Jailbreaking and what it teaches us about LLMs
LLM safeguards, the Swiss cheese model, persona hijacks, prefix injection, encoded prompts, many-shot attacks.
Liam Bai Blog RSS Feed
Liam Bai's personal website
Latest posts
A visual guide to Keytruda
PD-1/PD-L1, immuno-oncology, checkpoint inhibition, binding visualizations, antibody anatomy.
From Kolmogorov to LLMs: The compression view of learning
The Minimum Description Length (MDL) principle, Kolmogorov complexity, linear regression, data compression and learning.
Protein language models through the logit lens
Applying the logit lens to ESM-2. Logit visualization & interpretation, attention analysis, looking inside the mind of a protein language model.
Protein VAEs
Predicting variant effects with variational autoencoders. DeepSequence, EVE, EVEScape. Machine learning for clinical decisions, improving pandemic preparedness by predicting antibody escape.
An introduction to variational autoencoders
Predicting protein function using deep generative models. Latent variable models, reconstruction, variational autoencoders (VAEs), Bayesian inference, evidence lower bound (ELBO).
Protein Inception
Protein design by hallucination. DeepDream, Markov Chain Monte Carlo (MCMC), KL divergence, gradient optimization, scaffolding functional sites, SARS-CoV-2 receptor traps.
How to represent a protein sequence
Learning protein representations. Transfer learning, protein language models, contextual embeddings, Transformers, masked language modeling, BERT, UniRep, ESM, attention analysis.
What we can learn from evolving proteins
Predicting protein structure and function. Multiple Sequence Alignments (MSAs), the protein folding problem, the Potts model, Direct Coupling Analysis (DCA), EVCouplings.