Wednesday 15 February 2023

ChatGPT searchbots are broken, can NeevaAI fix it?

The long-foreseen AI chatbot war is officially on. And, while tech companies are racing to modernize their services, experts are warning about the dangers of rushing this ChatGPT frenzy into search.

Ad-free search engine Neeva is promising to fix AI-powered search, though, by combining the power of generative AI systems with its authoritative and unbiased search technology. 

This means that, unlike ChatGPT apps, its results are real-time. Plus, its AI summaries are backed up with sources' links so that users can check their reliability themselves. 

First launched in December 2022 in the US, NeevaAI beat the Big Tech giants Microsoft and Google in developing its own ChatGPT-like search assistant. Microsoft Edge's ChatGPT tool was, in fact, announced just about a week ago, to complement Bing's AI search. So too was Google's "experimental conversational AI service" Bard.

And now, NeevaAI has now enlarged its reach outside the US. Starting from February 13, 2023, users living in Canada, the UK, Germany, France and Spain can enjoy the power of its AI-powered search results. 

But, can NeevaAI really solve the AI-powered search conundrum? 

See more

From a search to an answer engine 

"Search, which is a daily use function, has increasingly become more about serving advertisers and not about serving users. Neeva flips the entire model around, and creates an AI powered, ad-free search that is entirely built for users," Neeva's CEO Sridhar Ramaswamy told TechRadar.

Bringing back a user-first experience was, in fact, the main reason why Ramaswamy quit his job as Google Head of Ads to develop its ad-free private search engine back in 2018. Now, with the explosion of software-like ChatGPT, he sees a further opportunity to better serve Neeva's customers. 

The algorithms powering generative artificial intelligence software, like ChatGPT, use large language models (LLMs) to create new content. NeevaAI combines this innovative capability with the authority and timeliness of its software to turn "a search engine into an answer engine."

Such a system is built to interrogate a page, understand its content and decide whether or not it might be useful and/or authoritative enough to be cited as a source. It does that in real-time as the web changes, to provide a synthesized single answer with linked sources pulled together from the most relevant sites for a query.

Ramaswamy explained that NeevaAI mostly uses its own models that its team can pre-train or fine tune for custom tasks like question answering or summarization. 

These are integrated with other systems, like GPT3.5 and Anthropic's Claude, for generating training data (used to train its own models) as well as super-specialized tasks in the overall pipeline. 

Despite this new accessibility of generative AI systems being one of the biggest innovations in tech of the last few years, the first iterations are far from perfect. 

Software like ChatGPT comes, in fact, with two significant downsides. First of all, its results are never sourced nor referenced. Second, these aren't retrieved for a real-time stack of data. The results will then be less relevant and more difficult to be verified. 

While some other Big Tech firms are launching an AI-powered search chatbot for trying to fix the latter point, Neeva's unique feature provides citation cards linked to its results. These swipeable cards highlight authoritative information about the researched topic, while suggesting important research questions for the searcher.  

"Our goal has been to responsibly integrate AI and provide authoritative answers that users can trust."

NeevaAI research query example

(Image credit: Neeva)

Ramaswamy believes that Neeva's ad-free model can be beneficial here, too.

He said: "Any time you move a product into a mode of asking a question and giving a single answer - ads collapse. What we are seeing is the tension both in realizing the full potential in this technology and the fact these companies have large revenues to protect.

"Not only that, but we were able to deliver this six weeks before big tech announced their own versions (which still aren’t yet actually available)." 

After receiving positive feedback following its launch in December, the company has now released its AI search assistant also across Canada and a few European countries. However, people living outside these nations could enjoy the feature as well by simply playing around with some settings.

Furthermore, Neeva is so far the only one in the game committed to support publishers and content creators - those suffering the most from a rise in AI chatbots. Besides offering them 20% of its topline revenue when their content is used to directly answer a query, Neeva is also exploring ways to help publishers leverage AI into their sites for a better user experience.

AI chatbot race: is NeevaAI really that good?

But, while it's true that Neeva probably opened the AI chatbot race, it isn't certainly the only player in the game today. 

Among the biggest names, Microsoft has just launched its new AI-powered Bing based on the OpenAI system. Google responded with Bard. The latter uses Google's own language learning model Lamda, which was deemed as 'sentient' by one of the engineers who worked on it. 

The smaller You.com browser has also integrated its site with a similar AI chatbot, and China-based search engine Baidu is rolling out its Ernie in March. Outside the search world, tech giant Meta already launched Blenderbot last summer across the US. 

Tech companies seem to be racing to stick their fingers into the AI pie. However, the results we are seeing so far are anything but perfect. 

That's because ChatGPT-like software comes with some intrinsic limitations.

For example, something known as jailbreaking can allow the AI chatbot to develop dangerous and abusive ways to reply to users. But, perhaps, the most problematic one is the tendency to sell misinformation as facts. This issue has already cost Google over $100 billion, as Bard responded to a query with false information during its launch event. 

See more

On that point, Ramaswamy said that Neeva works differently compared to pure generative AI because it uses a more constrained system which will decline to answer in case he cannot find a suitable reply. 

However, this doesn't mean that Neeva will never report misleading or false information. 

For instance, when we asked NeevaAI the same question that tripped up Google's Bard in its launch event - "What new discoveries from the James Webb Space Telescope can I tell my 9 year old about?" - we were presented with the same exact answer that Bard gave. 

This appears to be because NeevaAI found valid sources reporting the news, yet was unable to discern that these sites were actually reporting that this answer was wrong. 

"This is why citations are front and center in Neeva's AI answers," said Ramaswamy. "Just like no website is believable in its entirety, thoughtfulness and skepticism is essential to surviving in this modern world."

https://ift.tt/KNjaqlH

No comments:

Post a Comment