Answers

How models work

What is actually inside the models everybody uses, taken one idea at a time.

1. Why does the model get stuck repeating the same sentence over and over?The model is not stuck in a memory loop: once it has written a phrase, that phrase is in its context and makes writing it again more likely, so a decoder that always picks the top word cannot leave.
2. Why does diffusion add Gaussian noise instead of any random corruption?Diffusion adds Gaussian noise because Gaussian errors combine exactly: each step keeps the spread fixed and gives a closed-form noisy picture at any later step, which makes the reverse step learnable.
3. Why does the same request cost more tokens in some languages than in others?A token is learned from a corpus, so a language that was rare in the training text is split into many more pieces for the same meaning, and the price multiplier follows.
4. How do you turn the predicted noise into the mean of the previous image?Subtract a small scheduled fraction of the predicted noise from the current image, then rescale to undo the forward step's shrinkage.
5. Why does the forward noising process have to be a Markov chain?The forward process is made memoryless so that each step is a fixed random rule and any later state can be computed straight from the original picture.
6. Does it matter which layers I attach adapters to?Yes. For the same number of trainable parameters, adapters spread over more weight matrices, including the feed-forward ones, usually score higher than adapters concentrated on a few.
7. What do multiple attention heads actually give you that one big head doesn't?Several heads let a model track several relationships at once, each in its own lower-dimensional subspace, instead of squeezing every relationship into one weighted average that destroys information.
8. Why does a transformer need a cache that grows with sequence length?A transformer stores the key and value for every token it has already seen so later tokens can look back at them, and that store grows with the conversation.
9. Why do we divide the attention scores by the square root of the key dimension?The square root of the key dimension is the typical size of an unscaled attention score, so dividing by it stops softmax from becoming a hard pick-one.
10. Why is the REINFORCE gradient so noisy even though it is unbiased?Because unbiased only fixes the average, while one noisy return is multiplied by every action's score, so the estimate's spread grows with episode length and reward scale.
11. Why do tokenizers work on bytes instead of on characters?Tokenizers use bytes because 256 byte values can represent any text, while Unicode has too many characters to include safely.
12. What is actually new in a neural network compared with linear regression?Only three things are new: the slopes are computed by working backwards, the score is no longer a bowl, and the model builds its own inputs. The prediction, the loss, the step and the held out check are unchanged.
13. How many principal components should I keep?There is no single correct number: keep the smallest number whose combined variation you can afford to lose, and report the share you discarded.
14. Why can a completely made-up answer come with high confidence?Because the confidence score measures how likely the words are, and a fluent invention is built from likely words, not because the fact was checked.
15. What is the KL penalty against the original policy actually protecting against?It stops the policy from chasing high scores in parts of the answer space where the fitted reward is only guessing, by charging it for moving away from the model it started from.
16. Why can a linear recurrence be computed as a convolution?Because a linear recurrence adds up past inputs with fixed decaying weights, and that weighted sum is a convolution with the recurrence's impulse response.
17. What do you give up when you pool or downsample a feature map?You give up exact position: each block of the map becomes one number, so later layers see a coarser grid, gain reach, and lose the ability to say where a response came from.
18. What does it mean to use stride 2 in a convolution?Using stride 2 keeps the filter the same but moves its window two positions at a time, so it produces about half as many outputs in each direction and about a quarter of the work in two dimensions.
19. Why do CNNs use the same weights everywhere instead of learning a separate filter for each spot?They share one small filter across every position because a pattern such as an edge means the same thing wherever it appears; the huge cut in parameters is a consequence, not the main reason.
20. Why does a convolution make the image smaller if you don't pad it?A convolution shrinks the image because the window must fit completely inside it, so the first and last few positions have no full window. Padding adds extra values at the border so those positions can be used.
21. Why does logistic regression use a sigmoid instead of a straight line?The sigmoid turns an unbounded straight-line score into a probability between zero and one, and it pairs with the log score so fitting keeps the same gradient shape.
22. What is the difference between the inputs and the weights in a linear model?The input is supplied by each example and changes from row to row; the weights are the model's adjustable numbers, chosen once by fitting and shared by every example.
23. What does temperature actually do to the next-word probabilities?It divides the model's raw scores by a constant before they are turned into probabilities, so every ratio between two candidate words is pushed further apart or pulled closer together.
