Social media runs on outrage, and few things get people clicking faster than a scary health headline backed by “scientific proof.”

You’ve probably seen these posts on X, LinkedIn, or Facebook. Someone claiming to be an “expert” or whistleblower insists that a common everyday product is secretly making you sick. They drop a few alarming numbers, point to a real scientific study, and rely on a simple trick: hoping you’ll take their word for it because the original paper looks too complicated to read.

Most of us don’t have the time or training to dig through 20 pages of dense scientific jargon. As a result, people accept these wild claims as facts. That’s how single posts end up with millions of views, thousands of shares, and dangerous health trends built on cherry-picked data.

AI often gets a bad rap for making things up or spreading fake news. But when it’s built with strict scientific rules, AI can actually be our best defense against fake science. Here’s how our AI tool, Tessa, exposed two viral posts claiming sunscreen causes skin cancer.

The Viral Claims: “Sunscreen is Poison!”

Recently, a wave of social media posts started targeting sunscreen, claiming it’s a dangerous scam that causes skin cancer and blocks essential Vitamin D. Two specific posts blew up:

1. The “Chemicals in the Bloodstream” Alarm: A post with over a million views cited a 2020 clinical study, claiming sunscreen active ingredients soak into your blood in under 24 hours at levels up to 500 times above FDA safety limits. The post argued that these chemicals act as hormone disruptors that cause skin cancer.
2. The “300% Cancer Surge” Claim: Another viral post cited a massive study of over 470,000 people, claiming it proved sunscreen users had a 292% higher risk of melanoma and a 140% higher risk of basal cell carcinoma.

To an average reader, these posts sound totally legit. They cite famous medical journals and huge studies. But what did the actual scientific papers say?

What Tessa Found: Solid Science, Terrible Interpretation

We ran both papers through Tessa, our AI system built on T.E.S. (Transparency, Explainability, and Significance).

Tessa doesn’t just skim a paper. It looks at hundreds of data points to create a trustworthiness score (TScore) from 0 to 100. Here’s how these papers scored:

• The Blood Absorption Study: Received a TScore of 82 (Green), showing it was a solid, well-run clinical trial.
• The UK Skin Cancer Study: Received a TScore of 72 (Green), showing it was a strong, well-designed observational study.

Both papers were actually good, reliable science. So why were those viral social media posts so dangerously wrong? Because the influencers cherry-picked isolated statistics and made up wild conclusions that the researchers never actually said.

Unpacking the Lies

Using Tessa’s deep-dive analysis, we stripped both studies down to their core facts to uncover the truth.

Case 1: “Absorbed in Blood” Doesn’t Mean “Toxic”

• The Claim: Active ingredients absorb into your blood and act as toxins that give you cancer.
• What the Science Found: The study tested 6 ingredients and found that blood levels did go above the FDA’s 0.5 ng/mL threshold after application.
• The Plain Truth: Just because a chemical enters the bloodstream doesn’t mean it’s harmful. The FDA set that 0.5 threshold as a trigger to do more safety testing, not as a toxic limit. In fact, the study authors and the FDA specifically noted that people should NOT stop using sunscreen, because protecting your skin from UV rays is crucial.

Case 2: The “Sunscreen Paradox”

• The Claim: Using sunscreen dramatically raises your chances of getting skin cancer.
• What the Science Found: Researchers studied how DNA repair genes work together with lifestyle habits like using sunlamps or getting sunburns as a kid.
• The Plain Truth: While the raw numbers showed that frequent sunscreen users had higher cancer rates, the researchers never said sunscreen causes cancer. Instead, they explained what scientists call the “Sunscreen Paradox”:
  • Who wears sunscreen? People with pale skin who burn easily, or people who already had skin cancer in the past, report using sunscreen far more often.
  • A False Sense of Security: Many people put on sunscreen and then stay out in the blazing sun for hours longer than they should, or they don’t put on enough.
  • What the authors recommended: The researchers urged people to reapply sunscreen frequently and seek shade, not dump their sunscreen in the trash.

How Objective AI Gives Power Back to You

When we built Tessa at Siensmetrica, we wanted to give everyday people the power to spot trustworthy science without needing an advanced degree.

Tessa instantly cuts through the confusion by giving every paper a color-coded Score, so you can tell how solid the research is with just a glance.

But we don’t just hand you a number and expect you to take our word for it. If you’re curious (or skeptical) Tessa reveals the exact evidence behind that score. Nothing is hidden, and there are no black-box secrets. If Tessa gives a study a perfect sub-score for its experimental design, it shows you why: pointing directly to the study’s real-world data, clinical trial setup, and measurement methods.

Scientific researchers love these detailed breakdowns because they reveal the true quality of a study and put doubts to rest. We think regular people deserve that exact same clarity, transparency, and peace of mind. We want to give regular people the power to spot trustworthy science, without needing a medical degree.

