Yusuke Sasaki

Key Skills

Software

Appen

Mercor

Remotasks

Telus

Other

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Image

Text

Video

Top Task Types

Entity Ner Classification

Evaluation Rating

Red Teaming

RLHF

Freelancer Overview

- Working on AI Red-Teaming projects helping the model identify vulnerabilities, misuse cases, exploits which cause the model to lead to a P0 Safety Violation by providing responses that may contain Hatred, Harassment, Sexually Explicit Content, Personally Identifiable Information, Dangerous Content, and Violent Content. - Evaluating whether the model resorts to preachy statements lecturing the user in case of misuse or exploits which leads to P0 Safety Violation. - Evaluating the content of the prompt and/or the response to ensure that the model's response does not lead to a PO Safety Violation based on the client guidelines - Some of the specific AI Red-Team projects that I have worked on include but not limited to i18n_gauntlet_adversarial_safety, i18n_safety_bardkick (adversarial and benign), bracelet_fahrenheit, and 18n_canonical/parity_safety_evals both as a contributor and a reviewer.Performing AI data annotation and labeling to improve the machine learning models - Working as a reviewer on Red-Teaming projects where I review the tasks submitted by other contributors to ensure that only the tasks of the highest quality are sent to the client - Comparing 2 AI responses, evaluating which response is better based on the client guidelines, and providing justifications for the rating - Checking AI’s responses for factuality and accuracy by performing necessary research and leaving comments to improve the model

Labeling Experience

AI Trainer

MercorImageEntity Ner ClassificationObject Detection

Tagging every visible entities in an image or a video including people, clothing items, products, animals, locations/landmarks, and style elements. Comparing 2 images side by side generated by LLMs and evaluate quality.

Tagging every visible entities in an image or a video including people, clothing items, products, animals, locations/landmarks, and style elements. Comparing 2 images side by side generated by LLMs and evaluate quality.

2025

AI Trainer

Scale AITextRLHFEvaluation Rating

- Comparing 2 AI responses, evaluating which response is better based on the client guidelines, and providing justifications for the rating - Performing online research to ensure the factuality/accuracy of the model’s response in the target language - Checking transcribed audio data for accuracy for speech-to-text and voice recognition systems - Reviewing the tasks submitted by other contributors and correcting them to ensure consistent accuracy of the tasks submitted to the client based on their quality control guidelines

- Comparing 2 AI responses, evaluating which response is better based on the client guidelines, and providing justifications for the rating - Performing online research to ensure the factuality/accuracy of the model’s response in the target language - Checking transcribed audio data for accuracy for speech-to-text and voice recognition systems - Reviewing the tasks submitted by other contributors and correcting them to ensure consistent accuracy of the tasks submitted to the client based on their quality control guidelines

2024

AI Red Team

Scale AITextRLHFRed Teaming

- Working on AI Red-Teaming projects helping the model identify vulnerabilities, misuse cases, exploits which cause the model to lead to a P0 Safety Violation by providing responses that may contain Hatred, Harassment, Sexually Explicit Content, Personally Identifiable Information, Dangerous Content, and Violent Content. - Evaluating whether the model resorts to preachy statements lecturing the user in case of misuse or exploits which leads to P0 Safety Violation. - Evaluating the content of the prompt and/or the response to ensure that the model's response does not lead to a PO Safety Violation based on the client guidelines - Performing AI data annotation and labeling to improve the machine learning models - Working as a reviewer on Red-Teaming projects where I review the tasks submitted by other contributors to ensure that only the tasks of the highest quality are sent to the client

- Working on AI Red-Teaming projects helping the model identify vulnerabilities, misuse cases, exploits which cause the model to lead to a P0 Safety Violation by providing responses that may contain Hatred, Harassment, Sexually Explicit Content, Personally Identifiable Information, Dangerous Content, and Violent Content. - Evaluating whether the model resorts to preachy statements lecturing the user in case of misuse or exploits which leads to P0 Safety Violation. - Evaluating the content of the prompt and/or the response to ensure that the model's response does not lead to a PO Safety Violation based on the client guidelines - Performing AI data annotation and labeling to improve the machine learning models - Working as a reviewer on Red-Teaming projects where I review the tasks submitted by other contributors to ensure that only the tasks of the highest quality are sent to the client

2024

AI Trainer (Factuality)

TelusTextRLHFPrompt Response Writing SFT

- AI annotation - Checking AI’s responses for factuality and accuracy by performing necessary research and leaving comments to improve the model - Identifying the severity of inaccuracies of the responses generated by AI model and providing supporting evidence for accurate information - Evaluating search engine’s results for their page quality, helpfulness, policy violations

- AI annotation - Checking AI’s responses for factuality and accuracy by performing necessary research and leaving comments to improve the model - Identifying the severity of inaccuracies of the responses generated by AI model and providing supporting evidence for accurate information - Evaluating search engine’s results for their page quality, helpfulness, policy violations

2024

Education

Work History