Twenty questions on distinguishing the waves of AI, from Turing to reasoning models.
01What did Turing propose as a substitute for 'can machines think?'
- A. A definition of thinking
- B. The imitation game
- C. The Turing machine
- D. A neural network
Reveal answer
B. Turing replaced a philosophical question with an operational test.
02What year was the Dartmouth Conference?
- A. 1949
- B. 1956
- C. 1965
- D. 1972
Reveal answer
B. The Dartmouth Summer Research Project was held in 1956.
03Which report triggered the first AI winter?
- A. Lighthill Report
- B. Fifth Generation Report
- C. Turing Report
- D. Nilsson Report
Reveal answer
A. Lighthill's 1973 report to the UK Science Research Council cut UK AI funding.
04What caused the second AI winter?
- A. Deep learning failed
- B. Expert systems and LISP machines collapsed commercially
- C. Perceptrons were disproven
- D. Government funding was cut
Reveal answer
B. The expert-systems industry and LISP-machine market both collapsed around 1987.
05What did AlexNet demonstrate in 2012?
- A. Symbolic AI works
- B. Deep CNNs decisively beat prior methods on ImageNet
- C. Reinforcement learning solves Go
- D. Transformers work for translation
Reveal answer
B. AlexNet's ~11-point top-5 error gap ended the debate about deep learning.
06What did the transformer remove?
- A. Attention
- B. Recurrence and convolution
- C. Backpropagation
- D. Softmax
Reveal answer
B. The paper's title is a hint: attention is all you need — no RNNs, no CNNs.
07What is a scaling law?
- A. A regulation on AI compute
- B. An empirical power-law relationship between loss and compute/data/parameters
- C. A hardware constraint
- D. A safety principle
Reveal answer
B. Kaplan et al. 2020 documented smooth power-law scaling of loss.
08What does RLHF stand for?
- A. Reinforcement Learning from Human Feedback
- B. Recursive Learning with Human Fine-tuning
- C. Random Learning for Human Formats
- D. Rule-based Learning for Human Fidelity
Reveal answer
A. RLHF trains a reward model on human comparisons, then optimizes against it.
09What does 'inference-time scaling' mean?
- A. Scaling GPUs at inference
- B. Letting the model think longer at inference to improve answers
- C. Batching more requests
- D. Distilling models for speed
Reveal answer
B. o1 and successors trade inference latency for capability by generating long reasoning traces.
10Which of these is a Chinese frontier lab as of 2026?
- A. DeepSeek
- B. Anthropic
- C. Mistral
- D. Cohere
Reveal answer
A. DeepSeek is the most prominent frontier Chinese lab, known for MoE and open-weight releases.