The Bottom Line

Sensational headlines will always get clicks, but when bad information pretends to be “science,” it can make people sick. Objective AI gives us the power to cut through the noise, spot cherry-picked data, and get the facts in seconds.

As for the online panic? No, sunscreen doesn’t give you cancer. Please wear it, reapply it, and enjoy the sun safely. Your dermatologist will thank you.

See The Tessa Analysis for Yourself

Interested in how Tessa scored these papers and why? Here are the links to the Summary analysis for both (includes links to the published articles):

• Gene-Environment Analyses in a UK Biobank Skin Cancer Cohort Identifies Important SNPs in DNA Repair Genes That May Help Prognosticate Disease Risk.
• Effect of Sunscreen Application on Plasma Concentration of Sunscreen Active Ingredients: A Randomized Clinical Trial.

Since the commercial release of ChatGPT back in 2022, medical AI has lived in an uncomfortable middle ground stuck somewhere between “whatever” and “have to have it”.

There’s been plenty of promising AI “stuff” — strong academic papers, impressive demos, limited, task-specific successes. But very little that clinicians or researchers could genuinely trust or want to use day-to-day.

That may finally be changing.

The growing interest in a medical large language model called Lingshu-7B is a meaningful signal. On Hugging Face, it’s been downloaded more than five times as often as the next most popular medical model, ClinicalBERT. Not because it introduced yet another a “breakthrough” architecture, but because it reflects a deeper shift in how medical AI is being built.

That shift is simple, but profound: Medical AI is moving from siloed pattern matching toward staged knowledge accumulation, much like human clinicians learn medicine.

The Shift From One-Off Training to Cumulative Understanding

Most early medical AI systems followed a familiar path:

1. Train on a scoped dataset.
2. Optimize for a specific task.
3. Publish results.
4. Move on.

Chest X-rays. Skin lesions. Retinal scans. One domain at a time.

This approach works for academic benchmarks, but it fails in real-world medicine, where clinicians integrate imaging, lab data, patient history, evolving research, oh, and uncertainty, all at once.

The Lingshu model takes a different approach.

Instead of treating each medical task as an isolated problem, it’s trained in multiple stages, with each stage building on what came before:

• Foundational alignment that teaches the model what medical data is without degrading general language understanding
• Deep multimodal alignment that connects images, text, and structured data across domains
• Instruction tuning that reflects how clinicians reason, explain, and document
• Rigorous evaluation across many clinical contexts using a unified framework

This mirrors how humans learn medicine. Foundations first. Then specialization. Then experience. Then judgment based on all the previous.

Why This Matters More Than Model Size or Leaderboards

The most important thing about Lingshu isn’t its parameter count or benchmark scores, it’s that developers are actually using it. And that tells us something critical about where healthcare AI is heading.

Clinicians and builders don’t want clever experiments. They want systems that reduce uncertainty, lower evaluation burden, and work reliably across real clinical situations.

The fact that AI developers are finally taking note and responding means that Medical AI is moving out of the experimental phase and into a status where evaluation, thoroughness, and trust matter more than cleverness and novelty.

The Same Shift Is Happening in Biomedical Research

At Siensmetrica, we see the same transition happening on the research and analysis side.

Our Tessa Neural Networking and AI platform is built on a similar principle: knowledge accumulated in stages and evaluated using scientific rigor. Instead of treating each paper or dataset in isolation, Tessa:

• Accumulates knowledge over time
• Applies transparent, repeatable scientific methodologies
• Evaluates new research, especially work that is early, novel, or not yet fully vetted

This matters because biomedical research dissemination and discussion now moves faster than traditional validation pipelines can keep up with. By the time a study completes peer review and replication, it may already be influencing clinical thinking, funding decisions, or patient care. Having a knowledgeable AI rather than a single or general purpose on-hand is a game changer.

Like Lingshu, Tessa isn’t trying to replace experts. it’s designed to support better judgment by bringing accumulated knowledge and clear evaluation to situations where humans are overwhelmed by volume and speed.

AI Is Reaching the Front Lines of Medicine

The bigger story here isn’t about a single model or platform. It’s about a transition.

AI in healthcare is moving beyond academic experiments, demos, and clever apps and into practical use by people who treat patients, interpret evidence, and make real decisions under time pressure.

Practical tools are rarely the flashiest. But they are the ones that learn in stages, show their work, make uncertainty visible, and earn trust through evaluation rather than claims. That is where medical AI is heading, and it’s exactly where it needs to go.

