BlockBeats News, June 16th, AI startup Probably has completed a $9 million seed round of financing, led by a16z. The company is working on building a more rigorous approach to reduce "hallucination" and factual errors in Large Language Models (LLMs).

The company's first product is a data science tool that can quickly generate analysis results from complex datasets. Each output is accompanied by references and a full audit trail to enhance transparency and verifiability. This design is becoming the standard direction for an increasing number of AI products.

To prevent errors from seeping into the results, Probably has built a sophisticated "guardrail system," which founder Peter Elias describes as a "data science exoskeleton." This system uses a deterministic validator to check the initial output of the large model. If the result is inconsistent with the data, it is directly rejected and regenerated. Additionally, this validation mechanism also retrains the model in reverse to reduce future errors.

Click on the original article link below to join the BlockBeats · Feishu AI News Channel, which provides 24/7 monitoring of global AI trends and news.