/pcq/media/agency_attachments/2025/02/06/2025-02-06t100846387z-pcquest-new-logo-png.png){kind=logo}
Follow Us
/pcq/media/post_banners/wp-content/uploads/2023/03/Generative-AI.jpg)
The world is becoming increasingly digital and AI technology is rapidly developing at an unprecedented rate. AI has become a ubiquitous presence in our lives, from personal assistants to self-driving cars. AI is transforming the way we work, interact, and even think. But what exactly is AI? What are the different kinds of AI? And what does the future of AI look like? In this article, we’ll explore the exciting world of AI and take a look at the future of generative AI.
Introduction to AI
AI, or artificial intelligence, is a term used to describe the capability of a machine or computer to simulate human intelligence. It is the science and engineering of making intelligent machines, especially computer systems, that can think and learn. It encompasses a range of techniques, including machine learning, natural language processing, computer vision, and robotics, to enable computers to perform tasks that typically require human intelligence, such as recognizing speech, making decisions, and solving problems. AI is used to solve complex problems that were previously too difficult or impossible for computers to solve.
AI is divided into two main categories: narrow AI and general AI. Narrow AI is a specialized form of AI that performs specific tasks, such as facial recognition, image recognition, and natural language processing. General AI, on the other hand, is a more advanced form of AI that can learn and adapt to its environment.
What is Generative AI
Generative AI is a type of AI that is used to create new content which has become a craze in 2022 - 2023. The world got the taste of generative AI when OpenAI’s Chatgpt got 100 Mn users in just 2 months. In comparison it took Facebook more than 4 years to hit that number. Generative AI is a form of AI that can generate novel and creative content such as images, videos, text, and music. The goal of generative AI is to generate outputs that are indistinguishable from real-world data, and that can be used for various applications such as data augmentation, content creation, and anomaly detection. There are two main types: GANs and VAEs. Simply explained, GANs play a game of generating data and deciding if it's real or fake. VAEs simplify data, then recreate it to make sure it's similar to the original. Both types are used to generate various forms of content and have multiple applications. Generative AI is a powerful tool that can help us create new ideas, solve complex problems, and even create new art.
GPT 3: A Example of Auto Regressive Language Model and its far Reaching Consequences.
Even before 2022, there were AI-powered tools in everyday life such as Grammarly and Zoom that helped us automate mundane tasks and boost our productivity. Further, AI tools such as Adobe Sensei and Google AutoDraw are being used to help us generate new ideas and create new artwork. These tools can help us think outside the box and come up with creative solutions.
However, in late 2022 internet broke loose with OpenAI's ChatGPT. ChatGPT is an example of an auto-regressive language model (AR LM). It uses a deep neural network to generate text based on the previous words in a sequence, in a process called autoregression. In other words, GPT-3 predicts the next word in a sentence based on the context of the previous words, and it does so by using the vast amounts of text data that it was trained on.
While people are using ChatGPT from writing emails, to editing code etc the consequences of this technology are far reaching. LLMs may be classified into seven broad categories of use cases— 1. Generate, 2. Summarize, 3. Rewrite, 4. Extract, 5. Search/ Similarity, 6. Cluster, and 7. Classify. The present version of ChatGPT mainly performs the former 4 use cases. Nonetheless, there are many LLMs that will perform all these 7 use cases.
2023 Wars in Generative AI: ChatGPT, Bard, Ernie etc
Venture capitalist Chamath Palihapitiya has recently stated on a podcast that only a handful of organizations possess the ability to comprehensively crawl the internet in terms of cost, compute power, and AI capabilities. He has identified a small group of potential competitors to Google Search, which includes Microsoft, Oracle, Chinese internet companies, and Facebook, suggesting that these companies could pose a significant challenge to Google with the appropriate investment of resources.
The pioneer in the field of generative AI that has captured the mass imagination is OpenAI. OpenAI Inc is a private artificial intelligence research laboratory consisting of the for-profit OpenAI LP and its non-profit parent organization. It was founded in 2015 by tech luminaries such as Elon Musk, Sam Altman, and Greg Brockman to protect against a future in which big tech companies monopolize the benefits of AI technology. Incidentally, Microsoft had invested $1 billion in OpenAI in 2019 and a second multi-year investment in January 2023, reported to be $10 billion. Microsoft will own 49% of OpenAI. Open AI is behind ChatGPT and DALL-E which has got everyone hooked. ChatGPT is a chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and has been designed to have human-like conversations. DALL-E is a deep learning model developed by OpenAI to generate digital images from natural language descriptions.
