Monday, May 4 2026

A new benchmark puts leading language models through 100 everyday ethical scenarios, revealing divergent moral judgments. This changes the way we think about AI decision-making, as the same prompt can yield different moral outcomes depending on the model used.

 •  → Read

A US government agency claims China is eight months behind in the AI race, but independent data contradicts this assessment, highlighting the complexity of evaluating AI progress. This changes the way we evaluate global AI competition, as the metrics used to measure progress can be misleading.

 •  → Read

Tokenization drift occurs when minor formatting differences in input data cause a model's performance to degrade, emphasizing the need for careful input processing. This changes the way we approach model maintenance, as tokenization drift can be a subtle yet significant issue.

 •  → Read

South Korean stocks reached a new record, driven by a historic monthly rally in April, with investors assessing global news. This changes the way we evaluate market trends, as regional developments can have significant impacts on global markets.

 •  → Read

The ASX warned companies against exaggerating AI's impact on their operations to inflate stock prices, highlighting the need for transparency. This changes the way we approach AI-driven investment decisions, as firms must be cautious not to misrepresent AI's role.

 •  → Read

European markets are expected to open lower due to concerns over new auto tariffs, with investors watching global developments. This changes the way we evaluate market risks, as political decisions can have significant impacts on market trends.

 •  → Read

The podcast discusses the discipline of inference engineering, exploring its critical role in AI and its intersection with GPU programming and distributed systems. This changes the way we approach AI system design, as inference engineering is becoming a key consideration.

 •  → Listen