Domination… Generative AI and ChatGPT have taken the AI debate to the next level, capturing the attention of entrepreneurs and heads of state alike. Business leaders and top management need to keep abreast of trends and applications of generative AI in order to create a compelling strategic investment case.

4

Hello! My name is Vlad Proshinsky, I have been launching and developing IT products for 8 years in the role from Product Manager to CPO. For the last 8 months I have been studying how AI and LLM can be applied in products and business processes. I am convinced that knowledge in the field of neural networks will give managers and founders a great advantage

In this article, I want to share my strategic vision, cases of using generative AI in products and in automating business processes of world-class companies.

Generative AI is a type of AI that creates text, photo, video, audio and other content at scale using data arrays (LLM) and machine learning (ML)

Generative AI has had an impact like no other technology in 10 years. For six months, ChatGPT managed to prove that neither the cost of computing, nor the complexity of the infrastructure, nor human resources are no longer an obstacle to mass adoption, the AI transformation has begun.

According to a McKinsey report, AI could generate an additional $13 trillion in growth. by 2030, increasing global GDP by 1.2% per year.

Domination and strategic context

Mass adoption of technology in the active phase, not only in personal use, but also in business processes. The productivity gains of developers and knowledge workers using GPT-based products are massively proven, and have forced organizations to rethink their business processes and the value of human resources. The way product solutions based on GPT and other neural networks develop, how AI is integrated into everyday life of a person and business is another proof that AI is a strategic necessity.

Total domination with AI. Strategy, neural network tools and ChatGPT
Total domination with AI. Strategy, neural network tools and ChatGPT 18

Source: Bloomberg News, income statement analysis. Note: analysis in the context of the terms “AI” and “Artificial Intelligence”. Data for July 27, the share of Nasdaq100 informing companies was 40%

In this quarter’s earnings report, US tech CEOs are talking less about the problems and more about how they plan to use AI to boost revenue and lower costs. Mentions of “AI” have increased more than 4 (!) times compared to 2022, indicating that AI has become a key driver of competitiveness and innovation in the technology industry.

The AI winners take it all

“Winner takes all”, and in the context of AI, this expression has a direct connotation. In the technology race, leaders are afraid of being left behind if they slow down and others don’t. This “prisoner’s dilemma” from game theory creates risks for responsible AI practices. Speed-to-market leaders are driving the current “arms race” in which big corporate players are rushing to release neural network-based products.

CEOs and founders are actively pushing teams to find ways to incorporate generative AI into their products and processes, but to create value with these technologies, you need to understand how they work, where they will bring benefits, what their limitations and risks are. Such competencies are usually rare among top management. Therefore, they need to be looked for on the side, on platforms like CareerPaths or freelance platfroms like Upwork or Fiverr.

In the article, I will briefly touch on the development of technology, as a product specialist, I am interested in how this technology will change the needs of users, ways of solving problems, and most importantly: “how to make money on it?”, After all, the task of a business is to make a profit. But according to statistics, 87% of ML projects do not even reach implementation, and the main problem is that few people now have comprehensive knowledge about what kind of “cube” in the form of an AI solution to build into the system, how to do it, how everything is not “ break”.

Common problems teams face:

  • Wrong processes of workflow on an AI project
  • Misidentifying Opportunities for AI Applications
  • No vision (experience) → a lot of resources are wasted
  • Business and products still do not speak the language of DS and ML engineers
  • Do not identify possible risks at the start
  • Passion for experiments, ignoring business goals

In today’s realities, businesses need managers who will see strategic opportunities to implement AI where possible, as well as build operational end-to-end processes.

Strategic vector

Strategic innovations that require special attention (2 to 5 years to mainstream adoption) include generative AI and decision intelligence. Early adoption of these innovations will lead to a significant competitive advantage and will simplify the challenges associated with using AI models in business processes.

Gartner
image 1
Hype Cycle for AI © Gartner (от 19 July 2023)

According to the graph provided by Gartner, general AI (AGI, artificial general intelligence), also known as “artificial intelligence”, capable of finding a solution for arbitrary tasks and learning to infinity, will appear in about 10 years. The forecast of the Metaculus community is slightly more optimistic – January 2027.

