Mind Machine AI

India faces a critical shortage of agentic AI professionals, with demand projected to double by 2026. The current talent pool of under 100,000 struggles to meet the growing need, driving up salaries and competition among GCCs, IT firms, and startups. The market for AI agents is projected to grow significantly, driven by sectors like autonomous vehicles, smart manufacturing, and healthcare.

economictimes

Suna By Kortix AI

Adobe Firefly, which revolutionized the creative industry in under two years, has generated over 20 billion assets worldwide. Today at Adobe MAX London, they unveil ed the latest Firefly release, unifying AI-powered tools for image, video, audio, and vector generation into a single platform with new capabilities.

The latest image generation model, gpt-image-1, is now available in the API, enabling developers to integrate high-quality image generation into their tools and platforms. The model can create images in diverse styles, follow custom guidelines, and accurately render text.

OpenAI gpt-image-1

Two undergrads from Korea, with just three months of AI tinkering, have unleashed Dia—a 1.6B parameter open-source speech model that mimics podcast-style dialogues, rivaling Google’s NotebookLM. Built using Google’s TPU Research Cloud, Dia allows users to customize tones, insert natural speech disfluencies, and even clone voices, all operable on consumer-grade GPUs with 10GB VRAM. Available on Hugging Face and GitHub, it offers impressive performance, though it lacks safeguards against misuse. Nari Labs plans to expand Dia’s capabilities and language support, aiming to add a social layer atop their synthetic voice platform.

Nari Labs

A good look at the state of MCP by Charlie Graham. The current clients like Claude, cursor & VS Code seem crude but the future is bright. MCP’s are like “chat apps” that these clients run. MCP clients could dominate the future as the search engines do today. The gen1 apps lack security and results vary depending on llm interpretation of your query. But we are at gen 1 of this tech and in the past few weeks the count of MCP servers on MCP.so has exceeded 10k. If you want to dip your toes into AI app building MCP is where the action is.

mcps- gatekeepers and the future of AI

An amazing essay on building software for the AI Era. A lot of today’s software is akin to the “Horseless carriage” which refers to the early motor car designs that borrowed heavily from the horse-drawn carriages that preceded them. Peter Koomen brilliantly shows the folly of today’s AI app builders and suggest some creative ways of getting beyond the AI slop with how AI is being used in email apps.

AI Horseless Carriages

Interesting insight from @karpathy on X that sees the llm as your customer

Are we pivoting to a world where we depend on llm’s and products are built llm first ?

Building a magical AI code editor used by over 1m developers in 4 months: Inside Windsurf

Google has released new quantization-aware trained (QAT) versions of its Gemma 2B and 7B models, enabling state-of-the-art performance while running efficiently on consumer-grade GPUs. These models are designed to maintain accuracy even after 4-bit quantization, thanks to techniques like QLoRA and SmoothQuant. Notably, they outperform competitors like Mistral and LLaMA 2 across multiple benchmarks, and offer open weights for local deployment via platforms like Hugging Face and NVIDIA TensorRT-LLM. This positions Gemma QAT as a major leap toward democratizing high-performance AI inference for individual developers and small teams.

Google Developer Blog

Chungin “Roy” Lee, a 21-year-old former Columbia student, raised $5.3 million for his startup Cluely, which offers an AI tool to “cheat” on exams, sales calls, and job interviews. The tool, originally called Interview Coder, was developed by Lee and his co-founder, Neel Shanmugam, and led to their suspension from Columbia. Cluely’s manifesto compares the tool to inventions like the calculator and spellcheck, while a launch video featuring Lee using the AI assistant on a date sparked both praise and criticism.

techcrunch

ByteDance has unveiled Seaweed-7B, a 7-billion-parameter AI video generation model that delivers high-quality, real-time video with synchronized audio, rivaling larger models like OpenAI’s Sora and Google’s Veo at a fraction of the compute cost. Seaweed-7B supports text-to-video, image-to-video, and audio-driven synthesis, producing up to 60-second 720p videos at 24fps in real-time. It features advanced capabilities such as multi-shot storytelling, precise camera control, and lifelike human motion with synchronized lip-sync. The model’s efficiency stems from innovations like a 64× compression VAE, hybrid-flow Transformer architecture, and a progressive training strategy, reducing training costs by two-thirds compared to similar models. Seaweed-7B is currently in closed testing, with potential applications in AI filmmaking, education, gaming, and virtual assistants.