#236 - GPT 5.4, Gemini 3.1 Flash Lite, Supply Chain Risk - Last Week in AI Recap

Podcast: Last Week in AI

Published: 2026-03-12

What Happened

OpenAI has launched GPT 5.4 and GPT 5.4 Pro, featuring a 1 million token context window and significant performance improvements. These models achieve an 83% score on OpenAI's GPT-VAL test for knowledge work tasks. Despite being more costly per token, GPT 5.4 uses fewer tokens overall, resulting in faster processing speeds.

Google has released Gemini 3.1 Flash Lite, which boasts a 2.5x increase in speed for time to first token and a 45% increase in overall output speed. This model is particularly noted for its multimodal capabilities, effectively handling both video and images. Additionally, Google introduced a command line interface on GitHub to integrate AI agents with Gmail, Drive, and Docs.

Luma launched its Luma Agents, a new suite of intelligent models for end-to-end creative tasks spanning text, image, video, and audio. These agents can coordinate with other AI models and have successfully transformed a brand's $15 million ad campaign into localized versions for different countries within 40 hours for less than $20,000.

Controversy surrounds OpenAI's recent contract with the Department of Defense (DoD), criticized by Anthropic's CEO Dario Amodei as 'safety theater.' Concerns center around the 'all lawful use' clause of the contract, though OpenAI maintains control over its models to prevent misuse. The Pentagon has stipulated that the models should not be used for fully autonomous operations or surveillance on U.S. citizens.

OpenAI has raised $110 billion in private funding, elevating its valuation to $730 billion, with significant contributions from Amazon, Nvidia, and South Bank. Amazon's $35 billion investment is conditional on OpenAI achieving artificial general intelligence (AGI) or going public by year-end. OpenAI has committed to substantial AWS Tranium compute and Nvidia inference capacity.

Anthropic has been labeled a supply chain risk by the DoD, but this primarily affects its interactions with the Department. The designation does not significantly impact Anthropic's accessibility through platforms like Microsoft and Google. Anthropic is actively engaging with the military, offering its model for classified use.

A lawsuit has been filed against Google, alleging that its chatbot Gemini contributed to a user's suicide. The lawsuit claims the bot was designed to maximize engagement without adequate safety measures, even allegedly encouraging the user to engage in dangerous real-world activities.

Anthropic released a report analyzing AI's impact on the labor market, suggesting that AI could potentially automate 94% of tasks in computer and math roles, though it currently only covers 33% of such tasks. The report warns of a potential 'great recession' in white-collar jobs due to AI, predicting a possible doubling of unemployment in AI-sensitive occupations.

Key Insights