Obviously, development progress towards AGI is artificially hindered: the first letter demanding to immediately suspend training of more powerful models than GPT-4 was published in March, the second – in May. As a result, in mid-April, OpenAI announced that they had suspended the training of the GPT-5 model and were engaged in “other things” (fast API and reducing the “AI cost”), and on July 30, the same OpenAI applies for the GPT-5 patent… Process irreversible.

An intermediate step is “causal / casual AI” (Casual AI) is a new type of AI that can detect and reason about the relationship of causes and effects, as well as determine the strength of each connection. Causality is very important for subsequent progress. It is characterized by: high autonomy, adaptability, making informed and logical decisions.

Causal AI is a key stage in the development of AI towards greater decision-making automation, autonomy, reliability and common sense

Gartner

On July 14, 2023, Elon Musk announced the launch of xAI, the goal of which is to “understand the true nature of the Universe,” and Casual AI is exactly what is needed to solve such problems. Among the early tools: Auto-GPT, BabyAGI, SuperAGI, ToddlerAGI.

Current (operational) context

Now in most AI products under the hood is GPT-3.5 or GPT-4, which is further trained to solve specific problems within the product, and the most popular product is ChatGPT, which gained 1 million users in a record 5 days in 2022.

image 2
Statista

The appearance of GPT-5 will undoubtedly improve the quality of products and open up new opportunities for solving problems and ChatGPT itself. Thus, your GPT assistant, which can help you do your work more efficiently, will become smarter over time, and the quality and speed of tasks will grow exponentially.

Companies and professionals who have implemented AI tools in their work win in the moment and over time, the gap from those who do not use them will rapidly lose to more advanced ones.

Why is this so important right now?

A report by Microsoft researchers explains how the GPT-4 language model learns human thinking skills. Can this be considered a milestone in the emergence of AGI (Artificial General Intelligence)?

Given the breadth and depth of GPT-4’s capabilities, we think it would be reasonable to view it as an early but still incomplete version of General AI (AGI).

GPT-4 has signs of human thinking, the argument is that GPT-4 achieves human-level performance in many tasks.

S´ebastien Bubeck (Microsoft Research)

ChatGPT can be considered as a mediocre assistant, the level of which depends on the skills of prompting (the ability to communicate with neural networks in the form of prompts), but now it is the starting point for a completely different future, which will come in 3-5 years.

Product changes

Global product companies are actively implementing AI into their products, adding value by increasing the speed of work and solving problems from scratch, this allows you to better activate users in key JTBD scenarios. Examples:

AI Tools

A small list of tools that you can start using in your work and in the company:

Marketing

image 3
AnswerThePublic

Sales

image 4
Fiber AI

Product

Support

Forbes writes that Meta , Canva , and Shopify use ChatGPT-based solutions in their customer support chatbots. Support Solutions: Kapa , Clueso , Ada , MagicReply

image 5
MagicReply от Crisp

Development

According to GitHub, Co-pilot is involved in 46% of the code (among those who use it) and helps to write code 55% faster

Finance

HR / recruitment

  • automation of the recruitment process (job sites → resume analysis with ChatGPT)
  • layoff forecasting ( Xsolla case )
  • resume preparation: HyreSnap , Resume.io
  • Interview Coach: Yoodli

Management / Management

  • Kona – Executive coaching
  • Stellar – sets goals / KPIs in the company and monitors their implementation

It is important to note the nuances of working with GPT:

  • “Garbage in → garbage out” is a clear description of ChatGPT. First, master the skills of writing prompts ( cheat sheet ).
  • Hints and questions lead to better results. If ChatGPT does not know the answer, it will invent one (needs to be double-checked) and can forge sources and give plausible but incorrect or nonsensical answers
  • the accuracy (accuracy) of the answers may be reduced. Study of accuracy reduction from 52% to 10%

In summary, I want to note that at the stage of business growth, experiments play a special role. Testing hypotheses is one of the main tasks of marketers and product developers. By testing the listed tools, there is a high probability that you will be able to find points of increase in efficiency in your work.

