Machine Learning & AI

Question 1

9 breakthrough technologies 2022

Answer 1

1. unhackable Internet: entangled particles transmitted end to end, unable to be read w/o disrupting content
2. hyper-personalized medicine: takes large team to develop treatment for rare condition; solutions using digital speed
3. digital currency: has potential if backed by real currency
4. anti-aging drugs: senolytics - remove senescent cells that create low-level inflammation, creating toxicity, and also benefits of
5. AI-discovered molecules: 10^60 chemical molecules possible; use machine learning to speed up process of finding possible drugs
6. satellite mega-constellations: blanket the world with high speed internet or junk-ridden minefield
7. tiny AI: lower carbon emission; increases speed; increases privacy (local storage)
8. differential privacy: inject noise into user data to increase privacy
9. climate change attribution: improving techniques to link weather to climate change; disentangling factors

Question 2

adversarial inputs

Answer 2

• aim at misclassification in order to avaid detection
• e.g. - malicious documents designed to evade antivirus, and email attempting to evade spam filters
• mutated input: attackers actively minimize classifers detection rate using undetectable payloads. To develop detection systems
  
  1. limit information leakage: don’t provide error codes or confidence values
  2. limit probing: limit how many paylods they can test (e.g. captha)
  3. ensemble learning: combing variable detection methods

Question 3

self-attention

Answer 3

Question 4

AutoML

Answer 4

• attempt to make machine learning available to people without strong expertise in the field,

1. automate repetitive tasks which enables a data scientist to focus more on the problem rather than the models
2. automate data pipeline components - helps to avoid errors that might slip in with manual processes

Question 5

baseline models

Answer 5

• calculate
  
  • accuracy
  • efficiency
  • cost
• examples
  
  • naive bayes
  • linear regression
  • Markov Model (NLP)

Question 6

batch normalization

Answer 6

• N ~ (0, 1)
• this is where activation are the most dynamic
• do this to every layer
• exponential-smoothed average
• removes the need for bias
• need batch and running meaning and variance
• regularizes like noise injection because it adds error term to norm

Question 7

bayes rule

Answer 7

1. p(y|x)=p(x|y)p(y)/p(x)
   
   1. likelihood*prior/normalizer
2. prior p(y)=instances/total

Question 8

bidirectional RNN image classification

Answer 8

• transpose
  
  • vertical RNN and horizontal RNN
• global maxpool on both

Question 9

bidirectional RNN

Answer 9

• make another recurrent unit but read in reverse
• concat hidden to get size h_t 2M
  
  • h_t=[h^–>,h^]
• many to one case
  
  • output is O=[h_T^–>,h₁^]
  • could also take max
• is able to predict first item in sequence

Question 10

bidirectional RNN architecture

Answer 10

Question 11

black swan AI attack

Answer 11

• sooner or later, an attack will throw off your classifier

1. develop incident response process
   
   1. have necessary controls to delay or halt processing when debugging
   2. know who to call
2. use transfer learning to protect new products
   
   1. pretrained models or public datasets
3. leverage anomaly detection
   
   1. e.g. - abuse of free tier to mine data (changes of uses of platform)

Question 12

cognitive computing

Answer 12

• perform specific, human-like tasks in an intelligent way using machine learning
• simulate human thought processes using a computerized model
• imitate the way the human brain functions

Question 13

collaborative filtering

Answer 13

• user rank: s(i,j). = µi +∑wii’{rij-µi’}/∑|wii’|
• pearson correlation (correlation b/w variables) - basically cosine similarity
  
  • w_ii’=∑(x_i-µ(x)_i)(y_i-µ(y)_i)/√(∑(x_i-µ(x)_i)²∑(y_i-µ(y)_i)²)
  • Ψ_ii’: numerator is over rating by both users
  • Ψ_i, Ψ_i’: denominator is over respective user

Question 14

collaborative filtering formula

Answer 14

Question 15

concatenation trick

Answer 15

• the weights of the input and hiddens can be concatenated
• size Mx(M+D)

Question 16

convolution

Answer 16

• acts as a filter; blurring, edge detection, etc..
• denoted I*K(image convoluted with filter/kernel)
• a(t)*w(t)=∫a(τ)w(t-τ)dτ continuous or a[x]*w[x]=Σa[x2-x]w[x]
• signal and kernel interchangeable
• can replace x with x’
• filters are the pattern were looking for, spikes are occurance

