| import {jewishLessons} from './jewish-content.js?v=study-reader-44'; |
|
|
| export const guides={ |
| cinema:{title:'How can a film produce a brain prediction?',steps:[['Input','A selected clip supplies visual features and, for the prepared collection, audio features. The note below the player identifies which inputs the saved prediction used.'],['Model','TRIBE converts those features into estimates of fMRI responses across a standard cortical surface. This is a spatial map of predicted measurements, not a diagram of individual firing neurons.'],['Read the result','Pause on a moment and use the response legend. Muting playback changes what you hear; it does not remove audio from the saved model prediction. Different videos also differ in their objects, movement, and speech.']],brain:'Seeing and hearing involve distributed systems. Occipital and temporal lobes are useful landmarks, but a film engages more than one region.',try:['Watch the default cat clip with its soundtrack.','Mute playback and explain why the saved map stays the same.','Choose another clip and compare a moment. Describe differences without assuming that a mood label caused them.'],limits:'The map is an AI prediction for an average subject. Mood labels describe the selected clips; the model does not read their feelings. A new picture-only prediction requires preparing a clip with Include its soundtrack turned off.',question:'What changes when you press Mute playback?',answers:['The saved prediction is recomputed without audio.','Only playback sound changes; the saved prediction stays the same.','The viewer must stop feeling an emotion.'],correct:1,feedback:'Playback controls affect what you hear. The saved model output uses the inputs listed below the player.'}, |
| mindeye:{title:'From recorded brain responses back to a picture',steps:[['Input','A volunteer viewed photographs in an MRI scanner. An fMRI response is a slow blood-oxygen-related measurement, summarized here as a vector of preprocessed voxel values.'],['Two activities','Guess the picture uses the authors’ published reconstructions. New fMRI decode feeds an authorized participant-1 response vector through the installed structural decoder and VAE to create a fresh coarse image.'],['Read the result','Compare overall colors, shapes, and positions before looking for fine details. A plausible picture can contain details that the person did not see.']],brain:'Visual information passes through many processing stages. MindEye2 learns a statistical mapping between recorded patterns and image representations; it does not recover an exact photograph stored inside a brain.',try:['In Guess the picture, predict the original before revealing it.','Keep the same case while switching between one and forty hours of participant training.','Run a recorded example in New fMRI decode and compare its coarse output with the matched original.'],limits:'Published refined images and newly generated coarse images come from different pipeline stages. Human matching scores measure your guesses. They are not model accuracy or evidence of thought reading.',question:'Why is the new output much blurrier than the research replay?',answers:['The app secretly chooses a different original.','A volunteer looked at a blurred picture.','New runs use the structural decoder; the replay includes published refinement.'],correct:2,feedback:'The fresh local branch builds coarse structure. The authors’ published replay also includes later refinement stages.'}, |
| mario:{title:'What does it mean for Mario to learn?',steps:[['Observe','Each learner sees a compact description of nearby tiles, pits, enemies, speed, and jumping state. It does not see the future or receive a hand-written jump plan.'],['Act and get feedback','The agent tries one of six actions for a few game ticks. Forward progress, coins, and the flag earn reward; a fall or collision can lose reward. Each agent plays its own copy of the real downloaded game engine.'],['Update','Tabular Q-learning revises an estimate for a situation and action. All agents share the table, so one agent’s experience can influence the others’ later decisions.']],brain:'Reward learning and action selection are useful connections to the basal ganglia. This Q table is a software learning rule, not a model of dopamine molecules or a reconstruction of human brain activity.',try:['Test the empty policy and note its progress.','Train 20 rounds, then test again with exploration off on the current level.','Reset and compare one agent with thirty-two. Keep the course and rewards fixed; distinguish rounds from total experiences.'],limits:'A successful training attempt may include random exploration. The clean policy test disables exploration and learning. Finishing Jump School is not evidence that the agent can finish every original Mario level.',question:'Why test with exploration turned off?',answers:['To measure the learned policy without random exploratory actions.','To erase the agent’s learning.','To switch on a scripted solution.'],correct:0,feedback:'Testing uses the learned action values directly and does not update them. Training results can include lucky exploratory actions.'}, |