Microsoft is integrating GPT 3.5 into its native search engine Bing in a phased manner. To be waitlisted for this service everyone is downloading Bing. As such once obscure search engine Bing seems to have resurrected in popularity atleast. According to many initial users; however, the AI that Microsoft integrated with Bing is not ready for human contact.
Google is seriously worried about its search monopoly due to the rise of
The world is becoming increasingly digital and AI technology is rapidly developing at an unprecedented rate. AI has become a ubiquitous presence in our lives, from personal assistants to self-driving cars. AI is transforming the way we work, interact, and even think. But what exactly is AI? What are the different kinds of AI? And what does the future of AI look like? In this article, we’ll explore the exciting world of AI and take a look at the future of generative AI.
Introduction to AI
AI, or artificial intelligence, is a term used to describe the capability of a machine or computer to simulate human intelligence. It is the science and engineering of making intelligent machines, especially computer systems, that can think and learn. It encompasses a range of techniques, including machine learning, natural language processing, computer vision, and robotics, to enable computers to perform tasks that typically require human intelligence, such as recognizing speech, making decisions, and solving problems. AI is used to solve complex problems that were previously too difficult or impossible for computers to solve.
AI is divided into two main categories: narrow AI and general AI. Narrow AI is a specialized form of AI that performs specific tasks, such as facial recognition, image recognition, and natural language processing. General AI, on the other hand, is a more advanced form of AI that can learn and adapt to its environment.
What is Generative AI
Generative AI is a type of AI that is used to create new content which has become a craze in 2022 - 2023. The world got the taste of generative AI when OpenAI’s Chatgpt got 100 Mn users in just 2 months. In comparison it took Facebook more than 4 years to hit that number. Generative AI is a form of AI that can generate novel and creative content such as images, videos, text, and music. The goal of generative AI is to generate outputs that are indistinguishable from real-world data, and that can be used for various applications such as data augmentation, content creation, and anomaly detection. There are two main types: GANs and VAEs. Simply explained, GANs play a game of generating data and deciding if it's real or fake. VAEs simplify data, then recreate it to make sure it's similar to the original. Both types are used to generate various forms of content and have multiple applications. Generative AI is a powerful tool that can help us create new ideas, solve complex problems, and even create new art.
GPT 3: A Example of Auto Regressive Language Model and its far Reaching Consequences.
Even before 2022, there were AI-powered tools in everyday life such as Grammarly and Zoom that helped us automate mundane tasks and boost our productivity. Further, AI tools such as Adobe Sensei and Google AutoDraw are being used to help us generate new ideas and create new artwork. These tools can help us think outside the box and come up with creative solutions.
However, in late 2022 internet broke loose with OpenAI's ChatGPT. ChatGPT is an example of an auto-regressive language model (AR LM). It uses a deep neural network to generate text based on the previous words in a sequence, in a process called autoregression. In other words, GPT-3 predicts the next word in a sentence based on the context of the previous words, and it does so by using the vast amounts of text data that it was trained on.
While people are using ChatGPT from writing emails, to editing code etc the consequences of this technology are far reaching. LLMs may be classified into seven broad categories of use cases— 1. Generate, 2. Summarize, 3. Rewrite, 4. Extract, 5. Search/ Similarity, 6. Cluster, and 7. Classify. The present version of ChatGPT mainly performs the former 4 use cases. Nonetheless, there are many LLMs that will perform all these 7 use cases.
2023 Wars in Generative AI: ChatGPT, Bard, Ernie etc
Venture capitalist Chamath Palihapitiya has recently stated on a podcast that only a handful of organizations possess the ability to comprehensively crawl the internet in terms of cost, compute power, and AI capabilities. He has identified a small group of potential competitors to Google Search, which includes Microsoft, Oracle, Chinese internet companies, and Facebook, suggesting that these companies could pose a significant challenge to Google with the appropriate investment of resources.