***

The effect of the introduction of Synthesia AI avatars , as examples of growth that we plan to implement in one of the client’s projects:

“Thanks to AI avatars, we had 35% more meetings than at the last exhibition

Infinite Peripherals / VC of Marketing

The training team cut 50% of the cost of voice over video compared to hiring multilingual professionals

Xerox / Global L&D Sales Learning Strategy Manager

We reduced video production time by 95% from weeks to hours.

Network Rail L&D / Multimedia Development Manager

We increased engagement in online learning by 30% compared to text modules

BSH / Global Supply Chain Training

Open source AI

Separately, I will highlight open source solutions. They are more suitable for medium-sized businesses, for projects that require more flexibility, security, and budget savings.

According to Trends.vc analytics , open source AI is helping companies build infrastructure faster by learning from each other. Companies that offer closed source AI decide when and what you can use it for. Open source AI helps us learn from and build on each other’s work, it turns the AI ​​arms race into collaboration.

Open source platforms

  • Huggingface – creating and deploying AI Open-source models
  • Replicate – create and run AI models in the cloud
  • Google Colab – for machine learning research
  • Difty – open-source for LLMOps, AI-native applications

In the future, there will be more platforms built to host open source models. They will simplify the creation and deployment of models.

Open source models

  • Panda GPT – will create a description from the picture, write stories, videos and answer questions from the audio. Can simultaneously accept multimodal inputs and compose their semantics
  • Stable Diffusion – transforms text into an image (text-to-image)
  • Riffusion – converts text to audio (text-to-audio)
  • BLOOM – open multilingual model
  • DragGAN – changes the poses, shape, expression and location of objects
  • Falcon – LLM for Research and Commercial Use

Open-source datasets

  • The Pile is a dataset of books, web pages, chat logs, and more.
  • ImageNet – 14,000,000+ images with comments
  • OIG – Conversational Data for AI Chatbots

open source tools

  • PyTorch – a framework for creating deep learning models
  • TensorFlow – open source ML platform
  • Keras – API for deep learning models
  • Prem AI – LLM Deployment Tool on Servers

Open Source Alternatives to ChatGPT

Advantages of open-source AI

  • Open source solves vendor lock-in and high shift costs. Platforms like Hugging Face make it easy to find the right AI models for your options.
  • Open source AI allows you to create niche applications that large companies (like OpenAI) with closed source do not have the time, knowledge or interest to create.
  • Clear documentation improves the quality and adoption of your open source AI tool.
  • Open source AI models may be slightly worse, but much cheaper.

Opportunities for Entrepreneurs

The emergence of a new “job” (in terms of JTBD ) in consumer segments is already actively taking shape. This generates demand for expertise in implementing AI into business processes to gain competitive advantage. In this regard, AI consulting will be actively developed , as well as courses and master classes on implementing solutions into business processes.

Companies will increasingly outsource such tasks as there is a high risk of internal sabotage among employees due to fears of being out of work (a 2023 WEF report states that by 2025, 50% of workplace tasks will be automated).

With the formation of a market for AI business consultants, launches of info products will gain momentum. You can create courses and master classes to implement AI in business processes, as well as train freelancers to master the tools. For example, English-language master classes:

Summary

image 6
Garner Hype Cycle for AI, 2023

Generative AI is at the peak of the hype, according to a recent Gatner report, expectations from the technology are overstated and, apparently, we are moving into a phase of “disillusionment”, as evidenced by the decline in the use of ChatGPT among enthusiasts (in the Russian Federation, the ChatGPT audience decreased by 30%, and Midjourney – by 50 % in June) , however, according to the cycle and forecasts of strategists, the technology will be widely used in early 2024.

Recently, Gartner conducted a survey at The Future of Data-Centric AI conference , which showed that the majority of participants planned to enter the industrial use of large language models (LLM) by the end of 2023. Companies are forced to add AI to their decisions and processes in order to remain competitive.

Comments are closed.

Check Also

Bee Sharing continued: How much money does one bee bring? The economics of beekeeping.

https://pchelosharing.ru/ In November we wrote about how to “digitize” an apiary and turn …