Assessing Generative AI, the New Go To Buzz
First came DALL-E for image generation, then ChatGPT for text, code, almost anything; and now we are in a tsunami of startup investment activity
By John P. Desmond, Editor, AI in Business
The term “generative AI’ is emerging as the go to for describing the new wave that includes ChatGPT, DALL-E and other AI systems capable of generating text, images, music, code and other types of media.
The release by the OpenAI lab in early December of ChatGPT, an advanced search engine that can answer questions in clear, concise prose, caused a sensation, with more than a million people signing up to explore its use in a few days.
That was followed by a gold rush centered on OpenAI, which is in talks to complete a deal that would value the company at $29 billion, more than twice its value in 2021, according to an account in The New York Times,
Deal-making around generative AI companies is in overdrive. Jasper, a generative AI startup founded in 2021, raised $125 million in October, valuing the company at $1.5 billion. Stability AI, an image-generating company founded in 2020, raised $101 million in October, valuing it at $1 billion. Smaller generative AI companies such as Character.AI, Replika and You.com have also attracted interest from investors.
Investors placed a total of $1.37 billion into generative AI companies in 78 deals in 2022, according to data from PitchBook quoted by the NYT. That was about equal to the total investment of the previous five years.
What primed the pump for the current crescendo of potential AI hype, was the release in the spring of 2022 of DALL-E, the image generation tech from OpenAI. The system enables a user to generate photo-realistic images by describing what they want to see. Entrepreneurs went to work to engage investors with new ideas, when months later, ChatGPT was released.
“There is definitely an element to this that feels like the early launch of the internet,” stated Lonne Jaffe, an investor at Insight Partners, to the NYT. His firm is encouraging companies to consider incorporating generative AI in their products and services. “It’s hard to think of a company that couldn’t use it in some way,” he stated.
One venture firm was quoted as estimating that 450 startups are now working on generative AI products or services
Roots of Generative AI
Some worry the hype around generative AI has moved ahead of reality.
It may help to trace where the tech came from. A Generative Pretrained Transformer (GPT) is a type of large language model (LLM) that uses deep learning to generate text as if from a human, according to an account from the World Economic Forum (WEF). They are referred to as generative because they can generate new text based on the input they receive; they are pre-trained from a large collection of text data, before being fine-tuned; and the transformer is a type of neural network architecture for processing input text and generating output text.
This evolution has led to the potential of generative AI for a much broader range of tasks than language translation, text summaries and text generation. Currently, the WEF authors stated, use cases under discussion include newly architected search engines; complex algorithm explanations; personalized therapy bots; tools to build apps from scratch; explaining scientific concepts; writing recipes and writing college essays.
DALL-E and other text-to-image programs are likely to have a major impact on how art, animation, gaming, movies and architecture are rendered. Optimists see generative AI assisting the creative process of artists and designers, and speeding up the creation phase. Pessimists see jobs lost.
Legal Challenge to How AI Collects Code
In complex sciences, generative AI has the potential to transform. The GitHub Copilot from Microsoft, based on OpenAI’s Codex model, suggests code and assists developers in completing programming tasks with a higher degree of automation. Copilot is also at the center of a class action lawsuit filed in US federal court that claims open source licenses are being violated. The suit claims “software piracy on an unprecedented scale,” in the way AI is being used to corral large sections of licensed code without providing credit.
Unveiled in June 2021, Copilot is trained on public repositories of code scraped from the web, many of which are tied to licenses that require anyone reusing the code to credit its creators, according to a recent account in The Verge.
“We are challenging the legality of GitHub Copilot,” stated programmer and lawyer Matthew Butterick, who filed the lawsuit with the help of the San Francisco-based Joseph Saveri Law Firm, in a press release. “This is the first step in what will be a long journey. As far as we know, this is the first class-action case in the US challenging the training and output of AI systems. It will not be the last. AI systems are not exempt from the law. Those who create and operate these systems must remain accountable.”
Another issue is the ability of LLMs to generate false or misleading content. “Such capabilities can be misused to generate fake news and disinformation across platforms and ecosystems,” stated the authors, Benjamin Larsen and Jayant Narayan, both project leads for AI and ML for the WEF. Given the widespread effect of generative AI, “it is important for civil society and for policymakers to weigh in,” the authors stated.
Hype Explosion Data Points
A recent account in Axios references a “hype explosion” around generative AI. Some facts:
At the CES trade show last week in Las Vegas, 579 exhibitors were listed under the AI category:
That number was more than double those listed under Metaverse (176), Cryptocurrency (19), and Blockchain (55) combined.
Venture capitalists invested over $2 billion in generative AI $2 billion in 2022, according to the Financial Times.
Lots of experimentation is going on:
OpenAI said in September that DALL-E 2 has over 1.5 million users, creating over 2 million images a day.
Tracking the Backlash
And a backlash is beginning:
AI art has been banned from Getty Images and multiple anime conventions.
Admins at Stack Overflow have banned text or code generated from ChatGPT, because of its propensity to create confident chatbots responses that are incorrect.
New York City public schools have banned access to ChatGPT on its devices and networks, out of fear students will use it to write essays.
Clément Stenac, Chief Technology Officer, Dataiku, an AI and machine learning company founded in Paris in 2013, recently based a column in Forbes on his experience asking ChatGLPT to produce a commentary on the impact of generative AI. It was an impressive response. He stated, “AI is starting to blur the lines between human and machine, getting us tantalizingly closer to something that previously seemed only possible in science fiction: a computer that is seemingly capable of human reasoning and creativity.”
He adds, “However, that is simply not the case.” The generative AI platforms are taking advantage of a human trait and not generating illuminating new ideas, he suggests. “It’s not that these systems are ‘aware’ or even close to independent conscious thought. It’s that humans are quite predictable, which, in turn, allows generative AI platforms to present an extremely credible imitation,” Stenac stated.
Today’s generative AI is powerful and can provide a pretty good imitation of it, but it is not all-knowing. “However, it is precisely its ability to provide a pitch-perfect presentation of human thought that can fool us into thinking that sentience has arrived. And herein lies the con,” Stenac stated, advising that its output requires close human oversight and review.
Read the source articles and information in The New York Times, from the World Economic Forum, from The Verge, from Axios and from Forbes.
(Write to the editor here.)