3 ms·

did you read it? do you understand how statistical significance works? the literal hundreds of queries all outputting no reddit citations (Perplexity and Claude

did you read it? do you understand how statistical significance works? the literal hundreds of queries all outputting no reddit citations (Perplexity and Claude still showing this pattern from what I've tested today) does not follow from your "easy answer." but, by all means, i'm sure you've got it figured out so, yea...don't bother with the research.

verdverm 8mo ago
I did read it, the issue is not statistics, it's understanding how API calls differ from a UI. eg. You can get citations with the search grounding API in VertexAI. Apples vs Oranges, when you call the API you interact with an LLM, when you use the UI you interact with an agentic system. (other than perplexity, I won't touch their stuff) Some APIs offer extended features beyond what you find in an OpenAI compat API, like VertexAI which has "grounding with search" which returns citations as extra fields in the response JSON. For OpenAI and Claude, this is a tool call that you need to set in the request payload. If you get a zero percentage result, it should trigger your "I should verify this is doing or not what I assume it is" Share your code and we can have a more informed discussion. I get citations through the APIs, so I find it very unlikely your analysis is accurate. Consider this preprint review if you like. Also, the APIs are less restrictive with content, not more. You can get them to easily talk about things the UI with guardrails will not. This is because your average user is in the UI while they give more freedom to those using the API because you are now responsible for the output being passed to the public. Seems there are a number of false assumptions in the submission. --- From the HN commenting guidelines > Please don't comment on whether someone read an article. "Did you even read the article? It mentions that" can be shortened to "The article mentions that". I get that you are trying to use HN to promote your product, however how you interact on here will impact reception. Additionally/fyi, another rule states not to use HN __primarily__ for self promotion. You're just getting started here, the guidelines and faq linked at the bottom of most pages will be helpful to you.

verdverm 8mo ago

I did read it, the issue is not statistics, it's understanding how API calls differ from a UI. eg. You can get citations with the search grounding API in VertexAI. Apples vs Oranges, when you call the API you interact with an LLM, when you use the UI you interact with an agentic system. (other than perplexity, I won't touch their stuff) Some APIs offer extended features beyond what you find in an OpenAI compat API, like VertexAI which has "grounding with search" which returns citations as extra fields in the response JSON. For OpenAI and Claude, this is a tool call that you need to set in the request payload. If you get a zero percentage result, it should trigger your "I should verify this is doing or not what I assume it is" Share your code and we can have a more informed discussion. I get citations through the APIs, so I find it very unlikely your analysis is accurate. Consider this preprint review if you like. Also, the APIs are less restrictive with content, not more. You can get them to easily talk about things the UI with guardrails will not. This is because your average user is in the UI while they give more freedom to those using the API because you are now responsible for the output being passed to the public. Seems there are a number of false assumptions in the submission. --- From the HN commenting guidelines > Please don't comment on whether someone read an article. "Did you even read the article? It mentions that" can be shortened to "The article mentions that". I get that you are trying to use HN to promote your product, however how you interact on here will impact reception. Additionally/fyi, another rule states not to use HN __primarily__ for self promotion. You're just getting started here, the guidelines and faq linked at the bottom of most pages will be helpful to you.