A.I. cognitive abilities

9 flashcards · Other · by githubkoma

1. LLMs are highly sensitive to small prompt changes.

   hint: ⌨️

2. Training data contamination is a key risk

   hint: ☣️

3. Machine psychology must learn from psychology’s reproducability crisis

   hint: ⚠️

4. Faux Pas Performance could be explained by hyperconservatism aquired in finetuning phase

   hint: 🪛

5. GPT models show only minimal performance drop, because doesnt suffer from cognitive load impairments.

   hint: 🤯

6. In Theory of Mind Tasks GPT performs good except in faux pas tasks

   hint: 1. 👍/👎 2. French Word

7. In Problem Solving Tasks Student performance drops as task complexity rises

   hint: ⬆️/⬇️

8. Questions • What are LLMs highly sensitive to? • What is a risk about training data? • Was/is there a problem with psychological studies? • How does GPT perform in In Theory of Mind Tasks? • How do students perform in Problem Solving tasks compared to A.I.? • Which major psychological limitation does A.I. not have in comparison to Humans? (regarding problem solving tasks)

   None required.