Welcome back to EmergingTech. One story that caught our attention comes from India, where everyone seems to be talking about GenAI. But who's actually using it in the real world? A new report looked at 42 of India's biggest companies and found that while many have moved beyond experiments, only a handful are truly leading the AI race.

This Week in AI

LATEST IN AI

🧠 OpenAI Built an AI That Learns From Real Work

OpenAI and Thrive Holdings built a tax-preparation AI that doesn't just do the work, it learns from it. Using Codex, the system turns real-world accountant feedback into improvements, helping it get smarter over time without engineers manually fixing every issue.

Our take: Most AI tools are like interns who need constant supervision. That's a much more interesting direction than simply making models bigger.

Why it matters: The AI processed 7,000 tax returns, reached up to 97% accuracy, and saved accountants about a third of their preparation time. More importantly, it shows how future AI systems could continuously improve from real-world use, making them more useful long after deployment.

Leonardo is showing off a powerful new creative workflow where GPT Image 2 generates an entire storyboard complete with characters, camera angles, shot sequences, timing, and visual direction. Seedance 2.0 then takes that storyboard and turns it into a fully animated scene.

Turns out India's biggest companies are finally moving past the AI testing phase. Banks, telecom firms, and manufacturers are already putting AI to work, while many energy companies are still figuring things out.

The Interesting Part: A lot of companies talk about AI, but it is often hard to tell what is actually happening behind the scenes. This report stands out because it focuses on real world evidence like deployments, hiring, leadership roles, and business outcomes.

Why It Matters: Corporate India has seen plenty of tech trends come and go. But every once in a while, a technology changes how businesses actually operate. For many companies, AI is starting to look less like a buzzword and more like one of those moments.

🌦️ AI WEATHER FRONTIER

A new arXiv study explores how far AI weather models can predict beyond the typical 15 day range and what actually breaks when they are rolled out over long time periods. It focuses less on short term accuracy and more on long horizon behavior.

How they tested it: Through long rollouts and simulations, researchers identify three failure modes: blow up where outputs become unstable, drift where forecasts slowly diverge from reality, and loss of seasonality where models stop capturing recurring climate patterns.

Testing Setup: To evaluate performance, the study uses Forecasting the Weather Beyond Two Weeks Using NVIDIA Earth-2 as a benchmark framework.

🎨 Prompt of the Week

A dreamy stylized illustration in a hand painted storybook style, with soft textured brush strokes and a slightly imperfect artistic feel. Create a simple scene with a strong central subject, warm natural lighting, cinematic composition, and a nostalgic mood. Use a muted yet vibrant color palette with orange, green, and earthy tones. Include a gentle sky with soft clouds, a peaceful atmosphere, whimsical details, high visual richness, and the texture of a traditional painting.

Our readers most liked story from last issue.

Anthropic has released Claude Opus 4.8, bringing stronger coding, reasoning, and agent capabilities while keeping pricing unchanged.