Deep learning expert says GPT startups could be in for a very rude awakening

Generative AI exploded into the mainstream last year. Led by Elon Musk, the co-founder of OpenAI – the creator of DALL-E 2, a text-to-image generator, and ChatGPT, an impressive text generation system – the industry has absolutely exploded as these generative tools and others, particularly the imaging systems Stable Diffusion and Midjourney, have dazzled investment firms and the general public alike.

“Generative AI is on track to not only be faster and cheaper, but in some cases better than what humans create by hand.” reads a blog post by top investment firm Sequoia Capital, published September 2022. “If we allow ourselves to dream several decades ahead, it’s easy to envision a future where generative AI is deeply embedded in the way we work, create, and play is embedded.”

But despite the high investment – an estimated $1.37 billion in 78 deals in 2022 alone – after The New York Times — that VCs are throwing at generative AI companies, not everyone in this space is convinced that these generative machines are truly the earth-changing force that both creators and investors think they are.

“The current climate in AI has so many parallels to 2021 web3 that it makes me uncomfortable,” wrote François Chollet, an influential deep learning researcher at Google and creator of the deep learning system Keras, in a glowing twitter menace. “Narratives based on null data are taken for granted.”

In other words, Chollet argues that in an eerily similar way to the blockchain bubble, hype – as opposed to solid data and proven results – leads the industry. And if Chollet is right, given the current situation over at Web3land? A failure in VC’s projected returns could have some dire consequences for the broader AI industry.

“Everyone expects ‘civilization-changing’ impact (and a 100x return on investment) over the next 2-3 years as a sure thing,” he continued. “Personally I think there is a bull case and a bear case. The cop case is much more conservative than what the middle person in my TL takes for granted.”

the bullfallhe believes is that “generative AI is becoming a widespread one [user experience] paradigm for interacting with most tech products.” But Artificial General Intelligence (AGI) – AI that operates at the level of a human or higher – remains a “pipe dream”. Startups built on OpenAI technology may not make us humans obsolete just yet, but they could well find a long-term role in certain niches.

the bear casemeanwhile, would be a scenario where large language models (LLMs) like GPT-3 would find “limited commercial success in SEO, marketing and copywriting niches”. ultimately prove to be a “complete bubble”. (He offers that imaging would be far more successful LLMs, but would peak around 2024 “as an XB/y industry”.)

Still, Chollet believes the most likely case is somewhere in between.

But even Chollet’s best-case prediction still holds path not in keeping with the VC craze, where acolytes are writing checks to match their optimism about the technology — OpenAI, for example, is in talks to secure an investment deal that would take the company’s value to nearly $30 billion.

“It’s the new ‘mobile’ kind of paradigm shift we’ve all been waiting for,” said Niko Bonatsos, an investor at venture capital firm General Catalyst NOW. “Maybe bigger too.”

To the credit of investors, the algorithms are Cool. Text-to-image generators are truly impressive and open up broad new creative frontiers for people without Photoshop skills. At least with GPT systems, it’s a lot of fun to play around with.

However, they also have many problems. ChatGPT, for example, doesn’t always get it right with the very assertive statements it delivers, and experts fear the technology could make it very easy easy and efficient generate misinformation. And although industry CEOs are open about these programs being still in its infancythe very real potential for destruction and blurred creative lines that they present is hard to ignore, even against the backdrop of a bright – if still largely imagined – future.

And according to Chollet’s reasoning, it takes more than a product that’s cool and fun, or even very useful for niche products, to really be a “paradigm shift.” VCs may be taking a far greater risk than they realize, feeding on both a hype cycle of immature products and it rather than making measured calls about a situationally promising, if still fairly limited, burgeoning market.

“The fact that investments are driven by pure hype, by data-free narratives and not by actual sales data or first-principle analysis,” concluded Chollet’s thread. “The circularity of it all – hype fuels investment, which fuels hype, which fuels investment… unsupported narratives somehow become entrenched as self-evident, common wisdom simply because enough people repeat them often enough.”

“Everyone is starting to believe in the same canon (especially those who call themselves adversaries),” he said.

More on ChatGPT: There is a problem with this app that recognizes GPT written text: it is not very accurate

#Deep #learning #expert #GPT #startups #rude #awakening

Leave a Reply

Your email address will not be published. Required fields are marked *