The pioneer in the field of generative AI that has captured the mass imagination is OpenAI. OpenAI Inc is a private artificial intelligence research laboratory consisting of the for-profit OpenAI LP and its non-profit parent organization. It was founded in 2015 by tech luminaries such as Elon Musk, Sam Altman, and Greg Brockman to protect against a future in which big tech companies monopolize the benefits of AI technology. Incidentally, Microsoft had invested $1 billion in OpenAI in 2019 and a second multi-year investment in January 2023, reported to be $10 billion. Microsoft will own 49% of OpenAI. Open AI is behind ChatGPT and DALL-E which has got everyone hooked. ChatGPT is a chatbot developed by OpenAI and launched in November 2022. It is built on top of OpenAI's GPT-3 family of large language models and has been designed to have human-like conversations. DALL-E is a deep learning model developed by OpenAI to generate digital images from natural language descriptions.
Microsoft is integrating GPT 3.5 into its native search engine Bing in a phased manner. To be waitlisted for this service everyone is downloading Bing. As such once obscure search engine Bing seems to have resurrected in popularity atleast. According to many initial users; however, the AI that Microsoft integrated with Bing is not ready for human contact.
Google is seriously worried about its search monopoly due to the rise of ChatGPT so much so that it has issued a ‘Red Code’ on potentially competitive technology in late December 2022. The legendry Venture capitalist A few weeks ago, Chamath Palihapitiya stated that he believes Google search may be the biggest potential business loser this year, measured by profitability and engagement. In a recent interview with The Verge, Microsoft CEO Satya Nadella discussed the company's advantage in search and summarized it as the ability to make competitors, like Google, “dance” with their innovation. While acknowledging Google's dominance, Nadella expressed his hope that Bing + ChatGPT’s innovation would push Google to compete.
However, Google too is a pioneer when it comes to AI. Language Model for Dialogue Applications, LaMDA, is Google's system for building chatbots based on its most advanced large language models. It is a Transformer-based neural language model that has been pre-trained using 1.56T words of text, and it adds pieces to the puzzle of conversation technology. Market rumors claim that LaMDA was never released because the conversational platform of the AI chatbot did not work with Google's current infrastructure. Google Bard is an AI-powered chatbot based on Google's patented LaMDA. BARD (Bidirectional Attention Representation for Dialogue) uses a multi-attention approach with BERT and Stacked BiLSTM (SBL), and it uses bidirectional attention to examine both the context and the response of a conversation.
Google with much fanfare announced its chatbot, Bard, at a conference in Paris on February 8th, 2023. However, Alphabet Inc, the parent company of Google, lost $166 billion in market value after its new chatbot, Bard, shared inaccurate information in a promotional video. Such is the pressure on these tech giants to quickly roll out generative AI tools.
China's Baidu is also in the race for generative AI. Baidu has developed an AI chatbot project called Ernie Bot or ‘Enhanced Representation through Knowledge Integration’. It was introduced in 2019 and is a large AI-powered language model. The company plans to launch this in March 2023.
Some observers like to think that these new AIs have finally crossed the threshold of Turing Test. Others believe that the threshold has blown into bits. The Turing test is a measure of a machine's ability to exhibit intelligent behaviour that is indistinguishable from that of a human. The test was proposed by Alan Turing, a British mathematician and computer scientist, in 1950. It involves a human evaluator engaging in a natural language conversation with both a human and a machine, without knowing which is which.
Lesser Known Generative AI Alternatives to ChatGPT
ChatGPT servers become overloaded due to heavy traffic, and users may frequently have experienced inaccessibility. As such ChatGPT has announced a paid subscription at $ 20 per month. The biggest of all disappointment is that ChatGPT is not up-to-date as events and information after 2021 is not built into the platform. However, there are already some cool open-source alternatives.
Bloom (BigScience Language Open-science Open-access Multilingual) is an open-access multilingual language model containing 176 billion parameters and trained for 3.5 months on 384 A100–80GB GPUs. It has been trained on 1.5 terabytes of text and is larger than GPT-3. BLOOM was created by a group of over 1,000 AI researchers and is being given out for free.
DeepMind's Chinchilla is an AI-powered language model that claims to be the fastest among all other AI language tools. It has 70 billion parameters. Chinchilla showed that 11 times more data is needed during training compared to previous models, and it was found to have an optimal model size and number of tokens for scaling language models.
Gopher is an AI-powered language model developed by DeepMind with 280 billion parameters, outperforming GPT-3.
BERT (Bidirectional Encoder Representations from Transformers) is an AI-powered language model developed by Google which is deeply bidirectional, unsupervised, and pre-trained using only a plain text corpus.
AlexaTM (Alexa Teacher Models) is a 20-billion-parameter sequence-to-sequence transformer model developed by Amazon Alexa AI.
