Serial entrepreneur Wayne Chung demonstrates an AI understanding engine called Reasoner, which he claims can produce much more accurate and interpretable results than large language models like OpenAI’s o1 series, and at a much lower cost.
The AI industry is working hard to build understanding capabilities into this technology, in part to approach the holy grail of human-level or superhuman-level AI, and in part simply to overcome the inaccuracies that plague today’s undergraduates. The so-called hallucinations of Generative AI are one of the main factors holding back enterprise deployments, as is the inability to explain how LLM reaches its conclusions.
Reasoner aims to solve these problems with neurosymbolic artificial intelligence, a fusion of neural networks (the technology behind generative artificial intelligence) and more traditional symbolic artificial intelligence, which is based on fixed rules, logic, and human-made representations of relationships between things.
Chang has a long history in technology, starting with his founding of the file sharing service i2hub in 2004. In 2011, he co-founded Crashlytics, a ubiquitous mobile crash reporting tool that was acquired by Twitterwhere he became director of consumer product strategy. He went on to co-found the AI-powered accounting firm Digits (which Google acquired with Crashlytics in 2017), and last year he founded Patented.ai, an intellectual property-focused AI tool that, it turns out, also served as a pilot implementation of the Reasoner engine.
High Stakes AI
Patented.ai offers the ability to conduct automated searches of patent documentation and source code, identify potential patent infringement cases, and identify innovations that may be patentable. Given the high financial stakes of patent cases and the extremely time-consuming nature of determining whether infringement has occurred, there are clear opportunities for anyone who can automate the process, but also great risks if the system goes wrong.
In an exclusive interview with FortuneChang said Patented.ai’s initial reliance on LLM alone proved fruitless — attorneys who played with the system immediately noticed flaws in its results and rejected it. The company has also tried other common methods, such as search-augmented generation, which relies on external data sources to improve LLM performance (Google uses RAG for AI search results), but this also did not provide the required level of reliability.
This prompted a change in tactics that led to the development of Reasoner. “We didn’t really set out to build a reasoning engine,” Chang says. “That was not our mission at all.”
Reasoner does use LLM to help interpret language in texts—Chang says it doesn’t depend on which model it uses—but Reasoner’s core concept is adaptive dynamic knowledge graphs.
Knowledge graphs are widely used in technology. For more than a decade, Facebook’s knowledge graph provided the framework for building relationships between people, while Google enabled Search to answer basic factual questions. These repositories of established knowledge are clearly useful for providing correct answers to IBM queries DangerThe winning AI, Watson, was built on a knowledge graph, but these usually need to be manually updated to add new facts or edit relationships that have changed. The more complex the knowledge graph, the more work it entails.
Chung claims that Reasoner eliminates the need for manual updates, instead offering the ability to automatically build accurate knowledge graphs based on unstructured text fed into the system, and to have these knowledge graphs automatically reconfigure as they are added or changed information. (It should be noted that Microsoft earlier this year introduced GraphRAG, an attempt to use LLM-generated knowledge graphs to improve RAG results.)
In other words, you can insert a bunch of legal documents, and Reasoner will interpret them to build a knowledge graph containing the concepts in the documents and the relationships between them—with “full traceability” so that it’s easy for a human to check whether the facts accurately reflect what what is in the documents. This is where the concept becomes useful far beyond patent litigation.
In the demonstration to FortuneChang showed how Reasoner can take dozens of different OpenAI legal documents (from user and developer agreements to branding guidelines and cookie notices) and map their interdependencies. In the demo, this allowed for concise and detailed answers to the question of how a user can use the differences between OpenAI’s US and European terms of service to “avoid liability for harmful AI results.” Each step in the reasoning was explained—the logical steps were clear even to the non-technical eye—and Reasoner offered follow-up questions about the implications of the problem and ways to mitigate it.
Chang says Reasoner could also be used in a variety of other applications, from pharmaceuticals and advanced materials to security and intelligence. As such, he claims it could beat out offers from various other AI startups, such as Hebbia (a document retrieval firm that raised a $130 million Series B in July) and Sakana (an Nvidia-backed scientific discovery organization that raised $214 million in the September round of series A).
The price of mind
But in terms of reasoning ability, the big beast at the moment is OpenAI and its o1 series of models, which take a very different approach to the problem. Instead of moving away from the pure LLM paradigm, o1 models use “chain of thought” reasoning. combined with search, methodically working through a series of steps to arrive at a more weighted answer than OpenAI’s GPT models could previously provide.
The o1 models generally provide more accurate answers than their predecessors, but Chang claims that Reasoner’s results are even more accurate. There aren’t many benchmarks — Reasoner may release its own early next year — but based on the recently released Frames DocBench and Google dataset, Chang said Reasoner achieved more than 90% accuracy, while o1 couldn’t overcome 80%. This result could not be independently verified at the time of publication.
He also said that Reasoner’s approach has significantly reduced costs. OpenAI charges $15 per million tokens (the basic unit of AI data, equivalent to about 1.5 words) of input and $60 per million output tokens, while Reasoner costs 8 cents per million input tokens and just 30 cents per million output tokens cents. “We haven’t yet determined how we want to price that,” Chang said, adding that Reasoner’s “structural cost advantage” would allow users to be charged per result or per verified discovery.
Chang’s claims are certainly big, but Reasoner’s team is small, with about a dozen employees, mostly in the US. So far, the company had just $4.5 million in funding before a seed round last year with investors including the likes of Baseline Ventures founder Steve Anderson, Y Combinator MD Ali Roughani and Operator Collective founder and CEO Malloon Yen. “I’ve been very fortunate with some success in my history, so I wasn’t too worried about funding,” Chang said. But the entrepreneur expects to hire more employees soon as Reasoner grows.
Chung said Reasoner, which received $1.8 million in bookings in the third quarter of this year, will publicly release its tests and demo in the first quarter of 2025, allowing people to upload their own datasets and test the company’s claims. The firm will also release a software development kit to allow others to embed the Reasoner engine into their AI applications and agents. (Chang says the engine is lightweight enough that it can run on even the latest iPhones and Android devices without requiring an Internet connection.)
“We want to make sure that we release it in a way that we start building trust and confidence right away,” Chang said.
