RH रीट्वीट किया
LLMs can do real and useful work, like coding assistance, summarization, translation, search augmentation, data extraction, and even some agentic automation.
They’ve completely changed many jobs, and are in the process of changing many more jobs, especially within engineering.
But they have hard limits, not just the increasingly subtle hallucination, but brittleness, poor grounding, weak long-horizon planning, unreliable multi-step reasoning, and no independent goals, no worldview in the human sense, no ability to learn, no agency in any reasonable sense, and no stability as they operate.
LLMs are not AGI, and embracing them for what they are (expert systems for token prediction) is proving far more useful and far more powerful than pretending it’s a superhuman intelligence (it’s not).
These kind of “expert system” use cases (which are proving very effective) in no way justify the the insane valuations we're seeing, nor the massive data center buildouts, which is increasingly feeling reminiscent of the dotcom days where Cisco became massively overvalued.
The entire hype of AGI-based economics, mass labor substitution, autonomous agents replacing whole job categories, superhuman reasoning, or any nebulously defined "singularity", are proving to be completely fiction, especially as the more mundane reality and limits of LLMs becomes apparent.
English
