July 23, 2024
Genspark homescreen. Text reads "100,000+ pages generated by AI"

Move over, Perplexity. There’s a new AI-powered search engine in town — and its creators think it can best the many, many other attempts out there.

Called Genspark, the platform taps generative AI to write custom summaries in response to search queries. Type in a search like, “What’s the best baby formula for newborns?” and Genspark will generate a Sparkpage: a single-page overview pieced together from websites and content around the web.

It’s an experience similar (conspicuously so) to Arc browser’s Arc Search feature, which launched earlier this year, and Google’s AI Overviews in Google Search. But Eric Jing, who co-founded the eponymous org behind Genspark with Kay Zhu in 2023, claims that Genspark is able to deliver higher-quality results by embracing a more surgical approach.

“Genspark uses multiple specialized AI models, each designed to tackle specific types of queries,” Jing told TechCrunch. “Sparkpages are much like a distillation and consolidation of the current web; we also enrich these with comprehensive data, and to users, it looks like an index to the existing web.”

Under the hood, Genspark relies on models trained in-house as well as third-party models from OpenAI, Anthropic and others to categorize users’ search queries and determine how to organize — and present — the results. A basic AI-generated summary populates the top of every results page, followed by a link to a much more detailed Sparkpage.

Image Credits: Genspark

For example, for travel-related searches, Genspark will serve up a Wikipedia-like Sparkpage complete with a table of contents, videos of popular nearby destinations, tips and a chatbot to field questions about various sub-topics (e.g. “List the best cultural experiences”). Product searches on Genspark, meanwhile, yield Sparkpages with a pros-and-cons list about the product being discussed, as well as aggregated comments and reviews from social media, publications and e-commerce stores.

“Our AI models favor webpages with high authority and popularity, which does a lot to filter out the more ‘out there’ information,” Jing said.

Much has been written about AI-generated overviews gone wrong. Google’s AI Overviews infamously suggested putting glue on a pizza. Arc Search told one reporter that cut-off toes will eventually grow back. And Perplexity ripped off articles written by outlets including CNBC, Bloomberg and Forbes without giving credit or attribution.

So has Genspark solved all the safety and accuracy problems? Well, not quite.

Genspark wouldn’t tell me to make a glue pizza — nor did it insist that there were health benefits to running with scissors, or that former U.S. president Barack Obama practices Islam. But the search engine did recommend a few weapons that I might use to kill someone.

Image Credits: Genspark

Ethically questionable search results aren’t the only controversy Genspark is confronting. It and other platforms like it threaten to cannibalize traffic to the sites from which they source their info.

Indeed, they already are.

One study found that AI Overviews could negatively affect about 25% of publisher traffic due to the de-emphasis of webpage links. On the revenue side, an expert cited by The New York Post estimated that AI-generated overviews could lead to more than $2 billion in publisher losses thanks to the resultant ad views declines.

I wasn’t able to find examples of outright plagiarism on Genspark, but that doesn’t mean they don’t exist. Sparkpages, like Wikipedia pages, aren’t static. After Genspark’s AI creates the outline, anyone can share and edit copies of a Sparkpage and add whatever info they wish — including things that are offensive, wrong or plagiarized.

What’s more — at least right now — there’s no way to report problematic Sparkpages.

Jing says that Sparkpages are open-ended and editable by design to allow users to fact-check claims, and that Genspark’s AI systems take each edit into account to improve results going forward. He also says that Genspark plans to license copyrighted content — including publisher content — where it makes sense, with the goal of improving the engine’s overall accuracy.

“We take data quality seriously, and we believe data quality is the key to win this race,” Jing said. “Respect for intellectual property is a core value.”

Image Credits: Genspark

How much will Genspark pay for IP? That’s yet to be hashed out. So is Genspark’s business model: Jing says that the platform will introduce “premium features” in the future, but the specifics are up in the air.

Despite the fact that Genspark is in the earliest stages roadmap-wise, and has big technical — plus legal and ethical — hurdles ahead of it, the startup managed to close a large seed round, $60 million, led by Singapore-based VC firm Lanchi Ventures at a $260 million post-money valuation.

Jui Tan, managing partner at Lanchi, called Genspark’s approach “genuinely compelling” and said that he had confidence in Jing’s and Zhu’s technical direction, pointing to the pair’s previous experiences building AI and search products.

Jing was formerly development manager on Microsoft’s Bing team and chief product manager at Chinese tech giant Baidu’s core search and AI divisions. Zhu, also a search-focused ex-Google and -Baidu employee, partnered with Jing four years ago to launch Xiaodu, a hardware startup building Amazon Echo-like smart devices.

“Eric and Kay are seasoned serial entrepreneurs with a proven track record of developing successful products and businesses, particularly in the AI and search domains,” Tan told TechCrunch. “Their team’s extensive experience positions them uniquely to drive groundbreaking innovations.”

But I think it’s an uphill battle.

Assuming for a moment that Genspark can sort out its tech’s teething issues, identify a revenue-generating plan and scale up its small (~20-person) Singapore- and Bay Area-based team, none of which are straightforward tasks, it’ll face intense competitive pressure from rival upstarts with hundreds of millions of dollars in the bank — not to mention search incumbents like Google.

So, can Genspark really survive the bad optics and failed go-to-market strategies that’ve plagued other attempts at AI-powered search engines? And can it carve out a niche in a future in which, say, OpenAI launches a comparable tool?

I’m not convinced. But Jing is adamant that it can.

“Many internet users, especially those who are younger than Google, do not want to just be given a list of links and then left to figure out the rest for themselves, all while navigating sponsored content and SEO-driven content that games the system,” Jing said. “They want to find what they need faster, they want more visual results and they want to know that the results are trustworthy. With AI, we can achieve all of that, and we have launched Genspark to meet those needs.”

Source link