| racer:{title:'How the fleet learns to follow your track',steps:[['The environment','The line you draw becomes the center of a circuit. Your first point defines the start. Each car receives sensor information about the road and its own motion.'],['Shared experience','Every car explores independently. Progress earns reward and leaving the road ends an attempt. Experience enters a replay buffer, and Double DQN updates the shared neural policy.'],['Measure learning','Watch recent attempt progress and completed laps. Then test one driver with exploration off. A large crash count alone does not show that learning has improved.']],brain:'The demonstration connects perception, action selection, memory for experience, and movement. The module switches represent software dependencies; disabling one is not a biological lesion experiment.',try:['Use the oval and test before training.','Train until several attempts finish, then test from the same start.','Reset and compare a narrow road with a wider road. Keep population and training experiences comparable.'],limits:'A best lap is a selected attempt, not an average success rate. A policy learned on one drawn circuit may not transfer to another. Starting a new track creates a fresh learner.',question:'What is the clearest sign of a better learned driver?',answers:['More crashes accumulated.','Higher progress in a comparable test with exploration off.','More colorful cars on screen.'],correct:1,feedback:'Use comparable tests and training budgets. A selected lucky attempt or a larger crash counter does not establish improvement.'}, |
| garden:{title:'Watch a tiny network learn your pixel patterns',steps:[['Make an example','Switch pixels on or off and assign the pattern a class. A class is the label you want the network to learn; the picture is its input.'],['Train','The network predicts a class, compares that prediction with the label, and adjusts connection weights to reduce its error. The garden displays the actual small network’s weights and activations.'],['Try a variation','Change part of a learned pattern and test it before adding the new example. That separates recognizing a variation from merely remembering an example it has already seen.']],brain:'Synapses are connections between neurons. Real plasticity involves many biological processes. Weight changes in this small classifier are a teaching analogy, not a simulation of a biological synapse growing.',try:['Teach two clearly different patterns with different labels.','Test a slightly changed version before training on it.','Deliberately give the same pattern conflicting labels and watch how predictions become harder to interpret.'],limits:'Training performance concerns the examples used for updates. To test generalization, use a pattern that has not been trained. Brightness is a display of the artificial model, not measured biological activity.',question:'Which pattern best tests generalization?',answers:['The exact pattern you trained on repeatedly.','A new variation that has not been used for training.','A blank canvas after you reset the model.'],correct:1,feedback:'Generalization means applying what was learned to a new example. A trained example mainly tests recall of that training input.'}, |
| pictionary:{title:'How a drawing becomes a guess',steps:[['Pixels','Your strokes become an image. The visual model proposes names freely, so its possible answers are not limited to a small built-in list.'],['Hypotheses','Qwen proposes interpretations and Perplexity Decider compares and refines them. A language hint adds another source of evidence. Uncertainty can change as you add a distinctive stroke.'],['Feature-map view','The animated layer tiles are illustrative filtered versions of your drawing. They show the idea of edges and features; they are not internal activation maps extracted from the remote models.']],brain:'The visual system builds useful representations from sensory input. Language and prior knowledge influence interpretation too. A neural network layer is not a one-to-one substitute for an anatomical brain region.',try:['Draw an ambiguous circle and predict the first guess.','Add one distinctive detail and request another guess.','Compare the same drawing with and without a written hint. Notice that the hint can help or bias the model.'],limits:'A plausible answer can still be wrong. API model scores and displayed hypotheses are not calibrated probabilities of a person’s intent. The feature-map visualization is explicitly illustrative.',question:'Are the displayed feature tiles Qwen’s real internal activations?',answers:['Yes, every tile is a neuron from Qwen.','Only when the answer is correct.','No; they illustrate feature extraction from the drawing.'],correct:2,feedback:'The tiles are an explanatory visualization. The recognition request is real, but this app does not receive hidden layer activations from the API.'}, |
