TechFlow news: Haseeb Qureshi, partner at Dragonfly Capital, posted that the hype around large language models (LLMs) has somewhat cooled down. People are beginning to realize they aren't perfect and can produce errors and incoherent outputs. There's now less overbelief in AI hype claims about what LLMs can achieve that currently remains unattainable.
Although LLMs have made significant breakthroughs in recent years—such as the major leap from GPT-2 to GPT-3 and the notable improvements from GPT-3 to GPT-4—Haseeb Qureshi believes GPT-5 may represent only a modest upgrade for most users.
Haseeb stated that in the future, we need to integrate LLMs with system engines so they can play more effective roles across broader application scenarios, such as handling specific tasks like spreadsheet processing, game engines, role-playing, and scriptwriting. This won’t require revolutionary breakthroughs but rather time and sustained effort.
Notably, many large tech companies are currently conducting internal experiments and research on LLMs.