Question 17

convolution versus cross-correlation

Answer 17

convolution: (f*g)[n] = ∑f[m]g[n-m]

cross-correlation: (f♦g)[n] = ∑f[m]g[n+m]

Question 18

convolutional neural network

Answer 18

• imitates the visual cortex in the brain
• invariant to shifts in features of image (translational invariance)
• convolution->pooling->convolution->pooling->fully connected layer
• 3d image has 3d filter
• multiply filter by input elementwise; pooling looses a dimension
• the features that are found actually resemble things in input (layer 1 shape, layer 2 body part, label 3 face)

Question 19

cross-entropy

Answer 19

• use for classification
• use for autoencoder
• entropy and variance are correlated to unpredictableness
• -T.mean(T.log(y[T.arange(y.shape[0]), t]))

Question 20

d(sigmoid(x))/dx

Answer 20

y*(1-y)

Question 21

d(tanh(x))/dx

Answer 21

1-y**2

Question 22

data lake/warehouse

Answer 22

• lake: large-scale pool of raw data without a concrete purpose
• data warehouse: repository for structured, filtered data that has already been processed for a specific purpose.
• data lakes were born out of the need to harness big data and benefit from the raw, granular structured and unstructured data for machine learning
• data warehouses for analytics used by business users.

Question 23

data pipeline

Answer 23

• encapsulate a number of processing steps required to prepare data for machine learning
• performing “data prep” operations such as cleansing data and handling missing data and outliers
• also transforming data into a form better suited for machine learning
• includes training or fitting a model and determining its accuracy
• automated so their steps may be performed on a continued basis

Question 24

data poisoning

Answer 24

• feeding training adversarial data to the classifier to polute training data to perform worse
• model skewing:
  
  • attackers pollute training data to shift learned boundary
  1. use sensible data sampling
     
     • don’t allow one user too much influence to system
     • use weight decay
  2. compare classifer to previous
     
     1. dark launch: compare two outputs on same traffic
     2. backtesting: A/B testing of fraction of traffic
  3. build golden standard dataset: classifier must accurately predicct
• feedback weaponization: use feedback systems to attack legitimate users and content
  
  1. verify feedback befor making decisions
  2. don’t assume benefactor is responsible: hackers cover tracks to penalize users

Question 25

deeper is better

Answer 25

less hidden units per layer and achieve better performance

Question 26

DeepMind wavenet

Answer 26

• text to speech
• uses CNN
• is used by Google Assistant

Question 27

dis/advantage full batch

Answer 27

• wont work on big dataset due to size
• maximize likelihood over entire set

Question 28

distribution ML pipeline tools

Answer 28

• Apache Spark
• Apache Airflow
• Kubeflow

Question 29

docker containers

Answer 29

• small, user-level virtual machine that helps data scientists build, install, and run code
• built from a script
• ability to version control a data science environment

Question 30

does data influence training time?

Answer 30

• training time does not depend on amount of data
• only the creation time increases

Question 31

dropout regularization

Answer 31

• drops random nodes during training
• hidden layer units rely on multiple input
• emulates ensembles
• approximates 2^(# neurons) networks
• MULTIPLY PREDICTION LAYERS BY P-KEEP AT END

Question 32

edge analytics

Answer 32

• data collection and analysis where an analytical computation is performed on data at the point of collection (e.g. a sensor) instead of waiting for the data to be sent back to a centralized data store
• IoT model of connected devices has become more established
• filter for what information is worth sending to a central data store for later use.

Question 33

ensemble network

Answer 33

• majority vote; better accuracy than one model; two methods (different features, same features)
• 100 features 1 million data points
• 10 networks of 10 features

Question 34

exponentially-smoothed average

Answer 34

• z(t)=(1-å)Y(t-1)+å*costs(t), (0 < å < 1)
• y(t) = z / (1 - decay ** t + 1)
• more recent values matter more
• thus better for non-stationary data

Question 35

face map

Answer 35

Face maps provide the ability to combine different types of relationships into one that you can do math on

• THEOREM is that if you combine relationships that are null or simpler, you reconstruct the larger system
  
  • Co-limits
• Validating: measure the low dimensional entropy and the high dimensional one
• Can iteratively combine low dimensional bad ones with high dimensional good ones until you fine the right one

Question 36

fancy method

Answer 36

• NLP task
• max pool hidden y in the RNN
• determine if some sequence is in the data
• has 1 output

Question 37

fastest AI

Question 38

gated recurrent unit

Answer 38

• similar to rate, you have to choose b/w
  
  • taking the old value
  • or taking the new value
• if reset gate r(t) = 0, its like beginning new sequence
  
  • but, h(t) will be a combo of h(t-1) and hhat(t)
  • has same form as update gate
  • is essentially a forgetting factor
• update gate weights previous
  
  • z(t) = act(x_tW_XZ + h_t-1W_hz + b_z)

Question 39

geospatial analytics

Answer 39

• handle geographic information system (GIS) data (e.g. GPS data) and imagery (e.g. satellite photographs)
• uses geographic coordinates as well as identifier variables such as street address and zip code
• create geographic models and data visualizations for more accurate modeling and predictions.