| arcade:{title:'Why do these neurons spike?',steps:[['Input','Game state values enter a small trained spiking network. These describe positions and motion needed to choose a paddle action.'],['Integrate and fire','A leaky integrate-and-fire unit accumulates input. Its membrane value decays over time; crossing the threshold emits a spike and resets the unit. The plots show actual model traces.'],['Action','A learned target-position readout guides the AI paddle. A controller limits acceleration, filters target changes, and holds a small stopping zone to avoid chatter. The model was trained against offline interception targets; that teacher is absent during play.']],brain:'Biological neurons communicate partly through spikes. This model captures only a simple threshold-and-decay idea and omits most cell biology and network complexity.',try:['Play with the default threshold and look at the spike raster.','Raise the threshold and predict how firing will change.','Press Restore optimal defaults, repeat a similar game situation, and compare both firing and paddle behavior.'],limits:'More spikes do not automatically mean better play. Changing the threshold changes an already trained model, so it may move outside the conditions it learned under.',question:'What happens when an identical input crosses a higher firing threshold?',answers:['It generally becomes harder to emit a spike.','The neuron must emit more spikes.','The game automatically retrains itself.'],correct:0,feedback:'With the same inputs, a higher threshold requires more accumulated membrane potential. The effect on play depends on the rest of the system.'}, |
| plasticity:{title:'Can a network adapt after a pathway is removed?',steps:[['Before','Test the trained network and record its predictions. This gives you a baseline before changing any pathways.'],['Disrupt','Disable a neuron or connection. A hard mask removes its contribution. Test again to see which patterns are affected by that particular change.'],['Adapt','Retrain the surviving paths. Compare the immediate disruption with the result after further learning. Recovery is an observed outcome; it is not guaranteed.']],brain:'Plasticity is the capacity of nervous systems to change with experience. This puzzle shows a small artificial network adapting under a constraint. Human brain injury and rehabilitation involve much more than a connection mask.',try:['Choose a pattern and note its baseline prediction.','Remove one important-looking path, then test before retraining.','Retrain the surviving network and check both that pattern and a different one.'],limits:'Changes in this toy network do not predict the effects of injury to a person. A recovered training example does not demonstrate recovery on every new example.',question:'When should you measure the immediate effect of removing a connection?',answers:['Only after extensive retraining.','Before retraining the surviving paths.','After resetting everything.'],correct:1,feedback:'Measure before further training to separate disruption from adaptation. Then compare how additional learning changes the result.'}, |
| attention:{title:'Automatic reading, selective attention, and working memory',steps:[['Color clash','You select ink color while ignoring word meaning. Matching and clashing trials are mixed. Responses are timed locally, and correct-response medians are reported separately.'],['Memory sequence','You maintain a sequence of locations over a short delay and reproduce it in order. Extra flashes, slower presentation, and replay change the task rather than revealing a fixed “brain score.”'],['Interpret carefully','Look at speed and errors together. Faster answers with many mistakes may not mean better attention. A short run can vary substantially from one attempt to the next.']],brain:'Attention involves selecting information relevant to a goal. Working memory supports keeping information available for a short task. Both depend on interacting systems rather than a single isolated storage box.',try:['Before Color clash, predict whether matching or clashing words will be faster.','Finish a full round and compare correct-response medians and errors.','In Memory sequence, compare two flash speeds. Mark when replay helped instead of mixing assisted and unassisted trials.'],limits:'These are small classroom demonstrations with browser timing, practice effects, and small samples. They do not measure IQ or diagnose attention or memory conditions.',question:'A faster median with many more errors most directly suggests what?',answers:['A possible speed–accuracy tradeoff.','A confirmed increase in intelligence.','A diagnosis from the browser.'],correct:0,feedback:'Speed and accuracy should be considered together. Short browser tasks are demonstrations, not diagnostic assessments.'}, |