24. When a model says it is 90% sure, what is that 90% a probability of?It is the share of the model's score mass given to that option out of the options it was offered, not the chance that the answer is correct.
25. Why does attention use softmax instead of just dividing each score by the sum of all the scores?Dividing by the sum can give negative weights and fails when scores cancel; exponentiating first makes weights positive and turns score gaps into ratios.
26. Why does k-means split one long stretched-out group into two?Because k-means keeps only one centre per group, so a long group's ends end up closer to a second centre and the group is cut across its length.
27. How do I know if I should increase or decrease the slope to reduce the error?The sign of how fast the loss changes with the slope tells you: positive means lower the slope, negative means raise it.
28. Why do we need a forward pass before we can backpropagate?The gradient is built from local derivatives evaluated at the values the forward pass computed, so the loss alone cannot supply it.
29. When should I fine-tune a model instead of just prompting it better?Fine-tune only when the model has been told clearly what you want and still will not comply reliably; missing facts need retrieval, and unstated instructions need a better prompt.
30. How does masking actually make attention causal?The mask is applied to the scores before they become weights, so forbidden positions end up with exactly zero weight and the rest are renormalised automatically.
31. Why does attention need positional encoding if it already looks at the whole sequence at once?Attention treats tokens as an unordered set, so position has to be added to the vectors; otherwise a sentence and its anagram get the same representations.
32. Why does linear regression square the errors instead of just adding their absolute values?Squaring makes large errors count much more than small ones, so it picks a different line from absolute value; absolute value is a different choice about costs, not a mistake.
33. How can a model remember a sequence without keeping every past input?A model can remember a sequence by keeping a fixed-size block of numbers and rewriting it at every step from its old value and the new input.
34. How does a gate let information pass through unchanged?A gate lets old information take a direct additive route through the steps, with a learned fraction deciding how much survives each step.
35. Why is 'hello' a different token from ' hello' with a space in front?A space is a byte that gets merged onto the front of the next word, so the tokenizer stores ' hello' and 'hello' as separate symbols with no shared structure.
36. What does classifier-free guidance actually add together, and why does it make the prompt stronger?It adds the unconditional prediction to several times the gap between the conditional and unconditional predictions, pushing past the conditional answer so the prompt is obeyed more strictly.
37. Why is the model's noise prediction really the score of the noised image distribution?The network's noise prediction, divided by the noise level and given a minus sign, is the direction that makes a noised image more plausible.
38. If you ban certain tokens, does that change what the model would have said?Banning tokens leaves the model's raw scores alone but changes the probabilities you sample from, so the text can change.
39. Why does a bigger batch make the gradient less noisy?Each step's gradient is an average, and an average's error shrinks with the square root of the number of examples, so four times the batch only halves the noise.
40. Why do tree ensembles still beat neural networks on ordinary spreadsheet data?Because each spreadsheet column means something on its own, which is exactly what a one-column-at-a-time split needs; where inputs only mean something in combination, networks win by tens of points.
41. How can I tell whether my clusters are real or just something the algorithm made up?Check whether clusters survive refitting on different rows and whether they predict a held-out column; a convincing 2D plot is weak evidence because such plots can invent clusters.
42. Why does a model cost the same to run no matter what you ask it?In an ordinary model every weight runs on every token, so the arithmetic per token is fixed by the model size, not by how hard the question is.
43. How does a decision tree pick the exact number to split a column on?A decision tree sorts the column, tries a cutoff between every adjacent pair of distinct values, scores each one by how mixed the two resulting groups are, and keeps the best.
44. What is Gini impurity actually measuring?Gini impurity is the chance that two rows picked independently from a group have different outcomes. It is zero for a pure group and largest for an even split.
45. Why does changing a few rows of the training data give a completely different tree?A tree picks the single best split at each step, so a tiny data change can swap a near-tied split and reshape the whole tree; that is normal, not a bug.
46. Why does the value of a state have to satisfy a recursive equation with the next state's value?A state's value equals the reward from the next step plus the discounted value of wherever that step lands, because every possible future divides that way and averaging over the first step leaves exactly the next state's value.
