Wednesday, November 22, 2023

AI Corner: Trying Out Anthropic "CLAUDE"

Since February, I've been a subscription user of Chat GPT.  With their recent management turmoil, I tried a similar competitor service, ANTHROPIC CLAUDE, in its free version.[*]   

I found CLAUDE did a good job of summarizing my recent detailed white paper on FDA proposal for LDTs (October white paper here.)  

In both ChatGPT and CLAUDE, you can now just click a "paperclip" button and upload a PDF.


Here's the CLAUDE link (once you're registered):

Here's home for Anthropic, where I email-registered for CLAUDE:

Free Claude gives you 5-10 questions per 8 hours, I think.  $20/mo Claude gives you much higher bandwidth.

Here's the test:

 Please summarize in detail this complicated white paper about a new policy proposed by FDA for regulating laboratory tests [upload it]. I am the author of the white paper, and I would like to see how you summarize it. I expect a detailed summary to be in the area of 500 words long. Thank you.


Here is a 475-word detailed summary of the white paper about the FDA's proposed policy for regulating laboratory tests:

The white paper analyzes a new proposal released on September 29th, 2023 by the FDA to regulate laboratory-developed tests (LDTs). The proposal would phase out the FDA's enforcement discretion and subject LDTs, including those solely used within a single lab, to FDA oversight as medical devices. This would require LDTs to comply with requirements like registration, quality systems, premarket review pathways, and labeling rules.

The proposal lays out a 5-year timeline for phasing in compliance. However, the white paper author, Dr. Bruce Quinn, argues that the dramatic costs of compliance, estimated by FDA to be around $50 billion over the first 5 years, make the timeline wholly unrealistic and the proposal “simply impossible to execute.” The FDA also projects over $300 billion in benefits over 20 years, but the white paper contends these are based on speculative assumptions about LDT accuracy improvements.

The paper critiques how the high costs were obscured in the publicly released documents and summary. It analyzes the underlying FDA assumptions and calculations, including estimating over 200,000 regulatory expert-years and thousands of FDA reviewers would be needed in Years 3 and 4 alone to handle the influx of LDT applications. It compares this to the total yearly profits of less than $4 billion for LabCorp and Quest combined.

The white paper further examines the FDA's claimed legal authority to regulate LDTs as devices, noting this remains a disputed view. It contrasts the FDA arguments with analyses suggesting labs conducting testing may fall outside device definitions and commerce clause restrictions on federal power over intrastate activities.

In conclusion, the paper argues the proposed rule is essentially unworkable as written. It expects court challenges if finalized unchanged, as well as continued debate over the appropriateness and legality of FDA reaching into lab operations. But it suggests there may be alternative approaches, short of the complex VALID Act legislation, that could address FDA’s concerns on LDT claims and LDT accuracy without imposing such dramatic bureaucratic and financial burdens.

[*] Both ChatGPT and Anthropic Claude have good free versions, while a $20 subscription to either one gets you more bandwidth and priority when servers are overworked.