I have been experimenting with artificial intelligence large language models (LLM) for about four years. The idea of artificial intelligence and robotics has been on my mind longer, and I have returned periodically to the topic.
I have noticed that the LLMs can be helpful for brainstorming and organization. They are also reasonably capable of editing and proofreading. But they are clearly not human, not thinking, and not generally and readily reliable.
My latest expirement pit various LLM against one another. I pinged them with the identical "prompt" (query).
"what rights does a worker injured at work in florida have"
ChatGpt (version GPT-5 Mini), a descendant of the early arrivals on the market and readily available for no charge, aptly identified the availability of medical care from workers' compensation, wage replacement, permanent disability benefits, vocational rehabilitation, and death benefits. There were some cautions about notice, the petition process, and more. Chat included a disclaimer, "AI output may not always be accurate. Check important info."
Claude, a product of Anthropic (version Sonnet 4.6), yielded the same as Chat GPT, but added the "right to report without retaliation," some deadlines for reporting, and statute of limitations (the statutory one, not the "two clocks" theory so recently made famous in Florida), and there was discussion of the right to hire an attorney. Claude, likewise, provided no citations for its answers.
Gemini, a Google product, provided detailed medical treatment information, including the "employer choice" of physicians, the entitlements to testing and prescriptions, and even the judicially created medical mileage right. Gemini noted the temporary disability process, the 2026 maximum rate, and the 7-day waiting period. The retaliation preclusion, some critical deadlines, and dispute processes were also noted. The Gemini answer had several hyperlinks to other information (e.g., the Ombudsman Office) and to source information. Some of those were to law firms (one of which I have never heard of before).
Grok is a tool developed by Twitter and owned by SpaceX. Grok led with the duty of timely reporting, overviewed medical with stress on the employer choice, outlined various indemnity categories, and was generally consistent with the quality and quantity of information provided on the other platforms. However, Grok also delved into what businesses must have coverage, attorney fees, and ready contact information for the Employee Assistance Office. It noted that it had consulted "35 sources."
I looked at those sources and found many to be credible, "first-party" sources, including the Florida Legislature, Florida Chief Financial Officer. It also cited some secondary (or tertiary) sources, such as law firm websites.
I asked Grok
"Would your response be different if I asked you to not rely on any commercial websites, specifically no law firms?"
Grok said, "Yes, my response would be somewhat different." I then asked if "the fundamental rights change." It responded with criticalities related to reporting, indemnity, and medical care. The differences included "avoid(ing) phrasing that could imply interpretation from law firm perspectives." It would also avoid "summarized examples" and would emphasize state-provided resources, such as the EAO, brochures, and similar.
Gemini similarly conceded its response "would change." It noted that the "core legal rights ... remain the same," but this directive "changes the tone from a practical summary to a strictly regulatory one." Directed to avoid commercial sites, it would rely on the statute, the Florida Administrative Code, and the Florida Department of Financial Services. This directive would also eliminate "strategy," "advice," or "pro-tips." Gemini noted this would drive it to "no longer point to educational articles written by attorneys." This would "remove() any potential commercial bias or 'sales funnel' language that law firms often use to encourage consultations."
Claude was different. It replied
"No, my response would not be different. The answer I gave you came entirely from my own training knowledge about Florida workers' compensation law — I didn't perform any web searches or consult any law firm websites to generate it."
Claude then added:
"If I had searched the web to answer your question, your request would be a reasonable one to make, and I could limit my sources to official government sources ... (or other non-commercial sources.)".
The lesson is clear. There are differences among the LLM in the marketplace. Some go readily to the internet for information and the risks that entail. Others rely on their training, which itself might as readily have included editorialized, sales, or marketing information in the guise of academic or government production.
What you glean from your inquiry is influenced by how and what you ask. It may depend for depth or breadth on which LLM you engage. It will be influenced by the currency of information and resources chosen by the LLM for formulating response. The simple fact is that both you and the LLM will likely make choices. Choose well, and check both your work and the results.