47. Why doesn't temperature scaling change which answer the model picks?Temperature scaling changes how confident the model sounds, not which answer it ranks first, because dividing every score by the same positive number preserves their order.
48. What is actually stored in the state that a model carries forward?It is a fixed-size block of numbers rewritten at each step from its old value and the new input, not a stored copy of earlier inputs.
49. When should a model refuse to answer instead of guessing?Refuse when the chance of being wrong times the cost of an error is higher than the cost of handing the case to a person.
50. Why isn't picking the most likely word at each step the best way to write a sentence?Because the most likely word now can lead to a much worse sentence than a slightly less likely word that sets up a better continuation.
51. What is wrong with just tokenizing text one character at a time?Nothing is ever unknown, but a passage becomes roughly five times longer than it needs to be, and attention cost grows faster than length, so you pay several times over for the same text.
52. Why can't a language model just treat each whole word as one token?Because a fixed list of words cannot contain a word nobody has written yet, and enlarging the list never catches up, while the model still has to put a number to every input it is given.
53. Why do models fail at counting letters or handling numbers and code?Models often fail at letters, digits and code because the text is chopped into pieces before the model sees it, hiding individual characters and place values.
54. Why do tokenizers split words into smaller pieces instead of keeping whole words?Tokenisers split words into pieces because a fixed list of whole words is enormous and still cannot handle names, typos or new words, while common fragments can build any word without an unknown entry.
55. Why do so many tokens end up routed to the same few experts?Because picking and training reinforce each other: an expert chosen more often gets trained more, which gets it chosen more often again, so the traffic collapses onto a few.
56. What does it mean if two words are close together in embedding space?It means training placed their rows in similar positions because the pieces appeared in similar company, so one can often stand in for the other in text; it is not a direct synonym score.
57. Why can't you just add new tokens to a tokenizer after training?Because each token id is tied to weights the model has already trained, so changing the vocabulary makes old numbers name different pieces while the model keeps the old meanings.
58. Why does the same word get a different vector depending on the sentence it appears in?The lookup gives the same word the same starting row every time; the different vector is built later by mixing in the words around it.
59. How does DBSCAN decide that some points are noise and just leave them out?A point is left out when too few neighbours lie close to it and no dense point is close enough to attach it to a group.
60. What happens to the model if most of the training data comes from one source?The model becomes specialised to that source and loses ground on everything else, because a training mixture is a set of shares that must sum to one.
61. Why can't I just measure accuracy on data that has no labels?There is no accuracy to measure, because accuracy compares predictions with true answers and unlabelled data has none.
62. What does a convolution actually calculate at a single point in the image?It multiplies each filter weight by the pixel beneath it, adds those products, and writes the single total at the matching position in the output.
63. Why does PPO clip the probability ratio?PPO clips the probability ratio so the objective stops rewarding a policy for pushing an action further in the direction it already favours, while still penalising pushes the wrong way.
64. Why train on a loss function instead of just using accuracy?Because accuracy is a pass or fail count that is flat almost everywhere, so it gives the optimiser no slope to follow, while a loss changes smoothly and tells every weight which way to move.
65. How do human pairwise comparisons turn into a reward function?Human pairwise preferences become a reward by fitting a scoring model so that preferred responses score higher, then using that score with a penalty for drifting from the starting model.
66. Why can't I fit temperature on the same data I trained the model on?Because training predictions are already too confident, fitting temperature there confirms that confidence instead of correcting it.
67. Why did my loss suddenly spike and can I recover without restarting?Your loss spike is probably one bad batch, and you can usually recover by waiting or rolling back to a recent checkpoint instead of restarting the whole run.
68. Why does a tree that gets every training example right do worse on new data?A tree that fits every training row has memorised individual cases rather than learned a pattern, so its accuracy on new rows falls.
69. Why does training loss keep falling while validation loss starts rising?Training loss keeps falling because the model is memorising the batches it is shown; held-out loss rises when that memorisation stops helping on data it has not seen.
70. If it's all just English text, why does where it came from matter?Because the same English words are a record of who was in a position to write them; each source filters for speakers, registers and subjects before you touch it, and no quality filter undoes that.
71. Can a better model make up for a bad training set?No. A better model can reduce error on what the training set contains, but it cannot supply a skill or fact the set never demonstrates.

The other subjects