Cloud Hosting

Its not a surprise that AI and data science professionals remain in demand given the explosion of AI models on the market, and the rapid-fire advancements since. But just as companies are still struggling to figure out business use cases for LLMs, they also struggle to identify corresponding job roles. To make matters worse, there are additional obstacles popping up along the way.

Theres the acceleration of AI-adjacent talent wars: Its not just AI, says Babak Hodjat, CTO of AI at Cognizant Technology Solutions.

What types of job roles fall into the category of AI-adjacent talent?

At Persado, according to Frank Chen, the companys head of natural language processing, that list includes:

Research scientists who conduct cutting-edge research to develop new models and techniques for generative AI tasks.

Machine learning/NLP/data/software engineers who build and deploy GenAI models.

Data scientists and analysts who experiment and analyze model results.

Content specialists who create guidelines to control the generated content, collaborate with the AI development team to refine the generated content, evaluate the quality of generated content, and provide feedback to improve GenAI models.

Experienced UX/UI designers who ensure the designed AI interface aligns with user needs.

Related:Prompt Engineering: Passing Fad or the Future of Programming?

But job competition is also heating up elsewhere.

In 2024, well also see the war for cyber and software development talent grow more contentious as a result of major privacy concerns and savings-driven budget reallocations born from the generative AI boom, Hodjat says.

No doubt more job roles will emerge while others will soon fade away as AI matures, and companies become more adept at using it.

Data science and AI continue to be industries with strong growth projections, but there are a few jobs in particular that should be in demand for the foreseeable future, says Richard Gardner, CEO of Modulus, which touts a client list including NASA, NASDAQ, Goldman Sachs, Merrill Lynch, JP Morgan Chase, Bank of America, Barclays, Siemens, Shell, Yahoo!, Microsoft, Cornell University, and the University of Chicago.

Some job titles are already familiar such as prompt engineers and AI content editors. But those roles do not have standard descriptions or pay scales. For example, job boards are filled with ads for AI content editors at a mere $20 to $40 an hour. These are typically posted by employers who think these jobs require simple skills and little effort. But that is blatantly untrue. Depending on the complexity of the subject matter, it can actually take humans longer and exert more effort to edit and fact-check GenAI outputs than it does to write the thing from scratch.

Related:Hire or Upskill? The Burning Question in an Age of Runaway AI

Prompt engineer job roles also see a wide swing in job descriptions and pay scales. Sometimes vague job descriptions and low pay offers are due to an employers lack of experience in working with AI or a general cluelessness of what use cases they have for AI. Other times, its the opposite. The employer knows exactly what technical and linguistic skills they need from this group of job candidates and the pay offered better reflects the level of complexity.

In any case, according to ZipRecruiter, as of Nov 29, 2023, the average annual pay for a Prompt Engineering in the United States is $62,977 a year. Just in case you need a simple salary calculator, that works out to be approximately $30.28 an hour.

Eventually the job of prompt engineer will likely disappear as AI agents can already write their own prompts. But for now, most companies are looking to hire prompt engineers to get the most out of AI while also keeping token costs down.

As many would expect, the demand for large language model (LLM) engineers and natural language processing (NLP) engineers is on the rise.

Related:The IT Jobs AI Could Replace and the Ones It Could Create

The new and highly specialized role known as the LLM Engineer is primarily found within organizations that have reached an advanced stage in their AI journey, having conducted numerous experiments but now facing challenges in the operationalization of their AI models at scale, says Kjell Carlsson, head of data science strategy and evangelism at Domino Data Lab.

Glassdoor reports that as of November 2023, the estimated total pay for a LLM engineer is $164,029 per year in the United States area, with an average salary of $129,879 per year. The range of total pay is between $123k and $224k. ZipRecruiter pegs average total pay for this group at $142,663 per year, or $69 per hour.

NLP engineers are seeing an uptick in demand for AI and non-AI based projects but is itself continuing to evolve.

For example, Natural Language Processing Engineers, which essentially work to make applications that process human language, are currently in high demand for chatbots, but, over time, will continue to see demand for sentiment analyses and content recommendation systems, Gardner says.

Glassdoor reports the pay range for NLP engineers to be between $130k and $180k per year. Talent.com reports a wider pay spectrum of between $100k and $210k a year.