Question 40

GPU acceleration

Answer 40

• use GPUs and CPU to hasten
• GPU database accelerates certain database operations

Question 41

graph

Answer 41

• represents a connection between a collection of entities (e.g. spending habits of consumers)
• vertex (nodes): node attributes such as age and height, and number of neighbors
• edge: is relationship between customer and product - edge identity and and weights

Question 42

graph database

Answer 42

• uses “graph theory” to store, map and query relationships of data elements
• collection of nodes and edges
• node represents an entity such as a product or customer
  
  • have: a unique identifier, a set of outgoing edges and/or incoming edges, in addition to a set of key/value pairs
• edge represents a connection or relationship between two nodes
  
  • have: unique identifier, a starting-place and/or an ending-place node along with a set of properties

Question 43

graph database image

Answer 43

Question 44

grid search

Answer 44

exhaustive method that loops through all possible combinations of variable

Question 45

GRU architecture

Answer 45

Question 46

how to fix overrepresentation of end

Answer 46

• only add end token n % of the time
• otherwise stop on second to last word

Question 47

how to prevent or cause same prediction every time

Answer 47

• dont model the initial word probability distribution
• p(w(0)) = softmax(f(‘start’)), w(0) = randint(V, p=p(w(0)))
• output probability instead of deterministic output

Question 48

increase CNN invariance

Answer 48

• modify training data
  
  • orientation,
  • color,
  • size, etc…
• can be used to increase one portion of data size if there is class imbalance

Question 49

indirect encoding

Answer 49

• field of neuroevolution
• analogous to pruning in lottery ticket
• experissive while reducing parameters

Question 50

item-item vs user-user

Answer 50

• item-item
  
  • choose items for user b/c liked similar items in the past
• user-user
  
  • choose items for user b/c liked by similar users
• uses the same algorithms just transpose the ratings matrix
• comparing items opposed to users provides more data, also it is faster

Question 51

k-fold cross validation

Answer 51

1. split into k groups
2. training groups [1:k]
3. training groups [0]+[2:k]
4. etc…
5. use t-test to compare intra

Question 52

KL divergence

Answer 52

• use to compare 2 distributions similarity
• gradient is the same as cross entropy/negative log-likelihood
• kl and cross entropy replaceable for back propagation

Question 53

L1 regularization

Answer 53

encourages sparsity (=0)

Question 54

L2 regularization

Answer 54

encourages small weights (~=0);

Question 55

latency & sensors in python

Answer 55

• write in pytorch or tensorflow
• rewrite model in C++
  
  • either rewrite completely or serialize it
    
    • pytorch tracing function
    • tensor flows graph mode

Question 56

linear regression regularization

Answer 56

• J = ∑(y - y^)²+0.5*lambda(||*||_F²+||*||_F²+…+||*||_F²)
  
  • ||*||_F²=Frobenius norm
  • * is a weight matrix
• punishes complexity
• improves error on unseen data

Question 57

local minimum

Answer 57

• is more likely a saddle point therefore is not really a problem
• slides off
• very low probability for all dimensions

Question 58

logarithmic sampling

Answer 58

how you look for parameters to ensure you get breadth instead of close values

Question 59

lottery ticket

Answer 59

• find lucky subnetwork with sparse network
• present in LSTMs and transformers
• fit on small device

Question 60

low-code/no-code

Answer 60

• ML applications w/ drag and drop components
• connect components to create a finished application
• many enterprise Business Intelligence (BI) platforms fall into this platform category

Question 61

LSTM algorithm

Answer 61

Question 62

LSTM architecture

Answer 62

Question 63

LSTM params

Answer 63

• Input gate:
  
  • params: W_xi, W_hi, W_ci, b_i
  • depends on: x(t), h(t-1), c(t-1)
• Candidate cell:
  