| brain:{title:'How to read this 3D brain',steps:[['Start outside','Rotate the folded cortical surface and identify front, back, top, and the two hemispheres. Lobe colors are broad teaching regions, not exact functional borders.'],['Look inside','Fade the cortex, split the hemispheres, or remove half the surface. Select a deeper structure and compare its position with surrounding landmarks. Internal shapes are schematic.'],['Connect parts','Follow an illustrated signal route, then use Find a part to practice locating regions. The animation explains a possible processing sequence; it is not recorded neural activity.']],brain:'Functions such as seeing, acting, and remembering are distributed across interacting networks. A region can contribute to several functions, and one function can involve many regions.',try:['Find the occipital lobe at the back, then the frontal lobe at the front.','Open the brain and locate the hippocampus, thalamus, and corpus callosum.','Try a location challenge with labels hidden. Explain the function before revealing the answer.'],limits:'The cortex is an anatomical standard surface. Lobe boundaries, locations of functional markers, internal structures, and animated routes are schematic educational guides, not a clinical atlas.',question:'Does highlighting one area mean that it performs its function alone?',answers:['Yes, every function belongs to exactly one area.','Only on the left hemisphere.','No; many functions rely on interacting networks.'],correct:2,feedback:'Region names are useful landmarks. They do not mean that complex functions are isolated in single boxes.'}, |
| drugs:{title:'Different ways to change a chemical message',steps:[['Normal signaling','A cell releases messenger molecules into a synaptic gap. Molecules can bind receptors, move away, or be taken back up. Compare the normal panel with the changed panel.'],['Selected mechanism','Some substances block receptors, some activate them, and some change release or recycling. Read the selected substance’s explanation before interpreting the animation.'],['Make a prediction','Pause, change one setting, predict the visible effect, and replay. The intensity slider only changes a dimensionless teaching animation.']],brain:'Chemical signaling depends on receptors, cell types, location, and time. A single substance can act through multiple pathways. A simple synapse scene can explain one mechanism without predicting a whole person’s response.',try:['Pick a substance and identify whether the illustrated effect concerns release, binding, or clearance.','Pause and predict what will accumulate or flash differently.','Replay the comparison, then remove the substance to restore the common baseline.'],limits:'The animation is not a dose calculator, a clinical simulation, or guidance on using substances. Real effects and timing cannot be read from the animation intensity.',question:'What does the animation-intensity slider represent?',answers:['A safe dose for any person.','A dimensionless setting in a teaching animation.','A measured blood concentration.'],correct:1,feedback:'It only controls the illustration. It does not represent a real dose, concentration, or individual effect.'}, |
| anatomy:{title:'Use the labels as a connected map',steps:[['Locate','Use the search box to find a region. Check whether the label refers to an exterior area, a deeper structure, or a broad system that contains multiple structures.'],['Explain','Read the biological role before the software analogy. The analogy explains this app’s design choices; it is not an anatomical definition.'],['Practice','Pick a function, name two structures that could contribute to it, and locate both in the 3D explorer.']],brain:'Anatomical names describe regions and structures at different scales. For example, the cortex includes lobes, and broader systems can contain several individually labeled structures.',try:['Search for a region related to memory.','Explain its role in your own words without reading the robot analogy.','Find the same structure in the 3D brain and compare its neighbors.'],limits:'This is an introductory checklist with schematic locations. Complex functions are distributed and do not map exactly onto software modules.',question:'Should the robot analogy replace the biological description?',answers:['Yes, they mean exactly the same thing.','No; the analogy helps explain the engineered activity.','Only when a module is switched off.'],correct:1,feedback:'Keep the real biological role separate from the app’s simplified software analogy.'} |
| }; |
|
|
| Object.assign(guides,{ |
| flappy:{title:'How does a flock learn the timing of a flap?',steps:[['Observe','Each bird receives its height and vertical velocity, both edges of the next pipe, both gap boundaries, and the following pipe and gap positions. These describe the game state; the agent is not learning to see pixels.'],['Try and update','The learner chooses Coast or Flap. Gravity and collisions run on every physics tick. Surviving, passing pipes, and changing gap alignment affect the reward; crashes lose reward. Each three-tick action supplies a real Q-learning transition.'],['Compare','All birds share updated estimates, but each has its own flight attempt. The best training bird may include lucky random actions. Test on four layouts outside the training sequence with exploration off to check the learned policy.']],brain:'This connects sensory information, movement timing, and reward-guided action selection. The learned action values are part of a software learning rule; it does not model a bird’s nervous system or establish that one brain region acts alone.',try:['Test the blank policy before training. Note the four layout scores.','Train 10,000 ticks, then repeat the same test. Compare all four scores rather