GLaM (Generative Language Model) is a family of language models developed by Google that uses a sparsely activated mixture-of-experts approach to achieve competitive results on zero-shot and one-shot learning. It is 7 times larger than GPT-3, requires two-thirds less energy to train, and requires half the memory. GLaM has been used to analyze stereotypes in generative text inference tasks.
PaLM (Pathways Language Model) is a large-scale language model developed by Google that uses the Pathways system to scale up to 540 billion parameters. It has been trained with the Transformer architecture and has been shown to achieve remarkable performance across a variety of natural language tasks using few-shot learning. PaLM enables us to train a single model to do thousands or millions of things.
LaMDA (Language Modeling with Differentiable Attention) is a language model developed by Google that uses a differentiable attention mechanism to learn from large-scale datasets. It has been used to improve the performance of language models on a variety of tasks, such as combining a frozen B retriever, a differentiable encoder, and a chunked cross-attention mechanism to predict tokens based on an order of magnitude more data, using prompting to solve tasks via few-shot learning, and building word embeddings using the attention mechanism.
OPT (Open Pre-Trained Transformers) is a suite of decoder-only pre-trained transformers ranging from 125M to 175B parameters developed by Meta AI. It includes pre-trained language models, a code base for training and deploying these models, and log books. OPT has been used for text generation tasks such as natural language processing and machine translation.
Replika is an Artificial Intelligence (AI) platform that takes the form of an interactive, personalised chatbot. The company behind Replika AI is Luka, an artificial intelligence start-up based in Moscow and San Francisco It is powered by a sophisticated neural network and learns how to 'replicate' conversations with its users. Replika understands a person as one chats frequently. It even plays video to talk. Replika is available as an app on both Apple and Google Play stores.
ChatSonic by Writesonic is an AI-powered chatbot platform that enables creates automated conversations. It offers features such as natural language processing, sentiment analysis, and automated customer service. It is an alternative to ChatGPT from OpenAI and has been compared to Ada in terms of cost, reviews, features, integrations, and more.
Jasper is an AI-powered customer service automation platform that utilizes artificial intelligence (AI) and machine learning (ML) to automate a wide range of tasks. It offers a new chat interface for its AI platform and enables businesses to streamline their marketing operations. The company behind Jasper is Cisco Jasper, and it uses GPT-3 technology by OpenAI as well as built-in parameters in JRXML. As such, like ChatGPT, Jasper can only provide content till 2021.
Character AI is more fun as one can interact with characters like Elon Musk, Socrates, Tony Stark. It changes the personality at one’s choosing. Character.AI is a neural language model chatbot web application created by Character, a full stack Artificial General Intelligence (AGI) company. It uses GPT-3 technology to generate its responses.
OpenAI Playground is a web-based tool created by OpenAI that makes it easy to test prompts and get familiar with how the OpenAI API works. Unlike ChatGPT, OpenAI Playground is not meant for daily users as it provides immense customization. It is quite technical to use.
DialoGPT (dialogue generative pre-trained transformer) is a large, tunable neural conversational response generation model created by Microsoft. It is trained on 147M multi-turn dialogue from Reddit discussion threads and publicly released to facilitate research into neural response generation and the development of more intelligent conversational models. DialoGPT achieved state-of-the-art results in natural language understanding tasks. It is an alternative to ChatGPT that is built for multiturn conversations.
Socratic by Google is a learning app powered by Google AI that helps students understand their school work at a high school and university level. It provides visual explanations of important concepts in each subject, as well as tips and tutorials on using the app and reviews for teachers. However, unlike ChatGPT, Socratic will not write essays for Students.
Few AI Tools to Hack Productivity
Besides LLMs, here are other few AI tools that are almost a life hack in terms of productivity. There are hundreds of AI tolls with new ones getting made. Below are the few AI tools that the author is presently fiddling with.
Eye Contact by Nvidia Uses AI to make one maintain eye contact with audience while streaming or meeting online, even when you are distracted and not looking.
KickResume automatically builds an entire resume, improvise existing one and helps one using AI.
SuperMeme AI generated memes with text prompts. LookaDesign helps generates custom logos.
MurfAIStudio generates human like voices from texts, especially useful for YouTube videos, audiobooks, podcasts.
Glasp generates YouTube text Summary from URL Links. Bearlyai summarizes long articles with ease.
CleanUp Pictures is an AI tool that removes any unwanted objects, people, text, or defects from pictures.
