Introduction: Why AI is essential for feedback today
This article explores how to leverage these technologies to automate grading, personalize suggestions, and fit within the time constraints of a modern classroom.
The new tools: From LLMs to decision models in 2026
d1-ultra: A high-speed, multimodal decision model
Launched by Liquid AI last October, d1-ultra combines text and image analysis in a single pass, delivering calibrated probabilities. For education, this means analyzing drawings, written responses, or videos in milliseconds and returning structured feedback.
- Speed: 1 forward pass per input
- Multimodality: text, images, short videos
Jev: The decision model for controlled responses
TypeSafe introduced Jev in mid-September 2026. Unlike a chatbot, Jev accepts only predefined responses and returns a label, score, or reason. It’s ideal for multiple-choice assessments, rubric grading, or curriculum consistency checks.
- Interpretability: you can read the exact reason for the decision
- Integration with Perplexity SPACE for secure execution
Perplexity SPACE: Sandbox for secure AI agents
Perplexity has open-sourced SPACE, a microVM that isolates each AI task. In classrooms, this ensures every grading prompt runs in a secure environment, protecting student privacy and complying with school regulations.
How to create effective feedback prompts
Even with powerful tools, a well-structured prompt is essential. Below is a template that works with LLMs and decision models in 2026.
<!-- Prompt for automatic essay correction -->
You are an expert evaluator. Analyze the following essay and provide:
1. Overall score (0-100)
2. Three specific strengths
3. Three concrete improvement actions
Write in Italian. Use positive and constructive language.
Essay:
"""[ESSAY TEXT]""""Key elements of the prompt
- Clear role (e.g., "expert evaluator")
- Structured output (score + strengths + actions)
- Desired language (positive, constructive, Italian)
- Text delimitation instructions to avoid extra information
Practical classroom use cases
1. Grammar and style corrections
Use an LLM like d1-ultra to analyze written assignments. The model returns specific errors, punctuation suggestions, and stylistic alternatives in less than a second.
2. Personalized study recommendations
Input: student performance data (e.g., incorrect answers). Output: a personalized study plan generated by a prompt that includes strengths and weaknesses.
3. Project assessment
Upload a student’s project image to d1-ultra, include a rubric with criteria (originality, technique, compliance). The model evaluates and returns a score and reason.
Advantages and limitations: When AI works and when it doesn’t
- Advantages
- Speed: immediate feedback 24/7
- Consistency: same metrics for everyone
- Personalization: adaptation to each student’s level
- Limitations
- Creative or open-ended content still requires a human reviewer
- Decision models like Jev only work with predefined answers
- Data privacy remains a priority; use SPACE for isolation
Practical tips for teachers: Steps to follow
- Define the purpose of the feedback (correction, motivation, study planning).
- Choose the tool: LLM for open language processing, decision model for structured assessments.
- Create a prompt with clear role, format, and language constraints.
- Test the first output with a small group of students.
- Iterate the prompt based on results, then deploy using the SPACE sandbox for security.
Conclusion: The future of AI-powered feedback
In 2026, advances in LLMs and, especially, decision models are transforming how teachers give feedback. Tools like d1-ultra, Jev, and Perplexity SPACE offer speed, security, and personalization that were previously impossible. By understanding each technology’s strengths and crafting well-structured prompts, you can integrate AI into your workflow, free up time for human interaction, and help every student improve faster.
Start with a small experiment, measure the results, and watch as AI becomes your silent ally for smarter, more responsive education.
**Important: We are in 2026. All temporal references must be current. Do not mention past years such as 2024, 2023, etc. The content must be fresh and relevant to the present.** **Current context (October 2026):** Here are some recent trends and news to inspire you: - Perplexity Open-Sources SPACE, a Firecracker MicroVM Sandbox for AI Agents: Perplexity has open-sourced SPACE, the sandbox platform behind every task in Perplexity Computer. The Perplexity SPACE sandbox gives each agent task i... [2026-10-17] - Liquid AI Releases d1-ultra: A Multimodal Decision Model That Scores 65 on the AA Decision Index: Liquid AI released d1-ultra, a text and image decision model that returns calibrated probabilities in 1 forward pass. It scores 65 on the Artificial A... [2026-10-16] - 20 Jev Alternatives: The Decision Models Taking On TypeSafe in 2026: TypeSafe launched Jev on September 15, 2026. It is a decision model, not a chatbot. You define the allowed answers, and Jev returns a label, a score o... [2026-10-17] Use this current information as inspiration to create an original and relevant prompt for 2026.