Skip to content
View ebt55's full-sized avatar
☺️
Living life
☺️
Living life
3 followers · 6 following

Achievements

Achievement: Pull SharkAchievement: Pair ExtraordinaireAchievement: YOLO

Achievements

Achievement: Pull SharkAchievement: Pair ExtraordinaireAchievement: YOLO

Highlights

  • Pro

Block or report ebt55

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ebt55/README.md

Ebin Babu Thomas

AI / ML Engineer · AI Safety (Mechanistic Interpretability & AI Control)

I build production LLM, RAG, and agent systems, and do independent research on how to audit and control model behaviour. Currently working at the intersection of mechanistic interpretability and AI control: combining internal model signals (probes, model diffing) with black-box monitoring, counterfactual resampling, and evaluation harnesses.

ebinbt.dev · hello@ebinbt.dev

Pinned Loading

  1. diarist diarist Public

    Local, speaker-separated meeting transcripts for Windows. Live or from a file, up to 8 speakers, nothing leaves your PC.

    Rust

  2. diffing-agent-bench diffing-agent-bench Public

    Sealed, preregistered benchmark for black-box model-diffing agents: five LoRA finetunes of Qwen3.5-9B (one null, three planted behaviours, one dropped backdoor), audited blind by my implementation …

    Python

  3. kobayashi-maru kobayashi-maru Public

    Does filling an AI agent's task list with impossible work make it cheat on the tasks it could still solve? A preregistered dose-response study over 8,959 agent runs, with every raw record published.

    Python

  4. whose-voice whose-voice Public

    Blind attribution of the principal behind a covertly poisoned training corpus, against 47 candidates. Secret Loyalties hackathon, Jul 2026; corrected Sep 2026.

    Python

  5. digital-grimace-scale digital-grimace-scale Public

    A preregistered test of whether false or hostile feedback moves an open model's answer margin. The primary test failed and is published.

    Python

  6. exactdoc exactdoc Public

    Turn PDFs into editable Word documents that look like the original — checked page by page in Word, LibreOffice and Google Docs.

    Python 1