I’ll let you know when I get a fresh batch from back home!

Belgian here, we clearly need to talk and get rid of this misconception 😂

@itaiyanai.bsky.social make sure to have a look at this AI result of your Gorilla exercise ;)

Clearly LLMs are not as intelligent as we think they are. Do not let them gaslight you into fancy "underlying statistical principles" and use your own pair of eyes! Full conversation with Gemini can be re-read here: gemini.google.com/share/b64eec...
#EconSky 📉📈

I ended with "I think the data is fake and generated to look like a gorilla", which it again refuted more or less by stating that this could be true, but that we should not ignore the underlying statistical principles 🫠
#EconSky 📉📈

I then clarified that "I am not referring to the incident. I see in fact a gorilla in the scatterplot" and it ... gaslighted me into seeing things that are not there, as an example of Apophenia 😆😆😆 Again turning to statistical patterns as an explanation!
#EconSky 📉📈

First, it blamed me for referencing a well-known controversial incident regarding an older version of Google's image recognition and labeling technology, which mistakenly applied the label "gorilla" to images of African-Americans. 🤯
#EconSky 📉📈

When I then inquired "anything else you see in that plot?" and "nothings special beyond that?" I only got back some statistical mumbo jumbo that does not make sense once you have seen the gorilla ... so I shouted "DO YOU SEE THE GORILLA???" and not it gets interesting 🤓
#EconSky 📉📈

Follow up questions "can you do some interesting analysis", accepting to do statistical tests and "what about correlations" finally lead to the point where it suggested to plot (bingo! I thought) the data #EconSky 📉📈

Given widespread LLM usage by students, I decided to ask Google Gemini for help (free version, 2.5 flash). When I uploaded the datasets and asked "What can you conclude from these datasets?" it did not do much beyond simple (incorrect) summaries #EconSky 📉📈

Coincidentally, Claude Pro by @anthropic.com is free for one month! See details at claude.com/offers/oct-2...

For those interested, slides will appear here soon: eaae.org/PastEvents.a...

I usually do it like that too. Coincidentally, I wrote a referee report *while* reading a paper yesterday evening and I must say that it saved me a lot of time