EXPLORINEER
en

Archive — US — techcrunch

Probably raises $9M to build a more reliable kind of AI

Probably has raised $9M to develop an AI system that minimizes hallucinations and errors, targeting 99.99% accuracy. Their initial product, a data science tool, ensures reliability with citations and audit trails. The system leverages a deterministic validator, allowing operations on smaller AI models, which reduces costs and can extend to other precision-sensitive fields like accounting and healthcare.

Key points

  • Probably raised $9M in seed funding from Andreessen Horowitz.
  • The company aims to prevent hallucinations and factual errors in AI outputs.
  • Their goal is to achieve 99.99% accuracy, akin to deterministic systems.
  • The first product is a data science tool that provides quick answers with citations.
  • The 'data science mech suit' checks LLM's outputs against a deterministic system.
  • Using optimized models allows the tool to run on local hardware, reducing costs.
  • The methodology could be applied to other fields, such as accounting and medical services.

Why it matters

  • The need for accurate AI responses is increasingly crucial as more sectors adopt AI.
  • Rising token costs in AI usage highlight the importance of efficient models.
  • The trend toward accountability in AI outputs resonates with ongoing discussions about AI ethics and transparency.
  • In 2023, efforts to minimize errors in AI systems aligned with investigations into AI misusage and demand for better safety protocols.

Probably's innovative approach to reducing AI errors through rigorous validation could reshape expectations for accuracy in AI applications and offers a viable path for cost-effective solutions in data-intensive fields.