New ChatGPT
OpenAI has released GPT4o. The ‘o’ is for omni, hinting that this fast new multi-modal version of the AI bot does a lot more than chat. It uses the same LLM (Large Language Model), but is a lot more potent. The original has a latency of about three seconds— reasonable for text chat, but too sluggish for speech. GPT-4o reduces that to a level suitable for verbal and visual prompts in real time, as the video demonstration proved. The system is shown a room and then asked to guess what might be going on. It guesses that the room is rigged for video work. It was then told an announcement was going to be made. It correctly postulates that this might involve OpenAI (the demonstrator is wearing a branded top). This was all pre-prepared, but still spooky. OpenAI claims that it is better at audio and visual translation than all the current rivals. The speed of AI development remains dizzying.