As my readers know, just a few weeks ago, I blogged about how Artificial Intelligence (AI) tools are starting to emerge which can help ordinary consumers/patients navigate the overly-complicated U.S. Healthcare space. For example, I wrote about a clever AI tool called "Fight Health Insurance" which will take commercial healthcare insurance denial letters and enable patients to scan those letters, answer a few questions, and get nearly-instant appeals letters they can snail-mail to their insurance company's appeals department which I blogged about at https://blog.sstrumello.com/2024/09/harness-ai-artificial-intelligence-to.html which are kind of interesting, are driven by actual patients and can potentially save you some time in a laborious aspect of the U.S. healthcare system, which is appealing "adverse decisions" denying care which your doctor has prescribed.
But while AI can compose seemingly well-written text (with occasional errors; the AI tools which exist as of autumn 2024 are certainly not foolproof), for some things which I use my computer for, I am finding many AI tools have certain guard-rails meant to protect companies' advertisers from their own AI tools. Those guard-rails act as impediments to accessing information in order to protect their own advertising dollars.
For example, I am not a paid subscriber to any healthcare publications, however, sometimes I have found that I can retrieve the entire text (or much) of some entire articles simply by going to the html version of the a webpage for an article using my laptop (powered by Microsoft Windows; I have no clue about doing it on Apple Macintosh computers; I've also been unable to do so on mobile devices such as smartphones or tablet computers) by hitting "CTRL" and "U" concurrently and then using the search function to look for some of the article text which appears on the preview. That method helps me find the text in seconds, then I copy the text (including the html formatting), paste it into Notepad, and then I would manually strip-out all of the html formatting code (eliminating any text formatting or links in the article), and voila, on certain publications such as Crain's Modern Healthcare, I have been able to find the text for entire articles. Why not harness AI tools to do that and save myself some time? In theory, it sounds nice, although it doesn't always work out, it depends on exactly how the publications paywalls function.
For example, last week, Crain's Modern Healthcare magazine published an article about the Wall Street response to the CEO shakeup at CVS Health. For an investment of about 5 minutes of my time, I was able to use that method to retrieve the full text of the article, and I did little more than copy the raw html and then manually stripped the html code out, or in this case, I used AI tools to strip-out the html formatting on my behalf.
Wall Street weighs in on CVS Health CEO shakeup
Lauren Berryman, Modern Healthcare by Crain's
October 18, 2024
https://www.modernhealthcare.com/insurance/analysts-cvs-health-karen-lynch-david-joyner-aetna
"It is no secret that our industry faces significant and dynamic challenges, and that CVS Health must make financial and operational improvements to drive elite execution and maintain our position as a leading healthcare company," Joyner wrote in a memorandum to CVS Health employees Friday, according to Bloomberg News. "I need your help," he wrote.
