AI can accurately predict a person's political orientation from images of their expressionless faces, raising privacy concerns. Read More

• The study analyzed 591 participants' facial features using an AI algorithm and successfully mapped them to their self-reported political views.

• Researchers warn that this ability of facial recognition technology poses "serious challenges to privacy" and is "more threatening than previously thought."

• The findings underscore the urgency for addressing the potential risks of biometric surveillance technologies that can reveal intimate traits without an individual's control or consent.

💰 Vietnam's leading tech company FPT plans to invest $200 million in an AI factory using Nvidia's chips and software. Read More

• FPT aims to boost AI research in Vietnam and develop AI applications, including generative AI and autonomous driving, at the planned data center factory.

• The $200 million investment is part of FPT's vision to turn Vietnam into an AI hub, through collaboration with Nvidia in technology, business development, and training.

• Nvidia and FPT signed a comprehensive strategic partnership, with Nvidia supporting FPT's plans to establish AI factories in Vietnam and other potential markets like Japan and South Korea.

Have any feedback? Send us an email