🧠 AI Crushes Hardest Finance Exam

Good morning, AI enthusiast. AI just passed the brutal CFA Level III exam, a test most top analysts fail, showing it can tackle multi-step financial problems with precision.

PLUS: How to create cinematic AI video prompts without guesswork

AI NEWS

🧠 AI Crushes Hardest Finance Exam

Several leading AI models have passed the notoriously difficult CFA Level III exam, a benchmark that less than half of human candidates cleared. OpenAI’s o4-mini led the pack with 79.1%, followed by Google’s Gemini 2.5 Flash at 77.3% (passing score: 63%).

→ Tested on both multiple-choice and essay questions simulating real-world portfolio management scenarios
→ 23 models evaluated from OpenAI, Google, Anthropic, Meta, xAI, DeepSeek, and others
→ Reasoning-focused AI models excelled on complex essay tasks, while simpler tasks are now commoditized

Why is important: AI is rapidly matching and surpassing human performance in highly specialized, cognitively demanding domains like finance. While humans still excel at contextual judgment and nuanced decision-making, these results signal that AI could soon augment or automate advanced analytical work in investment management.

AI Entertainment

🎬 Xicoia’s AI actress sparks Hollywood backlash

AI talent studio Xicoia confirmed its hyperreal actress Tilly Norwood is in talks with major Hollywood talent agencies, a move drawing sharp criticism from human actors.

→ Tilly debuted in a comedy sketch last month, built with a full backstory, evolving arcs, and unique voice
→ Founder Eline Van der Velden says studios have shifted from skepticism to deal-making “in just months”
→ Van der Velden: “The age of synthetic actors isn’t coming, it’s here.”
→ Several actors urged boycotts of any agency that signs Norwood

Why is important: This could mark a turning point in labor tensions as synthetic performers enter traditional talent ecosystems. If agencies start representing AI actors, it raises questions about contracts, royalties, and the very definition of stardom.

AI NEWS

🍏 Apple tests “Veritas” chatbot to reboot Siri

Apple is using an internal chatbot, codenamed Veritas, to stress-test features for a major Siri AI overhaul, after delays pushed the launch to March 2026.

→ Veritas experiments with searching personal data and editing photos via voice
→ Tests Apple’s “Linwood” system using both in-house and third-party AI models
→ Not planned as a standalone app, features will be embedded directly into Siri
→ Engineering issues prompted executive reshuffles and talent losses

Why is important: Apple is taking an iterative, internal approach to AI voice assistants, signaling how even tech giants are navigating the challenges of next-gen AI integration while maintaining product timelines.

HOW TO AI

How to create cinematic AI video prompts without guesswork

Promptefy is a free website that removes the hardest part of working with VO3 and other video models, prompt engineering. Instead of struggling with vague prompts like “sci-fi city scene”, you just click through dropdowns for style, motion, pacing, and effects. With one click, it generates a director-level prompt that tells the AI exactly how to shoot your scene.

🧰 Who is this for

• Creators making short-form AI videos

• Filmmakers experimenting with AI cinematography

• Marketers who need polished visuals for ads

• Anyone tired of typing random prompts and hoping for the best

STEP 1: Access Promptefy

Go to Promptefy. At the top, you’ll see the Input Concept box (large white rectangle in the middle of the page with the placeholder text “e.g., A futuristic city at dusk”). This is where you type your main idea.

For this tutorial, type: “A neon cyberpunk street at night”.

STEP 2: Pick Your Model and Style

Right below the input box, you’ll see a slider labeled Cfg Scale. Leave it as default (0.7) unless you want tighter or looser control over your scene.

Scroll down a bit, and you’ll find a dropdown labeled Select Model. Click it and choose your model (e.g., Gemini 2.5 Flash).

Next to it on the right, there’s a dropdown called Style. Click it to open options like Simple, Cinematic, Anime, Vintage, etc. Choose Cinematic if you want realistic movie vibes.

STEP 3: Add Camera Motion & Effects

Below the style section, you’ll see two more dropdowns: Camera Style (left) and Camera Direction (right).

• Click Camera Style → choose between options like Dolly, VR Precision, Static Tripod.

• Click Camera Direction → pick where the camera moves (Zoom In, Zoom Out).

Next row down, you’ll see Pacing on the left and Special Effects on the right.

• Click Pacing → select Slow burn for dramatic shots or Quick Cut Rhythm for action.

• Click Special Effects → choose from Practical Effects, Smoke, Lense Artistry, etc.

STEP 4: Generate Your Optimized Prompt

At the bottom, you’ll see the dropdown for Prompt Length. Select Medium for most texuse cases (you can go Short for quick clips or Long for detailed sequences).

If you want to add extras, type them in the Custom Elements box (placeholder: “e.g., neon signs, flying cars”).

Finally, click “Generate Video Prompt”, within seconds, it will generate a full cinematic prompt. Example output:

A cinematic, wide shot captures a rain-slicked neon cyberpunk street at the dead of night, bathed in the pulsating glow of myriad vibrant, physically-present neon signs and holographic advertisements that feel grounded within the gritty urban decay. The static camera holds steady, slowly beginning a deliberate, 'slow burn' zoom, gradually drawing the viewer deeper into the intricate, bustling details of the nocturnal scene, revealing the tangible textures and authentic grit of this futuristic cityscape through its practical effects.

ESSENTIAL BITES

• OpenAI added parental controls to ChatGPT, letting parents link accounts, set limits, and adjust features for teens. Available now on web, coming soon to mobile.

• Walmart CEO Doug McMillon said AI will change every job, with roles shifting but total workforce staying flat at 2.1M.

• Tencent released HunyuanImage 3.0, an open-source text-to-image AI model that rivals leading commercial systems.

• Google updated Gemini 2.5 Flash and Flash-Lite models, improving agentic tool use, efficiency, and instruction-following capabilities.

Hot AI Tools

• 🎆 HunyuanImage 3.0: Tencent’s open-source text-to-image model

• ⚙️ Factory: High-scoring software dev agent on Terminal-Bench

• 🎥 Wan2.5: Alibaba’s new multimodal image & video AI

• ⚙️ VibeSDK: Launch your AI vibe coding platform instantly

That's it! See you tomorrow

- Dr. Alvaro Cintas