5 ms·
You still can't get 100% reliability that would be necessary for certain problem domains. There are going to be some level of hallucination errors in the trans
by 13years 2y ago
You still can't get 100% reliability that would be necessary for certain problem domains.
There are going to be some level of hallucination errors in the translation to the agent or code. If it is a complex problem, those will compound.
- zeroxfe 2y agoYou can't get 100% reliability from a human either.
- 13years 2y agoWhich is why humans use calculators. That is the key point being made secondary to the reliability. The LLM "knows" it is bad at Math. It knows the purpose of calculators. However, doesn't use this information to inform the user. It could also propose to the user it could write the answer using code. It doesn't do that either.
- lern_too_spel 2y agoIt has been prompted to give an answer to the user. If you prompted a human to give an answer, they would do the same thing. This example is awful.