  • params: W_xc, W_hc, b_c
  • depends on: x(t), h(t-1)
• Forget gate:
  
  • params: W_xf, W_hf, W_cf, b_f
  • depends on: x(t), h(t-1), c(t-1)
• Output gate:
  
  • params: W_xo, W_ho, W_co, b_o
  • depends on: x(t), h(t-1), c(t-1)

Question 64

machine learning security

Answer 64

malicious modification of networks to worse output

Question 65

main problem with AI

Answer 65

• don’t have common sense
• networks find correlation, not causation
• example research: causal bayesian network

Question 66

major AI conferences

Answer 66

1. NeurIPS (NIPS): neural networks, but not exclusively
2. International Conference for Machine Learning (ICML): general machine learning
3. International Conference on Learning Representations (ICLR): really the first conference focused on deep learning.
4. Association for the Advancement of Artificial Intelligence (AAAI): more application based
5. IEEE conferences
   
   1. International Joint Conference on NN (IJCNN)
   2. IEEE International Conference on Fuzzy Systems (FUZZ-IEEE)
   3. IEEE Congress on Evolutionary Comput (CEC)

Question 67

Markov decision process

Answer 67

1. set of all states: measurements
2. set of all actions: actions the agent can do
3. set of all rewards: received at each step
4. state transition probabilities
5. discount factor (gamma)

Question 68

mixed data

Answer 68

multiple independent data types

• Numeric/continuous
• Categorical
• Image

Question 69

model stealing

Answer 69

• steal prediction models and spam filters in order to optimize against them
• e.g. - blackbox probing for stock mrket predictions
• model reconstruction: recreate model by probing the public API and refining own model using it as an oracle
• membership leakage: attacker builds shadow models that is used to determine whether a given record was used to train a model

Question 70

MSE versus cross-entropy

Answer 70

• MSE for regression
  
  • assumes target continuous and normally distributed
  • doesnt punish misclassification enough
  • vanishing gradient
• cross-entropy for classification
  
  • maximizes likelihood
  • decision boundary is large
  • converge faster

Question 71

naive bayes

Answer 71

• generative classifier (p(x|y) instead of discriminitive p(y|x))
• p(x|y)=Πp(x_n|y)
• naive because assume no covariance
• IID assumption valid if you use PCA

Question 72

natural graphs

Answer 72

DEFINITION: a set of points that have an inherent relationship between them

• Co-occurance: capture user behavior based on interactions with data

EXAMPLES:

1. Citation Graph: capture relationship between articles to other articles using citations
2. Natural Language: node represent entity and edges represent relationships between pairs of entities

Question 73

neural structured learning (NSL)

Answer 73

GOAL: optimized supervised AND neighbor loss to keep structural similarity to learn the structure; (bonus) requires less data

• Graph Regularization: idea is to train ANN with graph-regularized objective harnessing labeled and unlabeled data
• Adversarial learning: generate adversarial neighbors by keeping structure similarity to other samples. Why important?
• Adversarial structures: as opposed to graphs, they are implicitly inferred;
  
  • Use similarity between instances (using pertained embedder), else, If don’t have similar structures you create adversarials intended to mislead the neural network to producing the incorrect classification
    
    • Usually perturbations generated by reverse gradient

Question 74

neuro-symbolic

Answer 74

• symbolics: better at abstraction and reasoning
• ml: better at scalability and pattern recognition
• hybrid: understanding causal relationships

Question 75

neurons in brain

Answer 75

• 10^11 neurons in the human brain
• each is connected to 1,000 to up to 10,000 other neurons

Question 76

node classification methods

Answer 76

• deepwalk: derive embeddings from truncated random walk from graph data
• graph CNN: Use Convolution on linear layers; each layer equals expanding of network
• graph BERT: removed dependency on links

Question 77

noise injection

Answer 77

• type of regularization
• N ~ (0, var << 1)
• injection to inputs and weights

Question 78

NSL Architecture

Answer 78

Question 79

olfactory machine learning

Answer 79

• uses 4 layers
• finds the maximum signal strength
• one neuron can represent multiple smells
  
  • as opposed to visual cortex which one neuron is one pixel
• attempts coctail party problem
  
  • narrows in on particular sound signals in short period
  • disentangle conversations

Question 80

one-hot encoding

Answer 80

assign random numbers

i.e red = 1, blue = 2, and green = 3

Question 81

open-source libraries

Answer 81

• Tensorflow (Google): symbol math library; high level keras; c++ support
• Pytorch (Facebook): ML/DL
• Scikit-learn: classification, regression, and clustering algorithms
• Jax: Basically numpy with ML support and GPU, TU
• PySpark: extremely fast cluster computing for python using spark on standalone cluster
  
