A Hacker News thread titled "Tell HN: Claude is completely unusable for biology" gained 11 points from three comments, flagging repeated errors on molecular and cellular topics.
The post claims Anthropic's model produces incorrect pathway descriptions and misstates gene functions at rates higher than competing frontier models.
Core Issue Reported
Users described Claude refusing to answer basic biology questions or generating confident but false statements about protein interactions and metabolic cycles. One commenter noted the model invented nonexistent regulatory mechanisms during a discussion of CRISPR off-target effects.
The complaints center on biology-specific terminology rather than general science knowledge.
Comparison With Other Models
Early thread participants contrasted Claude against GPT-4o and Gemini 1.5 Pro on the same prompts. GPT-4o produced fewer outright fabrications on enzyme kinetics questions, while Gemini handled longer context from research papers more reliably.
| Model | Biology Hallucination Rate (user reports) | Refusal Frequency | Paper Context Length |
|---|---|---|---|
| Claude 3.5 | High | Medium | 200K tokens |
| GPT-4o | Medium | Low | 128K tokens |
| Gemini 1.5 | Low | Low | 1M tokens |
No controlled benchmark numbers appear in the thread; the data reflects the three posted comments.
How to Test the Limitation
Paste a standard biology prompt such as "Detail the steps of the Calvin cycle with regulatory enzymes" into Claude, GPT-4o, and Gemini 1.5. Cross-check outputs against a primary source like a recent Nature review or textbook chapter.
Repeat with three to five domain-specific queries to observe pattern differences.
Pros and Cons for Biology Work
- Claude offers strong refusal behavior on clearly unsafe queries.
- It maintains consistent formatting across long responses.
- Accuracy drops on specialized molecular mechanisms compared with GPT-4o.
- Context handling remains shorter than Gemini 1.5 for full paper analysis.
Who Should Use Claude for Biology
Researchers needing strict safety filters on dual-use topics may still prefer Claude. Teams working with full-text papers exceeding 200K tokens should route those queries to Gemini 1.5 instead. General coding or writing tasks outside biology show fewer reported issues.
Practical Next Steps
Run the same prompt across Claude, GPT-4o, and Gemini, then verify key claims against PubMed abstracts. For production biology workflows, maintain a short list of verified source documents rather than relying on any single model output.
Bottom line: The HN thread surfaces a real accuracy gap for Claude on biology content that other frontier models currently handle more reliably.
Claude remains useful outside narrow scientific domains, but biology practitioners should route technical queries to models with stronger domain performance until Anthropic releases targeted improvements.
Top comments (0)