PhD Users Question AI Tool Value

The hot take: this is less a verdict that AI is “bad” now and more a sign that the hype-to-utility ratio is collapsing for experienced users. Once you know how to use these tools, the remaining gains can be swallowed by verification overhead, cost, and occasional confident nonsense.

• –Claude Opus is priced at the premium end of the market, with Anthropic listing Opus 4.7 at $5 per million input tokens and $25 per million output tokens, so “fortune” is not far off for heavy use.
• –PhD and research work is a worst-case environment for hallucinations because a single bad answer can waste more time than the model saved.
• –The strongest AI use cases are narrowing to bounded tasks like drafting, summarizing, brainstorming, and code scaffolding, not as a source of truth.
• –The post signals a maturing market: users are becoming more selective, comparing models on reliability per dollar rather than raw capability.
• –The likely outcome is not quitting AI entirely, but shifting to cheaper models or using premium models only when the task really warrants them.