Research Blog

Stay up to date with content produced and managed by AIRES.

• TeenyTinyLlama: small open-source language models trained in Portuguese 🦙

  TeenyTinyLlama, a pair of compact, open-source language models trained in Brazilian Portuguese.

• Ethical Dilemmas in Software Development ⚖️🖥️

  An invitation to take part in research on the ethical dilemmas faced by Information Technology professionals.

• Aira-Instruct 🤗

  Aira-Instruct, a new series of language models fine-tuned via instruction-tuning and RLHF, released in Portuguese and English.

• Chatbots and Language Model Playgrounds

  The new version of Ai.ra and the AIRES Alignment Playground, tools for exploring chatbots and the alignment of large language models.

• Auditability and Black-Box Models: Brazil and Electronic Voting

  How redundancy, randomness, and auditing guarantee the robustness and transparency of the Brazilian electronic voting system.

• Welcome to Teeny-Tiny Castle

  Meet Teeny-Tiny Castle, AIRES's open-source repository with tools and tutorials on AI ethics and safety.

• Is LaMDA Sentient? TL;DR, No.

  A technical post deconstructing the claim that LaMDA, Google's language model, is sentient.

• Ai.ra - The AIRES Expert

  Meet Ai.ra, AIRES's expert chatbot on artificial intelligence, ethics, and AI safety.

• Data, Responsibly (Vol. 1) Mirror, Mirror

  Meet 'Mirror, Mirror,' a science comic from the Data, Responsibly series about digital accessibility and the biases produced by biased data.

• Can Women Be Cyborgs? Gender, Feminism, and Artificial Intelligence

  An essay on feminist epistemology and the implications of gender for the knowledge produced within Artificial Intelligence systems.

• Alignment with (Large) Natural Language Models

  Testing how pre-trained language models can produce undesired behaviors, and why alignment with human values matters.