  • Spark SQL for dataframes
  • MLib for ML (for spark cluster - runs on Hadoop, Apache Mesos, Kubernetes)

Question 82

overfit

Answer 82

• really complex model with many parameter may not do good
• bias variance tradeoff
• simpler models overfit less
• regularization
  
  • noise injection
  • dropout
  • batch normalization

Question 83

p(drop/keep)

Answer 83

1. probability of keeping or dropping a node
2. typically p(keep) 0.8 for input layers and p(drop) 0.5 for dropout layers
3. multiply by mask

Question 84

parity problem

Answer 84

• application: data transmission in communication system
• sends bitstream to reciever
• if odd 1’s add parity bit 1
• if odd 1’s even add parity bit 0
• if odd then there is an error

Question 85

penalty term

Answer 85

term that grows as the weight grows

Question 86

pooling

Answer 86

• a method using in CNNs to downsample (reduce size of features)
• take the maximum/average in a block

Question 87

pretrained models

Answer 87

• bigger model with more comprehensive pretraining data
• performs better at multiple downstream tasks
• fewer training examples
• saves money collecting more annotated samples

Question 88

rated recurrent neural network (RRNN)

Answer 88

• Z = rate
• weight two things
  
  • f(x(t), h(t-1)): output of RNN
  • h(t-1): previous value of hidden state
• element-wise multiplication
  
  • h(t) = (1-Z)°h(t-1) + Z°f(x(t), h(t-1))
• acts like a low pass filter
• Z can be calculated many ways
  
  • weight param
  • function of X
    
    • z(t) = f(W_XZx(T)+W_HZh(t-1)+b_Z)

Question 89

recommendation interfaces

Answer 89

• news feed
• product feed
• similar product
• associated products
• search engine
• advertisements

Question 90

recurrent unit (Elman unit)

Answer 90

• h(t)=f(W_h^Th(t-1)+W_X^Tx(t)+b_h)
• y(t)=softmax(W₀^Th(t)+b₀)
• relies on all previous states, no markov assumption
  
  • X(t): D, W_i: D x M
  • h(t): M, W_h: M x M
  • Y(t): K, W₀: M x K

Question 91

recurrent vs markov model

Answer 91

• using softmax with RNN is about optimizing joint probability
• longer sequences dont go to 0
• joint probability is implicit in model

Question 92

ReLU

Answer 92

• rectified linear unit
• biologically plausible due to asymmetry
• better gradient propagation (no pretraining prep needed for deep networks)
• however dying relu can occur when all activation is turned off

Question 93

same/half padding

Answer 93

• padding the input with 0 (~half the length of the filter) to allow the output size to equal input size
• filters usually odd size to allow both sides of input to have equal padding

Question 94

saturated

Answer 94

the region of a function (such as sigmoid) that are not dynamic

Question 95

selecting CNN layers

Answer 95

• depth increases through layers, height and width decrease (reverse for type 2 - image in output)
• convolutional on the side of image
• fully-connected on the side of vector
• fully connected layers can all be the same size (research has found that it does not overfit)

Question 96

self attention diagram 1

Answer 96

Question 97

self-attention diagram 2

Answer 97

Question 98

solving class imbalance

Answer 98

sample with replacement for until the smaller set is the same size as the bigger set

Question 99

spiking neural networks

Answer 99

activation potentials resemble biological neuron

Question 100

stochastic gradient descent (SGD)

Answer 100

• assume all data IID
• very slow
• error improves long-term (although it may or may not immediately)
• sample with replacement (sometimes)

Question 101

structured v. unstructured data

Answer 101

• tabular data versus raw data
• data that looks like an sql data with input and id neatly defined versus non-structured like an image

Question 102

supervised learning for recommender systems

Answer 102

• input demographics
  
  • age, gender, religion
  • occupation, education
  • location, race
• predict the users reaction
  
  • did they buy item
  • did they click on add

Question 103

truncated back propagation

Answer 103

• long sequences take a lot of time
• stop at certain number of time steps

Question 104

types of RNN

Answer 104

1. label for sequence
2. label for each layer
3. no labels

Question 105

unrolled RNN

Answer 105

Question 106

valid/full mode

Answer 106

1. valid mode: output n-k+1
2. full mode: output n+k-1

Question 107

vanishing gradient/exploding gradient

Answer 107

• sigmoid is at max 0.25
• a**n=0/inf
• learn very slowly/not at all

Question 108

weight initialization

Answer 108

1. tanh: 1/M1 or 1/(M1+M2);
2. relu: 2/M1;

• DRAWING FROM N~(0, 1)
• should normalize input X=(X-µ)/std

Question 109

why initialize weights randomly

Answer 109

prevents symmetry (like having one unit!)

