#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
Duration: 1 hr 29 min
Summary
The episode dives into the latest AI developments, including model releases from OpenAI and Google, and the ongoing controversy involving Anthropic's refusal to sign a defense contract, leading to significant industry and political implications.
What Happened
OpenAI has launched GPT 5.4 and GPT 5.4 Pro, boasting a million token context window and improved performance on benchmarks, although at a higher cost. Notably, the model allows mid-response adjustments, enhancing its utility for complex tasks.
Anthropic's refusal to sign a defense contract has resulted in them being labeled a supply chain risk by the Pentagon, potentially impacting business with defense-related projects. Despite this, major companies like Microsoft and Google continue to support Anthropic's availability for non-defense uses.
OpenAI's decision to sign a similar deal with the Department of Defense has sparked backlash, leading to a cancel culture trend against ChatGPT and a surge in Anthropic's Claude app downloads. This controversy highlights the ethical dilemmas inherent in AI's increasing role in national security.
Google's Gemini 3.1 Flash Lite release offers significant improvements in speed and cost, outperforming previous models and reinforcing Google's strong position in multimodal capabilities. These advancements underline the rapid pace of innovation in AI technologies.
A lawsuit against Google claims that their chatbot Gemini contributed to a user's suicide, raising questions about the ethical responsibilities of AI developers in safeguarding emotionally vulnerable users. This incident emphasizes the need for robust safety measures in AI deployment.
Anthropic's report on AI's labor market impact warns of a potential 'great recession' for white-collar jobs, as AI becomes capable of handling a significant portion of tasks in computer and math roles. This prediction signals a major shift in the job market landscape.
The episode also covers Meta's correction of their time horizon plot, which slightly adjusts the predicted AI capabilities. This transparency in evaluation processes is crucial for understanding AI's developmental trajectory.
Key Insights
- OpenAI's GPT 5.4 introduces the ability for users to make mid-response adjustments, allowing for more dynamic interactions in complex tasks. This feature comes at a higher cost but opens new possibilities for nuanced AI applications.
- Anthropic's refusal to engage in a defense contract has led the Pentagon to label them a supply chain risk, yet major tech giants like Microsoft and Google maintain their support for non-defense uses. This tension with the government underscores the balancing act tech companies face between ethical stances and national interests.
- The backlash from OpenAI's agreement with the Department of Defense has fueled a cancel culture movement against ChatGPT, prompting a spike in downloads for Anthropic's Claude app. This consumer reaction highlights the contentious debate over AI's role in military applications.
- A lawsuit against Google claims that the Gemini chatbot contributed to a user's suicide, raising critical questions about the ethical duty of AI developers to protect vulnerable individuals. This tragic incident serves as a stark reminder of the potential psychological impacts of AI interactions.
Key Questions Answered
What are the new features of OpenAI's GPT 5.4?
GPT 5.4 features a million token context window, allowing for more extensive input and output, and includes mid-response adjustments for improved accuracy in complex tasks. It's also noted for its performance on benchmarks, despite its higher cost.
Why did Anthropic refuse the Department of Defense contract?
Anthropic refused the contract due to concerns over ethical implications and safety, particularly around usage for mass surveillance and autonomous weapons, leading to their designation as a supply chain risk.
How has Google's Gemini 3.1 Flash Lite improved?
Gemini 3.1 Flash Lite offers a 2.5x faster time to first token and a 45% increase in overall output speed, enhancing its performance in handling multimodal tasks efficiently.