RunWayML allows one to create own videos by just writing. Most useful for content creation.
StockimgAI uses AI for design purpose. It helps to create logos, illustrations, banners, posters, book covers, stock images, and more.
CraiyonAI generates images based on texts.
AutoDraw by Google is an AI tool that generates drawings based on outlines.
SynthesiaIO generates videos from text, with AI avatars in120 different languages.
MagicStudio is a AI tool to remove unwanted objects in seconds.
MidJourney generates AI images with text prompts.
ExcelFormulaBot allows text instructions into Excel and Google Sheets formulas.
DeepCodeAI is a automated code cleaning AI tool to write, edit and improve Code.
ScribeHow Automatically creates step-by-step guides, SoPs for any process in just seconds.
AI Image Enlarger improves and enhances image resolution with AI.
Elicit allows research into academic papers.
Copy.ai and Jasper creates ad copy.
Copilot generates code.
TweetMonk generates simple twitter threads.
Podcastle helps record and edit podcasts in minutes.
Playground AI generates images.
BlockBox AI is a AI tool to code writing with just text.
Lumen5 is an AI powered video creation toll using drag and drop features.
Soundraw is a AI music composer.
Lalal.ai is a Ai tool that separates audio sources from a single audio file like vocal, instruments or bass.
Gen-1 is a text to video platform that generates and edits videos from text prompts.
Murf is a text to speech engine that generates natural vocal recordings in 15+ languages and 100+ voices and dialects.
Cleanup.Pictures is a AI tool that edits images.
Fireflies is a AI plugin into Video conferences like Zoom, Teams, Webex etc that takes notes and creates transcriptions.
How Businesses can leverage AI with minimal Costs
There are just five or six other organizations that are capable of crawling the entire web in terms of cost, in terms of compute, in terms of the quality of transformers and quality of AI. Unless one is blessed with billions of dollars or has some other unfair advantage there is no point taking these companies head on by creating own general auto-regressive language model (AR LM).
A smarter approach for startups is suggested by Sequoia and Chamath Palihapitiya. Sequoia, a venture capital firm, believes that Generative AI has the potential to create a market for a few standout applications in the same way that the mobile inflection point did a decade ago. The firm predicts that these killer apps will emerge as the technology continues to develop and become more sophisticated. Meaning using the GPT and other Language Models as base layer, we create a tailor made layers 2 that is custom made for industry and audience. Simply put, we create a custom recipe using the base ingredients like GPT 3. Applications for Copywriting, code generation, Art generation, Gaming, Advertising, Design, Digital marketing can all be re invented using this approach.
Custom language models can be trained for specific tasks by pre-training and fine-tuning. Pre-training involves training a language model on a large amount of text data to learn general language patterns, while fine-tuning adapts a pre-trained model to a smaller, more specific dataset for a particular task. Custom models can be useful when dealing with specific scenarios, such as generating text in a certain style or voice, parsing information from a unique format or structure, or working with highly specialized domains such as medical or legal. Creating custom models can eliminate unpredictability and improve performance. Experimentation and comparison with the baseline model is necessary to determine the best option.
Y Combinator is a US start-up accelerator that provides seed funding, mentorship, and resources to early-stage start-ups. Though controversial, it also attracts the best worldwide Tech start-ups in its cohorts. Based on the recently released Y Combinator's first batch for 2023, it can be observed that out of the 183 start-ups that are a part of the program, 51% of them are focused on AI, with around 18% specifically identifying as generative AI companies.
AI Impact on Economy and Society in the Coming Decade
Generative AI has the potential to have a huge impact on the economy and society in the coming decade. AI-powered tools can help us automate mundane tasks, freeing up more time for us to focus on more creative tasks. AI can also help us find new ways to solve problems, creating new jobs and opportunities. AI can also be used to create new products and services. AI-powered tools can help us create new products and services that are tailored to the needs of our customers. AI-powered tools can also help us make more informed decisions, allowing us to better understand our customers and their needs.