Here's a sample of the immediate reaction to CVS Health's blockbuster announcement:
On CVS Health's future
"Concurrent with today's CEO announcement, CVS Health confirmed that three-quarters of its businesses, led by Caremark and CVS stores, continue to perform at or above expectations, reflective of the hard work, skill and acumen of the people of CVS. Losses in Aetna's Medicare Advantage product, which Glenview estimates for 2024 will approximate $3.4 billion or -5% operating margins, reflect the poor decisions and risk management of a select few. We believe these issues are quite fixable with strong leadership and appropriate board oversight and risk-management." — News release from Glenview Capital Management, which owns an estimated $700 million stake in CVS Health
"At the end of the day this still comes down primarily to stabilizing Aetna and [Health Care Benefits]. ... But we still need to actually get there, and today that is not happening." — Research note from Michael Cherny and colleagues at Leerink Partners
"The most important fundamental remains TBD — the hoped-for recovery in Aetna." — Research note from John Ransom and Parker Snure at Raymond James
"The announcement also appeared to lean into the value of the integrated business model that CVS currently operates. So not only is a breakup announcement not coming in the near future, it may never come." — Research note from Julie Utterback at Morningstar
On Joyner's promotion
"Investors may have been hoping for new blood from outside the organization." — Utterback
"An internal promotion instead of a search process is curious, but we don't see it as a strong signal in the context of the existential questions facing the company." — Research note from Jonathan Palmer at Bloomberg Intelligence
"We believe the company's culture, governance and leadership should be strengthened by those with both appropriate industry experience as well as fresh perspectives. ... All material decisions impacting the future of CVS should be addressed by an enhanced board that incorporates such external perspectives. ... We look forward to engaging with David Joyner in the near future to discuss opportunities to enhance the performance, culture and governance of this iconic institution to drive value for all stakeholders." — Glenview Capital Management
On Lynch's exit
"The decision by Karen Lynch and the board to offer and accept her resignation is one we respect and support. ... We thank Ms. Lynch for her years of service to Aetna and CVS Health along with her candor and integrity in recent conversations and wish her well in her future endeavors." — Glenview Capital Management
"We certainly were not expecting a management change today, although we understand the rationale behind it following another quarter of underperformance. The heat has seemingly been on CEO Karen Lynch given CVS' recent struggles, heavily driven by the [Health Care] Benefits segment (Aetna) that she came from. It is hard, given the operational and stock underperformance, to say a change at the top is undeserved." — Cherny and colleagues
"We are not surprised by the management change given the execution shortfalls at CVS, especially at the Aetna medical insurer that Lynch previously led." — Utterback
"Prior misjudgments left the board little wiggle room to shore up prospects outside of finding new leadership." — Palmer
Nona Tepper contributed to this story.
I have come to realize some of these relatively new AI tools are coded to avoid enabling users from harnessing AI tools to navigate paywall exclusions, but they claim it is about copyright protection because the companies could be sued for copyright issues. But the thing is that there is absolutely NOTHING in U.S. copyright law specifying that users cannot simply ignore html formatting to locate article text hidden by paywalls; rather that is code which the companies [providing the AI tools] are using to protect their own lucrative advertising businesses.
Of note is that Alphabet's Google (whose AI tool is called Google Gemini) relies far more on advertising than does Microsoft (whose AI tool is called Microsoft Copilot) which is a more traditional software company. I had not extensively used alternative AI tools such as OpenAI's ChatGPT (the latter being jointly funded by Elon Musk, and it appears that AI tool is slightly less protective of advertiser interests; it is possible that other AI tools could be less protective of advertisers because they do not really have any). However, the html code reveals some publisher's article pages only list a short excerpt of the full article text anyway; if the publisher does not do so, then they are not really trying very hard to protect their content IMHO.
However, extracting text from an html page might be the type of task which Artificial Intelligence (AI) tools such as Google Gemini, Microsoft Bing Copilot or Open AI's ChatGPT might be well-suited to handle, and yet when I entered my query into those tools as follows, I received a variety of answers. Below was how those AI interaction played out. I typed the following text into the AI input (using several different AI tools).
I typed:
"Using the html version of the article found at https://www.modernhealthcare.com/insurance/analysts-cvs-health-karen-lynch-david-joyner-aetna please remove the html formatting and return the article text"
Google's Gemini responded:
"I don't have the ability to directly access and process information from websites" followed by a copy of the article's headline.
So, I tried again by rewording my request ever-so-slightly as follows:
Using the html version of the page found at https://www.modernhealthcare.com/insurance/analysts-cvs-health-karen-lynch-david-joyner-aetna please remove the html formatting except for the page breaks and return only the text
Gemini's response to that inquiry were slightly more helpful:
I can't access and process external websites due to security restrictions. However, I can help you achieve the same result if you provide the HTML content of the webpage. Here's how to remove HTML formatting except for page breaks:
- Copy the HTML content: Right-click anywhere on the webpage and select "View Page Source" or "Inspect" depending on your browser. Then, copy the entire HTML code.