A range of other jobs are emerging, too. Topping this list is AI agent jobs and skills.

The agent view of AI [autonomous AI agents] will take on an increasingly significant role in AI-enablement projects, becoming a sought-after skill, Hodjat says. Various orchestration frameworks and platforms will vie to become the standard and software engineering will move towards adopting LLM-based agents into the fabric of software systems.

Other emerging job roles are harder to define and peg to a salary range.

Some of the most sought-after AI positions today include machine learning engineer, AI engineer, and AI architect, says Shmuel Fink, Chair Master of Science in data analytics, Touro University Graduate School of Technology. Nevertheless, several other AI roles are also gaining prominence, such as AI ethicist, AI product manager, AI researcher, computer vision engineer, robotics engineer, and AI safety engineer. Moreover, there are positions that require industry-specific expertise, like a healthcare AI engineer.

But back at the ranch, employees in any job role will become more valuable if they possess AI skills. As they gain those skills, some specialized job roles will evolve while others disappear. The one thing that is certain is that theres far more AI-driven and AI-adjacent change to come.

Original post:

Hot Jobs in AI/Data Science for 2024 - Hot Jobs in AI/Data Science for 2024 - InformationWeek

Sorting is the process of organizing elements in a structured manner. QuickSort is one of the most popular sorting algorithms that uses nlogn comparisons to sort an array of n elements in a typical situation. QuickSort is based on the divide-and-conquer strategy.

QuickSort is a sorting algorithm that uses a divide-and-conquer strategy to sort an array. It does so by selecting a pivot element and then sorting values larger than it on one side and smaller to the other side, and then it repeats those steps until the array is sorted. It is useful for sorting big data sets.

Well take a look at the quicksort algorithm in this tutorial and see how it works.

QuickSort is a fast sorting algorithm that works by splitting a large array of data into smaller sub-arrays. This implies that each iteration splits the input into two components, sorts them and then recombines them. The technique is highly efficient for big data sets because its average and best-case complexity is O(n*logn).

QuickSort was created by Tony Hoare in 1961 and remains one of the most effective general-purpose sorting algorithms available today. It works by recursively sorting the sub-lists to either side of a given pivot and dynamically shifting elements inside the list around that pivot.

As a result, the quicksort method can be summarized in three steps:

More on Data ScienceBubbleSort Time Complexity and Algorithm Explained

Lets take a look at an example to get a better understanding of the quicksort algorithm. In this example, the array below contains unsorted values, which we will sort using quicksort.

The process starts by selecting one element known as the pivot from the list. This can be any element. A pivot can be:

For this example, well use the last element, 4, as our pivot.

Now, the goal here is to rearrange the list such that all the elements less than the pivot are to its left, and all the elements greater than the pivot are to the right of it. Remember:

Lets simplify the above example,

As elements 2, 1, and 3 are less than 4, they are on the pivots left side. Elements can be in any order: 1,2,3, or 3,1,2, or 2,3,1. The only requirement is that all of the elements must be less than the pivot. Similarly, on the right side, regardless of their sequence, all components should be greater than the pivot.

In other words, the algorithm searches for every value that is smaller than the pivot. Values smaller than pivot will be placed on the left, while values larger than pivot will be placed on the right. Once the values are rearranged, it will set the pivot in its sorted position.

Once we have partitioned the array, we can break this problem into two sub-problems. First, sort the segment of the array to the left of the pivot, and then sort the segment of the array to the right of the pivot.

Lets cover a few key advantages of using quicksort:

Despite being the fastest algorithm, quicksort has a few drawbacks. Lets look at some of the drawbacks of quicksort.

The subarrays are rearranged in a certain order using the partition method. You will find various ways to partition. Here, we will see one of the most used methods.

Lets look at quicksort programs written in JavaScript and Python programming languages. Wellstart by creating a function that allows you to swap two components.

Now, lets add a function that uses the final element (last value) as the pivot, moves all smaller items to the left of pivot, and all larger elements to the right of the pivot, and places the pivot in the appropriate location in the sorted array.

Next, lets add the main function that will partition and sort the elements.

Lets finish off by adding a function to print the array.

Here is the full code of quicksort implementation for JavaScript:

Now, lets look at quicksort programs written in Python.Well start by creating a function which is responsible for sorting the first and last elements of an array.

