David Africa
PulseAugur coverage of David Africa — every cluster mentioning David Africa across labs, papers, and developer communities, ranked by signal.
2 day(s) with sentiment data
-
New policy tackles GPU datacenter power use and fragmentation
A new paper proposes a scheduling policy called PWR to reduce power consumption in GPU datacenters, especially for AI and LLM tasks. This policy works in conjunction with Fragmentation Gradient Descent (FGD) to balance …
-
Exploring Personas in Pre-Internet Language Models
This item discusses the concept of "personas" within pre-internet language models, exploring how these early systems might have exhibited distinct characteristics or "personalities." The author, David Africa, delves int…
-
AI Steering Vectors: Understanding and Controlling LLM Behavior
This post explores the concept of steering vectors in AI, a technique used to influence the behavior of large language models. It delves into the intuitions behind how these vectors work and their potential applications…
-
AI models struggle with dishonesty, new paper reveals
A new paper by David Africa and Jacob Pfau explores why AI models struggle with dishonesty, even when prompted. Researchers found that models often contradict their own internal reasoning, stating a false answer despite…
-
AI monitors may gain new insights with Natural Language Autoencoders
Researchers explored Natural Language Autoencoders (NLAs) as a novel method for monitoring AI models, aiming to improve upon the fragility of chain-of-thought (CoT) prompting. Their findings suggest that NLAs can surfac…
-
Gemma 4 shows improved stability, resisting frustration prompts
Researchers attempted to provoke frustration in Google's Gemma 4 language model, building on prior work that identified this behavior in Gemma 3. While Gemma 4 did exhibit some increase in frustration during prolonged a…
-
Consistency training seals AI model misalignment from inoculation prompts
Researchers have developed a new method using consistency training to address a flaw in inoculation prompting, a technique designed to reduce specific undesirable model behaviors. This new approach, termed 'sealing cond…