As defined "ai fluency" - actually being good at working with ai is in most areas, a rare skill I think. At this stage probably meant you played with a certain task a lot & developed an intuition. Can't get it by just reading some piece for Lay person telling you LLMs are next token predictor etc

You not gong to get normal google latencies of course, this is closer to federated search speeds if you know what that is. Its kinda like a human using another specialised search engine, entering queries and looking at returned queries. Faster than that since this is a specialised API but still ...

Seems to do okay getting to the figure or table (3)

Results are hit and miss. Here I ask it find the last sentence of a paywalled wiley journal, you can see from the json query and response it does find the paper. But it struggles to find the chunk with the last sentence. (2)

I now see the same thing done for gen ai. I attended this cnorhunourse by consultant that flashed pictures & we had to guess real or ai generated in a quiz. Someone inevitably won but nothing was said on how to figure it out (3)

half a dozen of probs with this exercise, least the librarian never did any follow up to explain the point of exercise. You couldn't tell with 30s of looking unless you had domain knowledge. But there was a fairly easy way to pass this with high accuracy. Just check if title existed in database(2)

I always thought this was obvious. Reminds me years ago where I saw a information literacy librarian "teach about predatory journals" by showing first page of real & fake scigen generated papers and asking students to decide within 30s if they were real or fake. (1)

The cynical me would say this has always been the case even before "ai" ha

Shrugs never even heard of the 2 things you mentioned. Creating a huge comprehensive search index is hard. Even chatgpt search etc had to mostly use existing search indexes

I guess this time some librarians think 'ai' is all useless & publishers are adding rubbish and expecting librarians to excuse the issues?

The other issue is there seem to be some push back from librarians from publishers who say these tools are not perfect and librarians can play educator roles. I am a bit confused by the pushback because that has always been the case for all tech. (4)

I'm not saying don't teach chatgpt to do literature review (particularly if it can use MCP servers like Wiley, pubmed). But librarians can cater also for the specialists who need powerful tools. Eg I get faculty telling me they not happy with perplexity/gpt deep research... (3)

General tools usage has always been more than specialised eg Google> Google scholar> Scopus. More important q is - for specialists do more specialised tools actually perform better? Do you gain moving from openai/Gemini deep research to elicit/Undermind etc. I think answer for now is clearly yes(2)