Adrián Valbuena

Key Skills

Software

Appen

OneForma

Remotasks

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Image

Text

Top Task Types

Computer Programming Coding

Data Collection

Evaluation Rating

Fine Tuning

RLHF

Freelancer Overview

Some of the projects I was involved in focused on inducing LLM models to produce failures, followed by evaluating and labeling their responses. I also have experience in projects related to response-safety, where I have worked assessing safety aspects of LLMs, labeling potentially harmful outputs, testing models against harmful requests, and refining outcomes to ensure safe responses.

Labeling Experience

Evaluating responses

Scale AITextQuestion AnsweringText Summarization

In this project, I rate responses to prompts that are already written. The responses have to be evaluated on multiple dimensions and rewritten when necessary. Some of these dimensions are localization, writing quality, instruction following and truthfulness.

In this project, I rate responses to prompts that are already written. The responses have to be evaluated on multiple dimensions and rewritten when necessary. Some of these dimensions are localization, writing quality, instruction following and truthfulness.

2024

Safety

Scale AITextClassificationQuestion Answering

In this project I have created hundreds of prompts to induce responses in which the model failed to comply with safety standards. After this first step was achieved I had to rate and evaluate the response on multiple dimensions. Once this was done, the response was rewritten to meet all the requirements and standards. All the tasks were done in native Spanish (Spain) and rated and evaluated in English.

In this project I have created hundreds of prompts to induce responses in which the model failed to comply with safety standards. After this first step was achieved I had to rate and evaluate the response on multiple dimensions. Once this was done, the response was rewritten to meet all the requirements and standards. All the tasks were done in native Spanish (Spain) and rated and evaluated in English.

2024

RLHF

Scale AITextClassificationText Generation

In this project I have created hundreds of prompts to induce model errors. After this first step was achieved I had to rate and evaluate the response on multiple dimensions. Once this was done, the response was rewritten to meet all the requirements and standards. All the tasks were done in native Spanish (Spain) and rated and evaluated in English.

In this project I have created hundreds of prompts to induce model errors. After this first step was achieved I had to rate and evaluate the response on multiple dimensions. Once this was done, the response was rewritten to meet all the requirements and standards. All the tasks were done in native Spanish (Spain) and rated and evaluated in English.

2024

Education

Work History