Call for Position: DTIC-PLCT-2025-10 - Medium support technician: research assistant to develop a research plan on detecting and mitigating discriminatory biases, and prevent manipulation (“jailbreaking”) in Large Language Models (LLMs)

24 Jan 2025
Job Information

Organisation/Company

Universitat Pompeu Fabra - Department / School of Engineering

Department

Engineering

Research Field

Computer science

Researcher Profile

First Stage Researcher (R1)

Positions

Research Support Positions

Country

Spain

Application Deadline

23 Feb 2025 - 23:59 (Europe/Madrid)

Type of Contract

Other

Type of Contract Extra Information

Indefinite according to RD 32/2021 of December 28, 2021

Job Status

Full-time

Hours Per Week

35

Is the job funded through the EU Research Framework Programme?

European Union / Next Generation EU

Reference Number

MICIU/AEI/UE - CPP2023-010780

Is the Job related to staff position within a Research Infrastructure?

No

Offer Description

Project description: 

The project goal is to develop a research plan to improve the robustness and trustworthiness of Large Language Models (LLMs). In particular, we will focus on two aspects. First, whether the LLM, in response to regular inputs, generates text that can be understood as reflecting discriminatory biases, such as demeaning stereotypes. Second, whether the LLM is susceptible of being manipulated through prompts into outputting misleading or harmful text, i.e., “jailbreaking”.

Tasks to be performed:

Specifically, the position has the following objectives to work over 3 months in the work package WP1 (“Algorithmic fairness and discriminatory biases”) of this private-public collaboration:

• To study the state of the art in algorithmic discrimination in LLMs, in particular theoretical taxonomies of harm and practical real-world incidents involving LLMs in industrial settings.
• To consider the values that are reflected into these taxonomies of harm, and the extent to which those values could be expanded through the consideration of intercultural digital ethics.
• To propose participatory and experimental methodologies to expand and refine taxonomies of harm, adopting a practical perspective.
• To study the state of the art in jailbreaking in LLMs.
• To propose participatory methodologies that can lead to describe and thus prevent new attack modalities of LLMs.

The duration of these tasks is estimated over a period of 3 months.

Group and complement: Group 2 + Level from u to j based on experience and skills.

Dedication and working hours: Full time (35h/week).

Planned remuneration approx: From 32,591.06 € to 39,005.54 € gross per year based on experience and skills.

Financing fund: Proyecto PRESP05324 - MICIU/AEI/UE - CPP2023-010780 - Castillo, Carlos, “Habilitando Modelos de Lenguaje Responsables e Inclusivos”, financiado por MICIU/AEI /10.13039/501100011033 y por FEDER, UE.

Where to apply

Website

http://apply.interfolio.com/162445

Requirements

Research Field

Computer science

Education Level

Bachelor Degree or equivalent

Specific Requirements

• Bachelor in computer science or equivalent.

The expected start date is March 24st 2025, the job is in Barcelona.

Additional Information
Eligibility criteria

Selection criteria: The selection of the candidates will be made through evaluation of the curriculum and, where appropriate, with the carrying out a test and/or interview. Valuation will be as follows:

• Academic Training (0-30 points).
  • Postgraduate degree in computer science, data science, data analyst, or similar.
• Other professional training and experience, adequacy to the proposed profile (0-40 points):
  • Proven experience using large language models.
  • Proven experience developing large language models, a big plus.
• Other merits (0-30 points):
  • Research experience in algorithmic fairness.

The minimum score to pass the selection process is 75 points. The candidate with the highest score in the selection process will be offered the job.

Additional comments

For more information about the call, how to apply, the list of those admitted and excluded, as well as the hiring proposal, check the website: https://www.upf.edu/web/etic/convocatories-psr

Work Location(s)

Number of offers available

1

Company/Institute

Universitat Pompeu Fabra - ETIC

Country

Spain

City

Barcelona

Postal Code

08018

Street

Roc Boronat 138

Geofield

Contact

City

Barcelona

Website

https://www.upf.edu/web/etic/working-enginyeria

Street

Roc Boronat 138

Postal Code

08018

E-Mail

recerca.enginyeria@upf.edu
recruitment.engineering@upf.edu

STATUS: EXPIRED

• X (formerly Twitter)
• Facebook
• LinkedIn
• Whatsapp
• 
  More share options
  • E-mail
  • Pocket
  • Viadeo
  • Gmail
  • Weibo
  • Blogger
  • Qzone
  • YahooMail