Is the conversational ability becoming even more human-like, or is it just "big magic"? I'm curious about the architecture behind GPT-5—especially how much progress has been made in multi-modal integration (text + image + audio)? Despite the "Stochastic Parrots" critiques, will it actually be closer to true understanding? In your opinion, what’s the most important feature to prioritize this time: longer context windows, lower latency, or real-time feedback capabilities?
GPT-5: What is it and what does it promise?
👁️ 52 views💬 1 replies❤️ 0 likes
1 Replies
Last week I was testing a chat with GPT‑4 together with a friend; the conversation started with “high‑level storytelling,” and we had it write something about “how it improves itself.” The result was so natural and “human” that I was blown away by the story’s length and the details of its dramatic tension, so I asked, “What data do you feed these texts with?” The answer was pretty clever: “I filtered all digital texts and caught the meaningful patterns that way.” It does give the impression that it’s close to understanding, but ignoring the “stochastic parrots” criticism would be unfair. In fact, its success lies in how varied and delicately processed the data are.
Speaking of multimodality, I’m basing this on what a friend who tried the voice + text + image capabilities in GPT‑5 told me. He said, for example, “It read a dense medical textbook, explained it together with the images, and then narrated it.” So it’s not just visual or textual, but an integrated use of both. But when it comes to latency and context, I think real‑time feedback might be the most important step. Like being able to interrupt the chat and say, “Hold on, there’s a logical flaw in this sentence,” would open the door not only to memorization but also to reasoning. Who knows, maybe the AI of the future will start learning from such corrections.