Hacker News

Protostome

Tell HN: Claude is completely unusable for biology

I've found that doing basic biology research with Claude has become extremely difficult. I'm not talking about anything remotely related to developing a bioweapon, just straightforward immunology questions that almost always end up being flagged.

I completely understand the need for guardrails, but in practice they can significantly reduce the tool's usefulness for legitimate scientific research. It sometimes feels like the model is better suited for "vibe coding" than for serving as a serious research assistant.

What's more funny is that you can use claude code together with molecular design models like RFdiffusion and Alphafold to develop a real bioweapon, but since this is in the realm of software, the prompts do not get flagged at all. Those models can even run on consumer GPUs so the limit of access is almost not there.

Has anyone else had a similar experience?

the__alchemista day ago

I do a lot of bio and chem, and have been using a mix of Claude, Gemini, and ChatGPT. I have a reasonably well equipped mol bio home lab, and do a lot of structural bio/chem work in rust/CUDA.

I only recall 2 censorship problems:

  - ChatGPT when I asked about LNPs
  - Claude Fable in the brief period before it was removed: Would downgrade with the safety warning on any question related to bio or my code bases, but it didn't block me; just downgraded to an older model. (On every query though!)

Which sorts of questions did you have trouble with? For context, my questions have been about wet lab protocols (How much of X reagent to mix with Y samples using Z equipment, bacteria culture questions/steps, interpreting sequencing or gel results etc), recombinant DNA stuff, molecular dynamics and related work between proteins and small organic molecules, lipids, nucleic acids etc.

Protostomeopa day ago

Here is a question that was just flagged: https://imgur.com/a/k90ce00

This is absurd

skyberrys2 days ago

Well it's not biology related, but Claude is the only model that has told me things that seem benign are racist or self harming for general conversations. These were not at all in that direction. They were very basic queries about how to reactivate an old app account or a question about the ingredients in a food dish image.

PJHkorea2 days ago

Did you happen to start biology-related work immediately in a new session? Before starting, how about laying the groundwork for educational purposes using context for about 10 to 15 turns?

jlongr2 days ago

What do you consider to be straightfoward immunology questions?

source