Here’s Why Over 4 Million Professionals Read Morning Brew

• Business news explained in plain English

• Straight facts, zero fluff, & plenty of puns

• 100% free

See for yourself

Today:

• AI Self-Improvement Accelerates Industry Transformation

• Perplexity AI Eyes TikTok Merger

• ChatGPT Gets Personality Customization Features

• Runway Frames Brings Cinematic AI

• First AI Chatbot Returns Decades Later

Researchers STUNNED As AI Improves ITSELF Towards Superintelligence "OpenAI have 'broken out'..."

AI progress has recently surged, with breakthroughs in reasoning models like GPT-4 enabling self-improvement through knowledge distillation. This cycle allows models to refine themselves iteratively, leading to smarter, cost-effective systems. Experts anticipate a rapid intelligence explosion, where AI could automate its research, accelerating advancements.

With OpenAI and others prioritizing superintelligence over short-term products, this transformative era could redefine industries, human capabilities, and our understanding of intelligence itself.

WATCH THE VIDEO ON YOUTUBE

Perplexity AI Eyes TikTok Merger

Perplexity AI has bid to merge with TikTok U.S., proposing a new entity with ByteDance investors retaining stakes. The deal could value TikTok above $50 billion, adding video to Perplexity’s AI-driven offerings. With TikTok facing potential U.S. bans, President-elect Trump may extend negotiations by 90 days. ByteDance prefers mergers over sales, boosting Perplexity's chances. The merger could reshape tech and AI search markets amid rising generative AI competition.

ChatGPT Gets Personality Customization Features

OpenAI has launched new customization options for ChatGPT, allowing users to assign traits like "Chatty" or "Gen Z" and add personal details for tailored interactions. This revamp enhances the interface of the custom instructions menu but relies on the same prompt engineering for personalization. The feature, available on ChatGPT.com and the Windows app, excludes EU countries initially. OpenAI aims to boost user experience as ChatGPT garners over 300 million weekly users.

Runway Frames Brings Cinematic AI

Runway has launched Frames, an advanced AI text-to-image generator praised for its cinematic visuals and stylistic control. Available through subscription, Frames offers creative pros 19 preset styles and tools for custom designs. It ensures safety with content moderation and invisible watermarks while addressing ethical concerns like bias and copyright. Frames aims to empower artists, designers, and filmmakers with a consistent, versatile tool for visual storytelling, potentially challenging rivals like Midjourney.

First AI Chatbot Returns Decades Later

The original code for ELIZA, the first AI chatbot created in the 1960s by Joseph Weizenbaum at MIT, has been rediscovered and restored. This early AI milestone, originally written in just 420 lines of code, has been run for the first time in decades. ELIZA’s resurrection offers a glimpse into the origins of artificial intelligence and highlights its stark simplicity compared to today’s advanced language models like ChatGPT.

🧠RESEARCH

Inference-Time Scaling for Diffusion Models beyond Scaling Denoising Steps

This study investigates how diffusion models can improve performance at inference time by optimizing noise selection during sampling. Through experiments on image generation benchmarks, the research demonstrates that increased computation enhances sample quality, with customizable components offering flexibility for various applications. The findings extend scaling insights beyond traditional denoising steps.

OmniThink: Expanding Knowledge Boundaries in Machine Writing through Thinking

OmniThink introduces a machine writing framework that mimics human cognitive processes, iteratively expanding and refining knowledge for richer, more original content. By addressing limitations of retrieval-augmented generation, OmniThink enhances knowledge density while maintaining coherence and depth. Experiments and expert reviews confirm its effectiveness in generating high-quality, informative long-form articles.

Towards Large Reasoning Models: A Survey of Reinforced Reasoning with Large Language Models

This survey explores advancements in using Large Language Models (LLMs) for complex reasoning tasks, highlighting techniques like reinforcement learning and test-time scaling to enhance reasoning accuracy. It reviews methods such as automated data construction and reasoning-based learning, examines open-source projects, and outlines challenges and future directions in developing large reasoning models.

FAST: Efficient Action Tokenization for Vision-Language-Action Models

The FAST framework introduces a novel tokenization method using discrete cosine transform for robot actions in vision-language-action models. It addresses challenges in high-frequency, dexterous tasks where traditional methods fail. With FAST+, a universal tokenizer trained on 1M trajectories, the approach significantly improves training efficiency and performance, reducing training time by up to 5x.

CaPa: Carve-n-Paint Synthesis for Efficient 4K Textured Mesh Generation

CaPa introduces an efficient carve-and-paint framework for high-quality 3D asset generation, addressing issues like multi-view inconsistency and low fidelity. Using a two-stage process—geometry generation and high-resolution texture synthesis—it ensures structural consistency and cohesive results. With outputs ready in under 30 seconds, CaPa sets a new benchmark for scalable, practical 3D generation.

🛠️TOP TOOLS

Vercel AI Playground - Interactive online platform that allows users to explore, experiment with, and compare various AI language models directly in their browser. 

Myreader AI - AI-powered reading assistant that transforms how users interact with digital content.

3DFY AI - AI-powered platform that transforms text descriptions or single images into high-quality 3D models.

Arvin AI - AI-powered browser extension that integrates advanced language models like GPT-4o, Claude 3.5, Gemini 1.5, and Llama 3.1 to enhance users’ online experiences.

Fadr - AI-powered music creation platform that combines advanced technology with user-friendly tools to enhance musical creativity.

📲SOCIAL MEDIA

🗞️MORE NEWS

• Cisco AI Defense safeguards AI adoption with end-to-end security. It identifies assets, detects risks, mitigates threats, and enforces controls. Backed by advanced threat intelligence, it protects AI applications across the lifecycle.

• Character AI introduces games like Speakeasy and War of Words to enhance user engagement. Targeting entertainment, the platform integrates safety tools while users spend an average of 98 minutes daily exploring AI-powered characters.

• AI search engine Perplexity acquires professional social platform Read.cv, enhancing its corporate features. Read.cv users can export data until May 16, while Perplexity focuses on building consumer and social experiences through strategic acquisitions.

• OpenAI's CEO, Sam Altman, announced the o3-mini model will launch in a few weeks, emphasizing its improved reasoning capabilities. The AI community debates its impact, particularly in business operations and hiring processes.

• Google aims to reach 500 million Gemini AI users by year’s end, leveraging a premium subscription model and Workspace integration. Despite lagging behind ChatGPT, Gemini offers expanded AI capabilities to drive growth in competitive markets.

• Meta has unveiled SEAMLESSM4T, a speech-to-speech AI translator handling 101 languages and delivering synthesized translations in 36 target languages. Built on extensive multilingual datasets, it promises equitable, open-source translation capabilities for global use.

• AI-powered "kissing apps" are trending on social media, creating deepfake videos of people kissing without consent. Critics warn these apps normalize exploitative deepfakes, raising ethical concerns about privacy, consent, and potential misuse.