Question 110

why sigmoid?

Answer 110

• monotonically increasing
• is the output of binary logistic regression (neuron like logistic)

Question 111

zero-day inputs

Answer 111

not very often, but happen

• new product or feature launch: new funcationalities open new attack surfaces
• increase incentive: google cloud compute to mine bitcoin

Question 112

image-as-graph

Answer 112

each pixel represents a 3d RGB vector connected to 8 “adjacent” neighbors

Question 113

Text-as-graph

Answer 113

Represent text as a sequence of tokens in a directional graph

Question 114

Examples of Graphs

Answer 114

• Social networks: people as nodes and their relationship as an edge
• molecules: Atoms as nodes and covalent bonds as edges
• Citation networks:each paper is a node and and edge is a citation between one paper and another
  
  • Can even do word embedding of abstract as information about one node

Question 115

Graph vs. Node vs. Edge TASK

Answer 115

• Graph: predict the property of the entire graph (.e.g. smell of molecule, label,)
• Node: predict the identity of role of each node (e.g. member loyal to John or Jane, image segmentation, POS tagging)
• Edge: determine the relationship between objects in images

Question 116

representing graphs in ML

Answer 116

1. Tabular: often sparse
   
   • node: feature matrix N x D (nodes x features)
   • connectivity:
2. Lists: more computationally efficient
   
   • Nodes = [[0, 1], [1, 0], [0, 0]
   • Edges = [[2, 1], [1, 1,]]
   • Adjacent list = [[1, 0], [2, 0]]

Question 117

graph neural network

Answer 117

an optimizable transformation on all attributes of the graph (nodes, edges, global-context)

• graph-in, graph-out: architecture accepts graph as an input, with information loaded into its nodes, edges, and global-context, and progressively transforms the embeddings, without changing the connectivity of the input graph

Question 118

simplest GNN

Answer 118

• Learn embeddings for graph attributes (nodes, edges, global) W/O changing connectivity
• Multilayer perceptron on each component
  
  • N(n_f|n_0), E(e_f|e_0), G(g_f|g_0)

Question 119

Large Language Generative Models Sizes

Answer 119

• params: NVidia-Turing (570 b), GPT3 & AI21 Jurassic (175 b)
• tokens: GPT3 (300 b), chinchilla (1.4 t)
• human feedback training

Question 120

areas for improvement generative models 2022

Answer 120

1. memory - recall previous information
2. summarize - decompose info into elements
3. Model/Data quality assurance -
   
   1. 4.6-17.2 T tokens (books, literature, articles)
   2. NOT blogs, webpages, social media, spoken content
4. bigger is not better = more cost, impossible to beat, …
5. text-to-video

Question 121

affective AI

Answer 121

• emotion detection in image, video, text, sound waves
• current technologies average samples which leads to possible bias (e.g. - cultural and sex differences)

Question 122

2024 predictions

Answer 122

1. humanoid robots - maybe a foundational model
2. ML Ops - weights and biases (raised 200 million at 1 billion)
3. ethics and law tossed in generative AI mix
   
   1. still lacking basic knowledge
4. AI search using generative AI
5. mixed modal foundational models

Question 123

6 openAI competitors

Answer 123

whomever can run cheapest and best for the job

1. AdeptAl - ~660M Greylock - 20 employees - browser extension for instructional surfing
2. Al21 - 750 million - 160 employees - customize models for automated copywriting, summarize documents
3. Anthropic - 4 billion - 80 employees - general-purpose LLMs free of bias and toxicity
4. Character - 18 employees - create and interact with chatbots that can role play
5. Cohere - ~1 bil Tiger Global - summarizing documents, copywriting, and search
6. Inflection - 1.23B Grevlock - communicate with machines to relay our thoughts and ideas

Question 124

How to find bad labelled data?

Answer 124

1. Sort by model losses and finding ambiguous samples, AND sorting by confidence where model and ground truth disagree
2. Confident Learning: pruning noisy data based on prediction intervals (worse than aforementioned method)