Today on Don’t Fear AI

• Cutting-Edge prompt engineering techniques by Leading AI Startups

• Get ready for a career in AI with AWS AI & ML Scholars

• AI Model with Image Editing

• DeepSeek New Update Rivals Top Offerings from OpenAI and Google

• Perplexity Labs turns complex ideas into outputs like Reports, Dashboards, and Mini Web apps

Youtube

Cutting-Edge prompt engineering techniques by Leading AI Startups 

State-Of-The-Art Prompting For AI Agents explores cutting-edge prompt engineering techniques used by leading AI startups, based on insights from over a dozen companies. It covers how companies like Parahelp, Tropier, and Jasberry craft prompts by clearly defining roles and tasks, adding structured instructions, applying constraints, and controlling output formats using tools like markdown and XML-style tags.

Key takeaways include:

• Layered prompt architecture: Separating system, developer, and user prompts for modularity and control.

• Worked examples: Vital for guiding model behavior and increasingly automated through customer data.

• Advanced techniques: Including metaprompting (prompts that generate other prompts), prompt folding, and using hard examples (like unit tests) to handle complex tasks.

• Debugging strategies: Leveraging escape hatches, debug_info, thinking traces, and long context windows to refine outputs and reduce hallucinations.

• Evaluations (evals): Described as more valuable than prompts themselves, as they drive meaningful improvement and prompt design. These are developed through close user interaction, often by founders acting as “forward deployed engineers.”

Tutorial

Get ready for a career in AI with AWS AI & ML Scholars

AWS AI & ML Scholars is made up of two phases, the Challenge Phase and Nanodegree Phase. When you apply to the program, you will be auto-enrolled into the Challenge Phase, which will be open from May 28 to August 1. As you navigate the Challenge Phase you will gain foundational knowledge of Generative AI, and create your own app using AWS PartyRock.

After successful completion of the challenge phase, if you are selected to participate in the Nanodegree Phase, you will dive deeper into how Generative AI aligns with a future career in tech. When applying, choose between three courses

• Future AWS AI Scientist

• Future AWS Business Intelligence Engineer

• Future AWS AI Engineer

In the courses you will use Amazon GenAI tools to work through real world use cases.

The program covers the full cost of a 4 month Nanodegree through Udacity. The program is designed to help learners get started with AI and ML with no prior technical knowledge required.

Apply here

New Model

AI Model with Image Editing

Black Forest Labs has launched FLUX.1 Kontext, a suite of advanced multimodal image generation models that support both text-to-image and image-to-image editing with high speed and precision. Unlike traditional models, FLUX.1 Kontext allows users to combine text prompts with reference images for in-context, iterative editing while preserving character consistency, style, and detail.

Key Features:

• Multimodal generation: Accepts both text and images as inputs.

• Iterative editing: Modify and build on images step-by-step.

• Local editing: Target specific regions for changes without affecting the whole image.

• Photorealistic rendering and fast inference up to 8x faster than leading models.

• Style transfer and typography support.

Model Variants:

1. FLUX.1 Kontext [pro]: Combines local editing, text-to-image generation, and context-aware transformations with iterative capabilities.

2. FLUX.1 Kontext [max]: An experimental, high-speed model optimized for prompt adherence and visual consistency.

3. FLUX.1 Kontext [dev]: Open-weight research model (12B parameters) available via private beta for safe experimentation and development.

Check the new model here

DeepSeek New Update Rivals Top Offerings from OpenAI and Google

Chinese AI startup DeepSeek quietly released a powerful update to its R1 reasoning model (R1-0528) without fanfare, but with performance that rivals top offerings from OpenAI and Google.

Key Highlights:

• Major Performance Gains:

  • Math accuracy (AIME 2025) jumped from 70% to 87.5%.

  • Coding benchmark (LiveCodeBench) improved from 63.5% to 73.3%.

  • Reasoning improved by using 23K tokens per prompt, up from 12K.

  • Beats xAI’s Grok 3 mini and Alibaba's Qwen 3; nearly matches OpenAI’s o3/o4-mini.

• Efficiency Over Size:

  • DeepSeek released a distilled 8B parameter model (R1-0528-Qwen3-8B) that performs comparably to models 30 times larger, showing that smarter algorithms can beat brute-force size.

• Industry Impact:

  • DeepSeek’s success, despite limited access to top-tier hardware, forced Google and OpenAI to cut prices and release smaller models.

  • The MIT license allows unrestricted commercial use, fostering open innovation.

Link to details

Perplexity Labs turns complex ideas into outputs like Reports, Dashboards, and Mini Web apps

Perplexity Labs is a new feature for Perplexity AI Pro users that helps turn complex ideas into outputs like reports, dashboards, and mini web apps. It goes beyond basic search by integrating tools for deep web browsing, code execution, and visualizations.

Key Benefits:

1. Complex Project Execution – Combines multiple tools to create detailed reports and analyses.

2. Mini-App Development – Lets users build simple interactive web apps within the platform.

3. Asset Management – Automatically organizes files (charts, code, CSVs) in one place.

4. Advanced Research – Delivers deeper, more interactive results than previous modes.

5. Boosts Productivity – Automates repetitive tasks and streamlines workflows.

Details of Perplexity Labs