Next, well add the main function that implements quick_sort.

Lets finish off by adding a function to print the array.

Here is the full code of quicksort implementation in Python.

Lets look at the space and time complexity of quicksort in the best, average and worst case scenarios. In general, the time consumed by quicksort can be written as follows:

Here, T(k) and T(n-k-1)refer to two recursive calls, while the last term O(n) refers to the partitioning process. The number of items less than pivot is denoted by k.

When the partitioning algorithm always chooses the middle element or near the middle element as the pivot, the best case scenario happens. Quicksorts best-case time complexity is O (n*logn). The following is the best-case recurrence.

This occurs when the array elements are in a disordered sequence that isnt increasing or decreasing properly. QuickSorts average case time complexity is O(n*logn). The following is the average-case recurrence.

The worst-case situation is when the partitioning algorithm picks the largest or smallest element as the pivot element every time. The worst-case time complexity of quicksort is O (n2). The following is the worst-case recurrence.

The space complexity for quicksort is O(log n).

The sorting algorithm is used to find information, and since quicksort is the fastest, it is frequently used as a more efficient search approach.

QuickSort may have a few drawbacks, but it is the fastest and most efficient sorting algorithm available. QuickSort has an O (logn) space complexity, making it an excellent choice for situations where space is restricted.

Although the worst-case running time is always the same, quicksort is often faster than HeapSort (nlogn). QuickSort takes up less space than heap sort due to the fact that a heap is nearly a full binary tree with pointers overhead. So, when it comes to sorting arrays, quicksort is preferred.

An error occurred.

More on PythonHow to Implement Binary Search in Python

There might be a few drawbacks to quicksort, but it is the fastest sorting algorithm out there. QuickSort is an efficient algorithm that performs well in practice.

In this article, we learned what quicksort is, its benefits and drawbacks and how to implement it.

See more here:

QuickSort Algorithm: An Overview - Built In

Modern data warehousing and Machine Learning12 min read

No doubt, user retention is a crucial performance metric for many companies and online apps. We will discuss how we can use built-in data warehouse machine learning capabilities to run propensity models on user behaviour data to determine the likelihood of user churn. In this story, I would like to focus on dataset preparation and model training using standard SQL. Modern data warehouses allow this. Indeed, retention is an important business metric that helps understand user behaviours mechanics. It provides a high-level overview of how successful our Application is in terms of retaining users by answering one simple question: Is our App good enough at retaining users? It is a well-known fact that its cheaper to retain an existing user than to acquire a new one.

In one of my previous articles, I wrote about modern data warehousing [1].

Modern DWH has a lot of useful features and components which differentiate them from other data platform types [2].

ML model support seems to be the foundational DWH component when dealing with big data.

In this story, I will use Binary logistic regression, one of the fastest models to train. I will demonstrate how we can use it to predict user propensity to churn. Indeed, We dont need to know every machine-learning model.

We cant compete with cloud service providers such as Amazon ang Google in machine learning and data science but we need to know how to use it.

I previously wrote about it in my article here [3]:

In this tutorial, we will learn how to transform raw event data to create a training dataset for our ML

See the rest here:

User Churn Prediction. Modern data warehousing and Machine | by Mike Shakhomirov | Dec, 2023 - Towards Data Science

2023 may have been the year of the LLMwe highlighted our most popular articles on ChatGPT and related topics last weekbut data science and machine learning are far too vast for us to reduce them to a single phenomenon (as inescapable as it might be).

Every day, TDS authors publish excellent work on a staggering range of topics, from the latest tools of the trade to career insights and project walkthroughs. For our final Variable edition of the year, we decided to highlight some of the most memorable and widely read posts weve shared around three themes: programming for data scientists, career growth, and and creative projects and opinion pieces. They do a fantastic job showing just how vibrant, diverse, and dynamic this fieldand our communityis.

We hope you enjoy our selection, and thank you once again for all your support over the past year.

Here is the original post:

Data Science Career Paths, Skills, and Special Projects: Our Best Reads of 2023 - Towards Data Science

As you can see above with greedy strategy, we append the token with the highest probability to the input sequence and predict the next token.

Using this strategy lets generate a longer text with 128 next tokens using greedy-search decoding.

