REVIEWS / GENERAL / OWNER INSIGHTS

🦉 WE READ 584 OWNER COMMENTS

Hallucinate: what owners actually say

Users are deeply frustrated by AI tools that hallucinate, deceive, and underdeliver — but a subset of developers find real value when they learn to work around the flaws.

LEMMY · 432 HACKERNEWS · 75 YOUTUBE · 59 REDDIT · 14 STACKEXCHANGE · 4

What owners complain about

  • Deceptive AI customer support COMMON

    Cursor deployed an AI support bot named 'Sam' that was not labeled as AI, leading users to believe they were talking to a human. Users describe this as intentionally deceptive, and the company then attempted to 'control the narrative on reddit' when called out.

  • Hallucinated outputs can't be trusted COMMON

    LLMs generate fictitious information — fake legal cases and citations, wrong code, fabricated facts. The core problem users identify is that 'you don't know which 5% is wrong,' making output unreliable without thorough verification.

  • Visual quality degradation SOME

    Referring to Nvidia's DLSS/AI upscaling, users call it a 'slop filter' with ridiculous ghosting. The CEO's defense of the technology is met with mockery and accusations of justifying sunk R&D costs.

  • AI forced into unwanted products SOME

    Users compare AI proliferation to 'the reverse headphone jack' — shoved into everything whether people want it or not. Phone AI features are called 'bullshit' and not worth upgrading for.

  • Account and credit abuse systems FEW

    Cursor's free-credits-without-credit-card model led to abuse, forcing changes to how their VSCode fork handles user identification, which degraded the experience for legitimate users.

What owners love

  • Daily coding utility

    Some developers love Cursor and use it daily; entire companies are building workflows and MCP tools on top of it, embedding 'Active Curation' into their development process.

  • Unblocks stuck projects

    One user described being stuck on a long-term code project for various reasons; after a friend explained how to properly talk to the LLM, they made genuine progress by changing their approach.

  • Acceptable imperfection

    Users compare AI to IDE tab completion (wrong >5% of the time but still useful) and calculators — tools that are imperfect but valuable when you understand their failure modes.

Surprising patterns

  • Lawyers have been disbarred and faced contempt charges for submitting unvetted AI-generated legal filings with fictitious cases — users unanimously support harsh punishment for this.
  • Stack traces are now appearing with notes attributing issues to AI-generated code, marking what may be the first documentation of AI's footprint in bug reports.
  • Users are actively developing adversarial techniques to mess with AI support bots, including prompt injections that force bots to give proper answers or reveal their nature.
  • Even enthusiastic AI users explicitly warn that you must verify everything — the tolerance for errors exists only when users can identify and correct them, not when errors are hidden.

WHO SHOULD SKIP IT

Anyone who needs guaranteed accurate output — especially legal professionals, support staff requiring accountability, and users who can't or won't manually verify every claim, citation, or code block the tool produces.

3.1/10 GYIBB verdict
Full review →

Synthesised from 584 real owner comments across 5 platforms. Every point is grounded in the comments — no marketing, no AI guessing. How we do it →