LLMs Are Just Fancy Polish: Experts Claim Current AI Hits a Wall, Calling Tools 'Asbestos' Over Hype
LLMs prove useful for narrow tasks like automating GDPR deletion requests, but the technology fundamentally fails when accuracy is non-negotiable, especially in legal or complex coding scenarios.
The community splits sharply: some see productivity gains, citing tools baked into professional software like Adobe. Others call the entire movement an overhyped distraction, with 'TootSweet' labeling it 'asbestos' blocking real AGI progress. Specific failures were pointed out by 'CodenameDarlen,' who noted hallucination persists even when sources are provided. Furthermore, users point out that the human's role has shifted to paid 'evaluation,' meaning the expert must constantly vet the output.
The weight of opinion shows LLMs are not replacements for expertise; they are aids that demand intense human scrutiny. The fault lines exist between valuing narrow utility (like basic automation) and distrusting the foundational models' inherent inaccuracies, suggesting profit-driven infrastructure is forcing adoption despite known limitations.
Key Points
LLMs are limited to non-critical, assistance-level tasks.
General consensus dictates models fail when 100% accuracy is required (e.g., law, accounting).
The technology is seen by some as a distraction or outright sham.
'TootSweet' and 'CodenameDarlen' dismiss the hype, labeling it 'asbestos' or 'scams'.
Human review remains mandatory because AI output is often faulty.
'BA5B' stated that LLMs force a premium on 'evaluation,' keeping human responsibility paramount.
Specific model failures prove the unreliability of LLMs.
'CodenameDarlen' observed hallucination generating links without sourcing content from cited pages using Qwen 3.5.
AI integrated into existing professional software is proving genuinely useful.
'Hackworth' noted the utility of generative fill and AI noise reduction features within suites like Adobe.
Source Discussions (3)
This report was synthesized from the following Lemmy discussions, ranked by community score.