AI Applied: Covering AI News, Interviews and Tools - ChatGPT, Midjourney, Gemini, OpenAI, Anthropic OpenAI Launches GPT-5.4 with Mid-Response Steering
8 snips
Mar 10, 2026 They break down GPT-5.4’s standout features like mid-response steering, million-token context, and native computer use. Conversation covers real-time correction UX, first-hand impressions of mid-steering, and concerns about models being overly cautious. They also tackle regulatory debates around AI in regulated industries and compare OpenAI’s approach to competitors such as Anthropic and Gemini.
AI Snips
Chapters
Transcript
Episode notes
Mid Response Steering Changes Interaction
- GPT-5.4 introduces mid-response steering so users can correct the model while it is still composing an answer.
- Jaeden described interrupting a long generated reply (e.g., cruise vs airport context) and the model incorporating the new info mid-reply.
Million Token Context Enables Longer Workflows
- GPT-5.4 supports up to a million tokens of context enabling much longer conversations or documents.
- Jaeden highlighted this alongside improvements in web search and knowledge work for extended context tasks.
IV Air Bubble Query Triggered Overcautious Response
- Conor tested a medical safety question about IV air bubbles and experienced the model doing an extensive response then retracting.
- He described the model giving a full answer then pulling back with cautious safety behavior.