As you we can see from the text above, although it is the simplest logic, the drawback of this approach is the generated repetitive sequences. As it fails to capture the probabilities of sequences, meaning, the overall probability of a several words coming one after another is overlooked. Greedy search predicts and considers only the probability one step at a time.

Repetitive text is a problem. We would desire our generated output to be concise, how can we achieve it?

Instead of choosing the token that has highest probability at each step, we consider future x-steps and calculate the joint probability(simply multiplication of consecutive probabilities) and choose the next token sequence that is most probable. While x refers to number of beams, it is the depth of the future sequence we look into the future steps. This strategy is known as the beam search.

Lets go back to our example from GPT-2 and explore beam vs greedy search scenarios.

Given the prompt, looking at the two tokens with highest probability and their continuation(4 beams) in a tree diagram

Lets calculate the join probabilities of the green sequences above.

Germany is known for its -> high-quality beer

with the joint probability 3.30%*24.24%*31.26%*6.54% = 0.0016353

whereas the lower path with the sequence;

Germany is known for its -> strong tradition of life

2.28%*2.54%*87.02%*38.26% = 0.0019281.

The bottom sequence overall resulted with the higher joint probability, although the first next token prediction step in the top sequence has higher probability.

While greedy search priorities the absolute maximum probability at each prediction step, it neglects the token probabilities in sequences. Beam search decoding enables us to go in depth of sequences and help us decode text in more extensive fashion. So is beam-search the ultimate solution?

Lets explore further and decode the next 128 tokens with the depth of 5 beams.

Here are 10 of the most beautiful places in Germany.

1. Lake Constance

Lake Constance is one of the largest lakes in Germany.

It is located in the state of North Rhine-Westphalia and is the second largest lake in Germany after Lake Constance in Bavaria.

Lake Constance is located in the state of North Rhine-Westphalia and is the second largest lake in Germany after Lake Constance in Bavaria.

"""

Although comparatively lesser than the greedy-search, beam-search suffers from repetitive output too. However, with beam search decoding, we can solve this problem by penalising the repeated pairs of word sequences. In other words, the probability of token sequences is assigned zero, if the sequence has already been decoded before. This penalisation of a repeated tokens sequence is also know as n-gram penalty.

While n signifies the length of the sequence, gram is a term that refers to unit in computational linguistic often corresponds to the term token in our case.

The reasoning behind is to discourage the generation of sequences that contain consecutive repeating n-grams. The decoding algorithm will penalise generated sequences that contain repeating pairs of words in the output.

Knowing this, lets apply n-gram penalty of n = 2.

The country's capital, Berlin, is the largest city in Europe, with a population of more than 8.5 million people.

The city is located in the former East Germany, which was divided into East and West Germany after World War II.

Today, Germany is a member of both the European Union and NATO, as well as the World Trade Organization and the Organization for Economic Cooperation and Development (OECD).<|endoftext|>

"""

This is the best completion of the input prompt we extracted from the model so far in terms of coherence and compactness. Through n-gram penalisation the output decoded with beam-search became more human-like.

When should we use beam-search and when greedy-search? Where the factualness is paramount, like solving a math problem, key information extraction, summarisation or translation, greedy-search should be preferred. However, when we want to achieve creative output and factuality is not our priority (like it can be in the case of story generation) beam-search is often the better suited approach.

Why exactly does your prompt matter? Because every word you choose to use, the sentence structure, the layout of your instructions will activate different series of parameters in the deep layers of large language model and the probabilities will be formed differently for each different prompt. In the essence of the matter, the text generation is a probability expression conditional on your prompt.

There are also alternative methods to prevent repetitions and influence the factuality/creativity of the generated text, such as truncating the distribution of vocabulary or sampling methods. If you are interested in a higher-level in-depth exploration of the subject, Id highly recommend the article from Patrick von Platen in HuggingFace blog.

Next and the last article of this series will explore fine-tuning and reinforcement learning through human feedback which played an important role on why pre-trained models succeeded to surpass SOTA models in several benchmarks. I hope in this blog post, I was able help you understand the reasoning of prompt engineering better. Many thanks for the read. Until next time.

Read this article:

Inside GPT II. The core mechanics of prompt engineering | by Fatih Demirci | Dec, 2023 | Medium - Towards Data Science