Generative AI

Generative AI

Large Language models (LLMs) have been in news for last few years due to rapid advances in the AI technologies and models and availability of cloud scale AI infrastructure. They are used to generate human like text, images and video making them useful for various language generation, translation and chatbot type applications..

Applications

ChatGPT a recent chatbot application based on the one such language model GPT-3 (Generative Pre-trained Transformer 3) which has taken the world by storm since a free consumer version was released late last year. This initial success of OpenAI with a million signups in first 5 days even with a caution of incorrect, biased or even occasional harmful results has not stopped the euphoria about the power of the technology and its wider applications. Generative AI use cases are widespread from ideation and summarising to content writing e.g. emails, reports, essays etc. and even writing and debugging code. This will have an impact on variety of industries and service providers involved in knowledge work.For businesses out of the box use is limited but integrated business workflows incorporating Generative AI based technologies in a secure, monitored way could increase automation capabilities resulting in reduction in cost and increase speed of businesses. There is however a downside of factual errors due to hallucination, copyright infringements and sensitive business information leaks which will need Quality control in a business context. ChatGPT has taken a very public approach to the models developed to monetise the high costs of development especially the cloud computing costs to train the models. Most other players have been essentially using these applications to AI enable in-house products e.g. Google (Search, Workspace), Microsoft (Translation, caption) Salesforce (AI enabling CRM), NVIDIA (testing of new GPU hardware). There is an additional factor of reputational risk due to these models misbehaving, limiting common broader use cases. Recent examples of this are Facebook AI’s ‘Galactica’ targeted at researchers and students which had to be withdrawn three days after public launch in 2022 after biased and false results and Microsoft research’s ‘Tay’ in 2016 which was shut down in 16 hours after it generated racist content. Research in content filtering technologies and data is helping mitigate some of these issues including the use of Reinforced Learning from human feedback (RLHF) and even crowdsourcing of human feedback being used by OpenAI promoting “ChatGTP feedback contest’ to help refine its content filters.

Competitors

Open-AI is a research company with some of the top researchers in Deep Learning, Robotics, CNN/RNN with open source roots . It’s competitors are research teams of some of the largest Silicon Valley companies. Google (Deepmind and Brain) are pioneers in Machine/Deep learning research with expertise in unsupervised/generative, reinforced learning models. It also developed industry standard AI infrastructure i.e. TensorFlow but are lately doing lot of research in AI model general purpose algorithms which are then applied for a number of tasks e.g. LaMDA, Sparrow, PaLM, Imagen etc. Some of these models like LaMDA (Language model for Dialog applications) and Imagen (text-to-image generator) have performed very well in comparison with OpenAI in controlled environment. Facebook AI (FAIR) focusses on the Language models e.g. Galactica and computer vision applications and has a big focus on Metaverse applications for the parent company. They have also developed AI infrastructure with Pytorch. Microsoft Research ( natural language processing), IBM ( Watson cognitive computing ),Salesforce ( Einstein product suite),NVIDIA (Megatron) , Baidu etc. also have substantial AI research teams for specific applications for content, chatbot and image generation but only a few have successful business models. A number of start-ups have been incorporating this AI research using API’s in specific applications. Stability AI is one such company with a product ‘Stable Diffusion’ which is open-source and has been incorporated in consumer applications like Canva which has more than 100 million users. Hugging Face is another company which has created an AI community to democratise AI models and datasets to build applications.

Future

Google has been put on notice by the OpenAI work given their founding mission to “organise the world’s information and make it universally accessible and useful”. In a statement Google has hinted on similar innovations in 2023 to protect a mid-term threat to Search advertising revenue of $208 billion which was 81% of overall revenue in 2021. Microsoft investment in the OpenAI is interesting given the OpenAI’s ‘non-profit’ origins in 2015 with a mission of developing “digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return” transitioning to a ‘capped profit’ company allowing investor return up to 100 times of the original investment. With the initial investment of $1 billon in 2019 for exclusive licensing of GPT-3 models in cash and cloud credits for the use of Azure AI services it is planned additional investment of $10 billion in 2023 . Microsoft has incorporated OpenAI products i.e. DALL-E, ChatGPT, Codex in Azure OpenAI service including other internal products e.g. GitHub co-pilot for developers, PowerBI natural language and Microsoft Designer for creators. Microsoft Bing and Office products ( Word,Excel,PowerPoint) will also use the OpenAI technologies for image and content generation. While Google and Microsoft will compete to release chatbot products using cutting edge AI technologies the research in this area is very fast moving and it will be hard for them to keep pace with independent research companies. Researchers will push the boundaries to innovate which established Big Tech companies will not be able to achieve in the short term.

Conclusion

Strides made by Generative AI based applications do point to the fact that ‘Artificial Intelligence’ is going to be one of the key technology battlegrounds in this decade. We already have made rapid progress in the use of these technologies in various products and services in common use without even noticing. With the maturing of the AI technology and unlimited computing power we are going to see a more fundamental platform level shift which will have profound impact on ‘automation’ and ‘knowledge work’ in the most industries.