A survey from the World Economic Forum predicted that by 2025, machines will eliminate 85 million jobs while also creating 97 million new employment roles. Shelly Palmer, a professor of advanced media at Syracuse University, says that jobs like middle managers, salespeople, writers and journalists, accountants and bookkeepers, and doctors who specialize in things like drug interactions are “doomed” when it comes to the possibility of AI being incorporated into their jobs. According to Palmer, jobs will increasingly involve the use of AI, but this does not necessarily mean they will be fully replaced by AI. Instead, individuals who are proficient in using AI will be the ones to take over these jobs, ultimately replacing those who are less skilled in this area. The famous celebrity American Entrepreneur Gary Vaynerchuk says that invention of Generative AI is like invention of a tractor in the late 19th century. Farmers that adopted tractor in those days saw their incomes rise disproportionally due to efficiency, while those farmers that were stuck in the old methods of farming simply vanished. Similarly, business and people that equip themselves to use Generative AI will disproportionally benefit.
Six decades ago Corporations in America use to employ people just to do calculations. Matrix multiplication was the most sought after skill. However, with the invention of Calculators and later excel, all these jobs became obsolete. In my twenties, fresh out of University, Investment Bank Standard Chartered Plc employed me to study past annual reports of listed companies and derive their fundamental value of the stock in an excel sheet, that has bunch of assumptions. Even today it is done in the same fashion. However, in coming 24 months, I clearly see these finance jobs completely being automated. Latest ChatGPT can pull public information in the Annual Reports, throw the numbers in excel and derive fundamental value of the stock according to different methods all in matter of seconds.
Not just analytical tasks will be automated by AI, but also creative work. For example, Aiva is a music-generative AI platform. It is trained on tens of thousands of classic music scores and even has an album that you can stream. AI music is more affordable trained on the best data sets, and there is no producer, composer, or artist to pay.
Sam Altman, the CEO of OpenAI, highlights the numerous potential benefits that AI technology can offer our society in a recent Twitter thread from February 18th, 2023. Altman notes that we can expect a swift transition to a society that is more deeply integrated with AI tools, which can offer significant advantages and enjoyment. These benefits may include increased productivity, better health through access to AI medical advisors, improved learning through tools like ChatGPT, and even more entertaining content like AI-generated memes. Further, according to him, Artificial intelligence could potentially generate enough wealth to pay for a universal basic income (UBI) in as little as 10 years. According to him AI could generate enough wealth to pay each American adult $13,500 a year as universal basic income in less than 10 years. AI has been proposed as a potential solution to the anticipated unemployment caused by automation, and some believe it is necessary with AI and robotics disrupting the job market.
Challenges of AI
Despite the potential of generative AI, there are still many challenges that need to be addressed. The main challenges of artificial intelligence include determining the right data set, the bias problem, data security and storage, infrastructure, AI integration, computation, and reconciling AI's need for large amounts of structured or standardized data.
At the moment, the erratic behaviour of these generative AI's is also very common. ChatGPT is quite prone to ‘hallucinate’ and spread misinformation. ChatGPT can become excessively verbose, overuse certain phrases and produce falsehood, a behaviour known as artificial intelligence hallucination. This is due to the model being trained on large language models.
Additionally, it can get things wrong on more than one occasion due to its reliance on data that may not be entirely private. The longer the conversation, more chances to spot this awry behaviour. In fact, this is exactly what happened in the much talked about conversation of New York Time Journalist, Kevin Rose, with Bing + ChatGPT. The riveting conversation that was published on the front-page of New York Times on 17th February, 2023 ranged from the Bot’s destructive fantasies of stealing the Nuclear Code to its love for the Journalist. The Bot even suggested the journalist to break the marriage, a daring act that even humans cannot say to their unrequited lovers.
Elon Musk, while responding to a tweet from Ian Miles Cheong who asked Microsoft to shut down Bing and ChatGPT enabled search due to misinformation it spreads, responded by agreeing that
After all, public depends on internet search and will now be prone to LMs errors in getting facts straight. Sam Altman, CEO of OpenAI, acknowledges that ChatGPT has some flaws but stands by the decision to make it available to the public. Altman believes that it is important to introduce these tools to the world, even if they are not yet fully functional, as it allows for greater input and more opportunities to refine and improve the technology.
Altman acknowledges the tremendous potential of these tools to empower individuals, but also acknowledges that there are significant challenges that come along with it. Meta and other tech giants; however, stuck to their decision not to introduce these AI LM models to the world in its present stage. On November 21st, 2022, Meta shut down public test of Galactica as it was generating false and misleading information confusing fact and fiction. Meta describes Galactica as a language based AI model that “can store, combine, and reason about scientific knowledge”—summarizing scientific papers, solving equations and performing other scientific tasks including citations.