than choosing only the best bird.','Reset and compare low exploration with high exploration using the same difficulty and number of agents.'],limits:'This classroom adaptation supports endless pipes or a chosen pipe count and uses rectangular collision checks. It is not the original mobile binary. A finite test set illustrates performance but does not establish reliability on every possible pipe sequence.',question:'A bird passed many pipes during training. What should you check next?',answers:['The learned policy on new layouts with random exploration off.','Only the single luckiest attempt.','Whether more birds are drawn on screen.'],correct:0,feedback:'A clean test separates learned behavior from lucky exploratory moves and checks layouts absent from the training sequence.'}, |
| crossy:{title:'Why waiting can be a learned action',steps:[['Observe','The chicken gets its column, local traffic flags, and actual nearby car positions, speed, and length in the current and next two lanes. The flags describe blocked cells now and one hop ahead using visible car velocity. This representation makes traffic information available; it does not choose an action for the learner.'],['Choose a move','Wait, Forward, Left, Right, and Back are all available. A hop takes eight physics ticks. Cars continue moving, and collision checks happen during the hop. Reaching a new row earns reward; being hit, invalid moves, and excessive waiting lose reward.'],['Test the timing','Different training rounds use different traffic layouts. The shared table can learn that immediate movement is not always worth the risk. A four-layout test turns exploration and learning off and reports every result.']],brain:'Attention to motion, action selection, and timing contribute to everyday decisions such as crossing a road. This is a game with engineered sensors, not a simulation of human road safety or a guide for crossing real roads.',try:['Predict what the blank policy will do, then test it.','Train 10,000 ticks and compare the four test layouts. Use average distance for endless mode, or check distance and finishes for a fixed course.','Try the tighter-traffic setting with a new learner. Keep the population and training budget comparable.'],limits:'This is an independent Crossy Road-style game with original voxel models. Choose an endless road or a fixed length. Tests stop after at most 6,000 ticks so an endless test cannot run forever. It includes roads, grass, cars, and trees; rivers and trains are outside this version. A selected training success is different from a general success rate.',question:'Why might Wait be useful even though new rows earn reward?',answers:['Waiting automatically freezes the cars.','A short wait can avoid a collision before the next hop.','The policy is following a hidden scripted route.'],correct:1,feedback:'Traffic keeps moving. Waiting can make a future hop safer, but waiting too long also ends the attempt.'} |
| }); |
|
|
|
|
| Object.assign(guides,{ |
| 'neuron-signals':{title:'How input becomes a model spike',steps:[['Combine inputs','Excitatory drive raises the model’s target voltage, while inhibitory drive opposes it. A steady input and a repeated pulse train supply different time patterns.'],['Integrate and leak','The membrane voltage moves over time rather than jumping immediately to its target. The time constant controls how quickly it responds; leak draws it back toward rest after input falls.'],['Fire and recover','Crossing threshold records a spike, resets the membrane, and starts the recovery interval. The trace shows the calculated voltage and marks spike times.']],try:['With excitation 18 and inhibition 0, run 500 ms and record the spike count.','Set inhibition to 18 and run another 500 ms. Compare the two parts of the trace.','Reset the trace, select pulsed input, and predict the effect before running again.'],limits:'These are model variables. The spike marks and schematic axon are simplified displays, not cell recordings.',question:'If equal excitatory and inhibitory drives cancel in this model, what should you expect?',answers:['More firing because there are more inputs.','Voltage tends toward rest and sustained firing stops.','The app starts training a different neuron.'],correct:1,feedback:'The drives oppose one another in the equation. With no net drive, leak draws voltage back toward the resting level.'}, |
| 'blind-spot':{title:'How to investigate a gap in visual input',steps:[['Hold fixation','The + stays at a fixed point on the board. Looking at it with the selected eye helps keep gaze direction consistent.'],['Vary the target geometry','Dot spacing and your viewing distance change where the dot appears in your visual field. The app leaves the dot on the page at all times.'],['Record observations','The buttons record what you report along with the current eye, spacing, diameter, and line setting. These records describe the observation, not an automated detection of your vision.']],try:['Choose the right eye, close the left, and hold your gaze on the +.','Slowly adjust spacing or viewing distance. Log only what you actually notice.','Repeat with a line through the dot, or with the other eye. Keep other conditions as similar as possible.'],limits:'The display is not calibrated for retinal size, and the app cannot know where you are looking.',question:'When the dot disappears from your awareness, what has the page done?',answers:['It has made the dot transparent.','It has detected a change in your brain scan.','It has kept the dot visible; the change is in your perception.'],correct:2,feedback:'This activity never hides the target. The observation depends on gaze, one-eye viewing, and the target’s position in the visual field.'}, |
