Careers at Onton

About Onton

Learn more about

Onton⁠

and our space.

🔉 If you prefer listening/watching

Future of neurosymbolic AI (Alex + Zach from Onton)

⁠

LLMs won’t lead to AGI

⁠

Google is losing market share

⁠

📚 If you prefer reading

Our tech

We’ve engineered a powerful logic layer, backed by a fine-grained knowledge graph and reasoning engine, that extracts and makes explicit the statistical correlations encoded in LLMs. Our graph engine is 700x faster on natural language queries than RedisGraph, itself a SOTA graph database which outperforms competitors by up to 15,000x on some benchmarks. And the data model that underlies it is flexible enough to represent any piece of knowledge in full fidelity, thus overcoming the “overcrystallization” problem inherent to traditional RDF-like knowledge graph approaches.

Most recently, we’ve begun augmenting our knowledge graph via self-learning from queries and products. Soon, we’ll add user behavior and eventually multimodal data to answer even more questions. Ultimately we’ll have not just a hand-curated knowledge graph specific to product data, but a self-refining, self-correcting, self-augmenting one that encompasses all the background information and methods of reasoning needed to help users make any decision optimally.

Legacy search is getting worse

⁠

Degradation of the Internet⁠

⁠

Whole categories of searches have been so over-SEO’d and over-advertised that searching them on Google and Amazon

has become useless⁠

Today

the average shopper spends⁠

15 hours, across 12 different websites, over the course of 79 days trying to make a purchase decision.

Over the past four years, the amount of time it takes for someone to make a purchase decision online has

increased⁠

by 9 days.

LLMs are

creating exabytes⁠

of unstructured, unverified information daily, making information retrieval more difficult.

Search is fragmenting

We believe there is a very real likelihood that the search industry will fragment over the next several years. As consumers expect ever more useful results from their queries, we expect to see the emergence of specialized engines that index specific areas in a much more comprehensive way than Google is likely to do. - Union Square Ventures

⁠

https://every.to/napkin-math/vibes-based-search⁠

⁠

Millennials, Gen Z, + younger generations

are turning⁠

to new ways of searching

⁠

Disrupting Google Search⁠

⁠

Onton’s market opportunity

Even if we only stayed in home decor and furniture, we’d see a TAM mirroring Wayfair's, which

reported⁠

$12B in revenue last year and is one of the 250 largest companies in the country. We plan to expand across categories and tackle e-commerce search at large which easily presents a TAM of >$500B. Think Amazon 2.0, but without having to deal with warehousing/fulfillment complexities.

The timing is ideal for Onton to exist and grow rapidly. Online shopping was already growing significantly before the pandemic; COVID greatly accelerated that trend. Online retail sales rose 32% YoY in 2020. In 2023,

270 million people⁠

shopped or browsed products online in the US alone.

⁠

Most people start⁠

their e-commerce searches on Amazon, not Google.

LLM issues and challenges

Making sense of the

increasingly⁠

unstructured web is critical to enabling high-quality search. And this sense-making can’t come from LLMs alone. As Meta’s Chief AI Scientist

Yann LeCun said⁠

recently, “Auto-regressive LLMs [e.g. GPT] cannot be made factual, non-toxic, etc. ... It’s not fixable.”

Ilya Sutskever, co-founder of AI labs Safe Superintelligence (SSI) and OpenAI, told Reuters recently that results from scaling up pre-training — the phase of training an AI model that use s a vast amount of unlabeled data to understand language patterns and structures — have plateaued.

⁠

Sam Altman also discussed⁠

the limitations of large language models and that breakthroughs would be needed. What’s necessary is a solution that harnesses the power of models like GPT while taming their pitfalls.

“

Perplexity is a Bullshit machine⁠

”

⁠

“Getting from Generative AI to Trustworthy AI”⁠

(Doug Lenat and Gary Marcus)

“

Do large language models need sensory grounding?⁠

” (Yann LeCun)

“

Training on Generated Data Makes Models Forget⁠

”