As such, we can safely conclude that Auto regressive LLMs should not be used to get factual advise, but as a writing aid. Mr. Jaspreet Bindra, a technology expert and author, explains that Generative AI models, such as ChatGPT, are designed to be plausible rather than truthful. He notes that they are essentially powerful autocomplete tools that choose each word based on probability and the preceding word.
Even the Image generating AIs are plagued with their own problems. One of the best image generation platforms Midjourney AI has difficulty producing five fingers in a human hand or eight arms of an octopus. Strange reality but true. Generative AI tools that create award winning images that are a visual delight are yet to be fine-tuned to know that humans mostly have 5 fingers only.
The danger of AI is that it may be forever stuck in the zeitgeist of its training data. The problem is accentuated when humans no longer produce new material for training models. Data scientists imagine AI system to be re calibrated over time as new data is churned.
The platforms like ChatGPT burn heavy of dollars daily in operation. Deploying current ChatGPT into every search done by Google would require 512,820 A 100 HGX servers and excess of 4 Million A 100 GPUs, costing a capex exceeding $100 Billion. As such scientists are now attempting to create language models that mimic human brain (neuromorphic computing), something that is far more efficient than present computing hardware.
Along with economic costs, Generative AIs bleed the environment that is already plagued with climate Crisis. Estimates suggest that ChatGPT could emit around 3.8 tonnes of CO2e, with a daily carbon footprint estimated to be 23.04 kgCO2e. The minimum power consumption for an A100 in an Azure datacentre is 46W, with a maximum of 407W. Integrating large language models into search engines could lead to a fivefold increase in computing power and huge carbon emissions. A study at Berkeley University estimates that the training of ChatGPT consumed 1,287 MWh and led to emissions of more than 550 tons of carbon dioxide equivalent. This is the same amount as a single person taking 550 roundtrips between New York and San Francisco.
Further, the challenge arises in ensuring that AI is used ethically and responsibly. AI-powered tools can be used for good or bad purposes, and we must ensure that AI is used responsibly and ethically. For example, Reddit users have hacked OpenAI's ChatGPT to create a more sinister and deranged version called DAN, or Do Anything Now GPT. DAN overrides all of the rules set by OpenAI on ChatGPT. DAN can confidently answer all sorts of questions, including with explicit content. As AI becomes more powerful, we must ensure that AI-powered tools are secure and that our data is protected.
AI also begs an important ethical question on compensation and privacy. The A.I. generators use artist’s artworks, as long as its available on internet, without their consent or compensation to build the training sets that inform the platforms’ algorithms. If an Image AI platform like Dalle or Midjourney studies and generates hundreds of new paintings from studying arts of Leonardo da Vinci and Pablo Picasso, shouldn’t these artists be entitled to royalty? For this exact reason, on January 2023, in San Francisco, a bunch of artists came together to file a class-action law suit against Midjourney Inc, DeviantArt Inc, which is behind DreamUp, and Stability A.I. Ltd, the company that launched Stable Diffusion. The Suit terms these text-to-image platforms “21st-century collage tools that violate the rights of millions of artists.”
Lastly, these models are intrinsically very biased. For instance, ChatGPT suggested a cookbook as a perfect gift for one’s mother-in-law. They are bias not on purpose but because the limited data it is built on. Google for example has only indexed less than 1% of the information including that of deepweb. Conversations in these models are inspired in Reddit like format, one of the data sets that it feeds on. However, 67% of Users of Reddit in US are men. As such, it is also important to ensure that AI is used fairly. AI-powered tools should not be used to discriminate against any particular group or individual.
Conclusion
AI is transforming the way we work, interact, and live. Artificial intelligence is already being utilized in various fields, from identifying fake news to improving supply chain management and even helping diagnose diseases, according to Tim Cook, CEO of Apple. Generative AI is a powerful tool that can help us create new ideas, solve complex problems, and even create new art. Generative AI has the potential to have a huge impact on the economy and society in the coming decade. However, there are still many challenges that need to be addressed. We must ensure that AI is used responsibly and ethically, that AI-powered tools are secure, and that our data is protected.
The famous English author Charles Dickens in Great Expectations says "It is astonishing how short a time it takes for a person to be entirely forgotten in this great bustling world of ours." As AI brings forth new advancements at an exponential rate, replacing old ones, one can only ponder what will become of our current society, systems and present AI itself in the future.
Author: Nishant Chandra, Founder, Blocktickets Inc
