Separating Facts from 'AI Nonsense'

Hey, it's Matt. Welcome to AI Street Markets, where I highlight AI investing tools. Adjust your AI Street delivery preferences here.

Was this email forwarded to you?
Subscribe below to get AI Street straight to your inbox.

GUIDANCE

Extracting Reliable Data From Earnings Calls

Right now, ChatGPT is treated like a Swiss Army Knife: it can handle a lot of different tasks well enough but isn’t really tailored for specific jobs.

Subject matter experts are refining the tech for specific use cases.

One example: on earnings calls, generalist large language models sometimes confuse reported values with guidance — a pretty big error.

This isn’t exactly an AI “hallucination,” which fabricates data. The numbers are real, but they’re not answering the actual question.

Kris Bennatti, CEO and co-founder of Hudson Labs, calls these responses “AI nonsense.” This week, I spoke with Bennatti about how they solve this problem.

The former KPMG auditor and data scientist says the company’s fine-tuned approach can properly separate actual earnings from guidance.

In the below video, we go through Nvidia’s recent earnings release, which turned out to be a good example. The chipmaker posted strong results, with revenue up 78%, yet its stock sold off sharply. Why? Its forward guidance came in slightly ahead of expectations but not enough to sustain the market’s high hopes.

Hudson Labs requires users to specify companies and earnings calls before asking questions, ensuring the LLM pulls information only from relevant documents.

In this video, we cover:

• Extracting guidance figures from earnings calls with correct calculations

• Using “enforced sourcing” for more reliable AI results

• Tracking specific topics (e.g., “traffic” at Costco) across multiple earnings calls

• Comparing subtopics across different companies’ disclosures

• Leveraging LLMs to organize data rather than generate content

Hudson Labs, which was co-founded by Suhas Pai, has clients managing over a trillion dollars in assets, including some of the world’s largest hedge funds and pension funds. Its client base also includes insurers, accounting firms, and plaintiff class action firms.

Over the last few weeks, we've seen that specialized AI applications can provide important advantages for specific financial tasks, particularly where accuracy and domain expertise matter.

For readability, here’s what Hudson Labs produced on Nvidia’s guidance:

• NVIDIA expects total revenue for Q1 2026 to be $43 billion, plus or minus 2%.

• GAAP gross margin is expected to be 70.6%, plus or minus 50 basis points.

• Non-GAAP gross margin is expected to be 71%, plus or minus 50 basis points.

• GAAP operating expenses are expected to be approximately $5.2 billion.

• Non-GAAP operating expenses are expected to be approximately $3.6 billion.

• GAAP and non-GAAP other income and expenses are expected to be an income of approximately $400 million.

• GAAP and non-GAAP tax rates are expected to be 17%, plus or minus 1%.

• NVIDIA expects full year fiscal year '26 operating expenses to grow to be in the mid-30s.

BACKGROUND

If you’re new here, check out the last few editions on using AI for investment analysis, creating customized news feeds and tracking earnings call mentions:

• AI Due Diligence: AnalystAI

• Tracking Big Tech's AI Spending

• DeepSeek on Earnings Calls: From Zero to 100+ Mentions

• Demo: AI Agents in Market Analysis

• ChatGPT's Task Scheduling: Automated Market Analysis

• Effective Prompts For Investment Research

None of this is investment advice.