- Remove formatting: Paste the copied HTML into a text editor like Notepad or a similar program. Text editors typically ignore HTML formatting and display only the plain text content.
- Identify page breaks: Look for HTML tags like <br> or <hr> which represent line breaks and horizontal rules respectively. These can be interpreted as page breaks in the plain text version.
- By following these steps, you'll have the text content of the webpage minus the HTML formatting, with page breaks preserved.
On the other hand I tried the first version of my inquiry with Microsoft Copilot, which did slightly better, responding to my inquiry:
"I'm sorry, but I can't provide the full text of the article as it is copyrighted material. However, I can give you a brief summary of the key points if that would be helpful. Would you like me to do that?"
To Copilot's credit, when I responded: "Yes, please provide a brief summary", Copilot's brief summary was actually the bulk of the Modern Healthcare article text, therefore my assessment was that was fairly decent. Here was exactly what Copilot came back with:
"Sure! Here's a brief summary of the key points from the article":
- CEO Change: CVS Health has replaced its CEO, Karen Lynch, with David Joyner
- Lynch stepped down as CEO of CVS Health after facing numerous challenges during her three-year tenure
- Challenges Faced: Under Lynch's leadership, CVS Health encountered high medical costs, federal enforcement actions against its pharmacy benefit manager subsidiary CVS Caremark, store closures, and layoffs
- Wall Street Reaction: Analysts have mixed sentiments about the leadership change, but haven't significantly altered their long-term outlook on CVS Health
- Financial Performance: CVS Health has repeatedly missed earnings targets, leading to unrest among shareholders and activist interest
- The company also cut its financial expectations for the third time in August and announced plans to cut 2,900 workers
- New CEO's Message: David Joyner reached out to employees, acknowledging the significant challenges the industry faces and seeking their support"
Meanwhile, Open AI's ChatGPT responded slightly differently, with the following:
"I can't access external content like the article you mentioned. However, you can copy the text here, and I'll help you clean it up or summarize it as needed"
Hence, I could use a combination of copying the raw html article text, then using several different AI tools (such as using ChatGPT with a command of: "Please remove the html formatting code from this and leave only the text with relevant paragraph breaks in: <html>ARTICLE TEXT</html>" and entering it.
By using three different AI tools, I was able to generate the complete article text in just minutes, and the AI tools assisted me in doing it more quickly (but not more reliably than doing it manually). Key was to understand which ad-revenue and security quirks each AI tool had in place and use the AI tool best suited to my particular need to maximize my productivity.
Not every publisher's paywall protections for their article content works the same way. That means using the "CTRL" and "U" function will not necessarily even deliver the article text hidden behind a paywall, but some less-sophisticated publishers do that right now. I am trying it to harness the internet to get content I feel should otherwise be available.
As I noted, my feeling is that publishers who do not put their content completely behind paywalls and simply respond with an excerpt of the entire article text are kind of exposing themselves to what I just did by using the html version of their article pages to find their complete article text. That's really on them. Some publishers are much more militant about protecting their content. But AI tools demonstrated a very mixed ability to manage such inquiries. My perspective was that some were clearly better than others.
Personally, I have found going to my public library and checking out a book such as "Writing AI Prompts For Dummies" enables me to make my queries sufficiently detailed to effectively leverage some popular AI tools including Google Gemini, Microsoft Copilot and ChatGPT.
I have mixed feelings about each.
I tend to use Google Gemini most frequently (I type "bard.google.com" rather than "gemini.google.com" because its only four letters; whoever rebranded Google's AI tool made a big mistake with such a long name, but whatever), but as this example shows, I found better luck with Microsoft's Copilot and ChatGPT's large language model AI tools worked better in this instance than did Google's Gemini. But using each to determine which one is best suited can be helpful.
Who knows?
You might find combining their writing capabilities together will enable you to come up with the perfect article. But for more mundane tasks, using several AI tools for different parts of the job worked well.
No comments:
Post a Comment