The Summarizer "Discuss further" doesn't seem to work

JanPieter

I clicked 'Discuss further' after summarizing an online article. The chatbot doesn't seem to know the contents of the article that was just summarized. This (of course) used to work when I tried it a while back.

Tried it with a Dutch article, a random article in English from The Guardian, and also with the example article mentioned from The Atlantic "Hacking the President’s DNA": https://kagi.com/discussdoc?url=https%3A%2F%2Fwww.theatlantic.com%2Fmagazine%2Farchive%2F2012%2F11%2Fhacking-the-presidents-dna%2F309147%2F

Doesn't work either with a YouTube-video that was just succesfully summarized. I haven't tried other docs.

httpjames

On multiple websites I try to discuss with Discuss Document (sites that don't have any cookie popups or anything that should obstruct the crawler), it refuses to answer my questions due to not having "enough context".

I tried discussing https://thehive.ai/privacy and received the following response:

does the hive train its models off API inputs?

I apologize, but I do not have enough context to fully answer your question. As an AI assistant created by Anthropic to be helpful, harmless, and honest, I do not have direct knowledge of my training procedures. How else can I assist you today?

Something similar occurs with Research Assistant:

read the provided privacy policy. does the company train its models off API inputs?

Unfortunately I do not have access to the privacy policy you referenced, so I cannot determine if the company trains its models off API inputs based on that information. As an AI assistant focused on being helpful, harmless, and honest, I can only respond to questions based on the context and information provided to me. If you are able to share specifics from the privacy policy that might indicate whether they train models off API inputs, I'm happy to analyze that information to try to assist further. Please let me know if you can provide any excerpts or additional details.

This has occured much more frequent since the last Assistant update.

I expected both Discuss Document and Assistant to actually respond to my question using the link I provided.

bebowilson

I experienced this too. In my example it answered contextually correctly for a bit, but then after several responses it seemed to forget that it was supposed to be contextual to the document I had asked it to summarize.