| 'prediction-learning':{title:'How an expectation changes after an outcome',steps:[['Predict','Add up the learned values of the cues present on this trial. This is the model’s expectation before it sees the outcome.'],['Compare','Subtract that expectation from the selected outcome, 0 or 1. An unexpected outcome gives a positive error; an expected but omitted outcome gives a negative error.'],['Update active cues','Multiply the error by the learning rate and update only the cues that were present. A cue that already predicts the outcome can leave little error for a new cue to learn from.']],try:['Press Teach A and inspect its value and the trial log.','Press Omit the outcome and compare the new value and negative errors.','Run the blocking preset and compare A with B. Explain why B learned little despite appearing with the outcome.'],limits:'This cue-association rule has no action selection or biological dopamine cells. Its declining cue value is a model result, not a claim that a human memory has been erased.',question:'Why does B gain little value when it is added after A already predicts the outcome?',answers:['B creates little additional prediction error.','The model has a rule that permanently forbids B from learning.','No outcome is supplied during the compound trials.'],correct:0,feedback:'A already accounts for almost all of the outcome, so little shared error remains to change either cue.'} |
| }); |
| for(const lesson of jewishLessons){ |
| guides[lesson.id]={ |
| title:'How to read this source sheet', |
| steps:[['Read the named source','Begin with the fuller Hebrew or Aramaic passage and its reference. Read the English guide as a labeled paraphrase, then use the source link to check wording or wider context.'],['Examine the interpretation','The questions beside each passage help you identify evidence, compare readings, and explain your reasoning. Keep the author’s claim distinct from your own example.'],['Make a comparison','Use the Brain connection tab to explore a modern learning or reasoning idea. Then return to the text: a helpful comparison should not replace the passage’s own context.']], |
| try:['Read one source, then explain its point without looking.','Return to the passage and mark one detail your explanation missed.','Use the source-sheet notes to respond: '+lesson.reflection], |
| limits:'These are textual and cognitive comparisons. They do not establish that religious or philosophical terms name particular brain regions.',question:lesson.check.question,answers:lesson.check.answers,correct:lesson.check.correct,feedback:lesson.check.feedback, |
| primarySources:lesson.sources.map(s=>({title:s.reference,url:s.url})) |
| }; |
| } |
| guides.mario.try=['Choose one numbered practice level and create a new learner so the course stays fixed.','Test without exploration and record progress before training.','Train 20 rounds, then repeat the same test. Compare progress, finishes, and the amount of experience.']; |
|
|
| guides['brain-circuit']={title:'How a signal passes through the model circuit',steps:[['Drive the input unit','The input drive affects the first model neuron. Its threshold crossings record spikes and schedule events for the next unit.'],['Deliver delayed pulses','After the selected connection delay, an event adds a voltage pulse to the next enabled neuron. It can fire immediately if the threshold is crossed, unless it is recovering.'],['Compare model and anatomy','Each unit has a regional teaching label. Selecting it highlights a landmark in the 3D brain and explains the biological role separately from the simpler model role.']],try:['Run a 500 ms baseline and note the output-unit spikes.','Add inhibition of 20 at Visual processing, reset the trace, and run the same interval. Compare upstream and downstream counts.','Restore inhibition to 0, disable the Visual processing model neuron, reset, and repeat. Then restore all stages.'],limits:'This six-unit, feed-forward chain does not represent a complete brain circuit. Connection parameters are not fitted physiological measurements, and regional labels do not mean that one neuron performs a region’s functions.',question:'What does disabling a model stage let you conclude?',answers:['The model’s downstream signaling depends on this artificial stage.','A human with a change in that brain region would show exactly this output.','The remaining neurons have learned a new route automatically.'],correct:0,feedback:'The result concerns this particular model chain. There are no alternate pathways or learning rule in this circuit, and it is not a model of a human lesion.'}; |
|
|
