Symbolica AI is developing a toolkit to provide an alternative to conventional AI models

As reports about the rising costs of LLM training and the possible unfeasibility of the 1-trillion transistor chip emerge, some are turning to alternatives such as the traditional symbolic approach to AI. Conventional models like Claude, Gemini, or GPT learn by example and from statistical approximation. This is why they require vast amounts of data and the corresponding hardware to process it, because the more novel examples they encounter, the better their performance gets. In contrast, symbolic AI is a known approach based on the core notion that knowledge can be represented using symbols that can be transformed by following a collection of rules.

Ex-Tesla engineer George Morgan founded the startup Symbolica AI to harness the power of symbolic AI to combine it with breakthroughs in deep learning to build accurate models that require lower training time, cost, and data amounts. One of the key benefits of structured AI models is that their outputs are explainable and verifiable. Thus, with Symbolica AI, Morgan plans to mitigate the most common conventional models' pain points: unstructured outputs, hallucinations, and longer time-to-market. To do this, the company will offer a toolkit to build symbolic AI models, with the option to pre-train them to accomplish specific tasks such as code generation and theorem proving. Although Symbolica AI's business model is still not set, and the company is not disclosing any of its (potential) customers, the company has come out of stealth with a $33 million investment led by Khosla Ventures and backed by Abstract Ventures, Buckley Ventures, Day One Ventures, and General Catalyst.