The debate over reforming the National Institutes of Health is gaining urgency, we have some ideas

Writing on STATNews.com, the Good Science Project’s Stuart Buck outlines how administrative waste, irreproducible studies, and a risk-averse peer review system hold back progress. These challenges don’t just affect researchers, they also slow the pace of medical breakthroughs. At Siensmetrica, we believe reform is essential, but so is giving scientists better tools to navigate today’s overwhelming research landscape.

The NIH holds an enormous role in advancing medical science, but as Stuart Buck notes, much of that potential is weighed down by inefficiency and poor incentives. Three challenges stand out:

• Administrative waste: Researchers often spend close to half their time on compliance paperwork instead of science. That is wasted time and money.
• The reproducibility crisis: Studies are too often published without strong validation. Entire fields can be built on findings that don’t hold up.
• Peer review bias: Innovative work is often overlooked because the system favors “safe” projects with predictable outcomes.

These problems don’t just affect researchers, they also slow progress on medical advances that cost lives every day.

Attacking the Issue From the Ground

At Siensmetrica, we see this from the ground level. Scientists struggle to find reliable, relevant studies in a sea of questionable work. Our Tessa platform helps address this by scoring research across multiple dimensions: peer review quality, journal impact, author h-index, study novelty, reproducibility, as well as many more. This gives researchers a way to filter massive amounts of literature quickly and focus on the studies that are truly trustworthy.

In practice, this matters a great deal. When the volume of low-quality research grows, the best studies get buried. Researchers spend more time sifting through noise, which drains time and funding. Both are limited, and both are critical for advancing science.

Reforms like those Buck suggests—reducing bureaucracy, funding replication, and rethinking peer review—are a strong step toward fixing NIH’s structural problems. But tools outside government have a role too. Platforms like Tessa can help the research community cut through the clutter, reduce wasted effort, and accelerate discovery.

The future of science depends on both institutional reform and better tools for researchers. NIH reforms may clear some of the structural roadblocks. In the meantime, we can make progress by equipping scientists with better ways to navigate the research landscape.

Artificial Intelligence has tremendous potential, but there’s a long way to go between what might be possible one day and what is real today.

As the founder of a tech startup, I spend a lot of time talking about my company and the problem it’s trying to solve. Naturally, because artificial intelligence is such a hot topic these days, people assume that it’s a big part of the “magic” in our platform’s technology.

The truth is though, AI only plays a minor role. It’s not that it isn’t a useful technology, it’s just that, despite what the hype is telling you, it’s not a good solution for every type of problem. Actually, it’s not a good solution for most problems, particularly the complex ones in medical and scientific research we’re trying to solve.

Even for those problems that AI is good at, it still has a long way to go before the real value of AI matches the hype. Here are the key weak spots we see in AI today:

AI promises are overblown

Despite the high-profile entrepreneurs trying to create a wave of excitement and raise money to pursue their ventures, most of the promises and predictions are really based on hopeful thinking rather than reality.

Yes, it would be nice if AIs were like Tony Stark/Iron Man’s J.A.R.V.I.S. – conversational, fast, and 100% accurate. But the reality is, today’s AIs, even after extensive training, might get to 90% accuracy… most of the time. With additional, highly focused training, it might get to 95%. But that last 5% will probably never happen.

Admittedly, 95% is pretty good if you’re working on a 5th grader’s book report, or trying to learn how many bluefin tuna are caught off the coast of Mexico each year. But when it comes to subjects where “good enough” isn’t – and there are all kinds, ranging from aircraft design to life-saving biopharma research – it’s nothing you’d want to trust.

AI is power hungry and expensive

Finally, AI is incredibly expensive. Running an AI requires immense computational resources which, in turn, requires immense amounts of electricity to power, heat and cool the hardware. Even big tech companies like OpenAI, Microsoft, and Meta publicly admit those costs are so high that their AI initiatives struggle to be profitable. Additionally, the more complex the AI, the more expensive it is to run.

Typically, we would expect the costs of technology to decline over time. However, the real costs of AI aren’t in the software or hardware, they are in the power and bandwidth it takes to run it. Demand for AI is so high that global data center capacity, which will double over the next five years, will still lag demand. Moreover, electricity consumption is expected to increase almost 25% a year through the end of the decade. That double-whammy is expected to keep prices sky high through at least the end of the decade.

AI isn’t good at most tasks

Beyond the hype problem and expense, there are simply many kinds of real-world problems that AI just can’t solve.

A lot of the appeal in AI is the idea that it’s some sort of super smart Star Trek computer that can answer questions and solve problems better, faster and cheaper than a human. However, AI isn’t actually “intelligent”, it’s just a really big statistical analysis system. It’s very good at giving you the most likely answer based on a statistical match with the questions you ask it, but it’s completely useless when it comes to “understanding” the question.

For example, if you show a person a “no trespassing” hanging upside down and ask them “can you fix that?” the person will have no problem recognizing that the sign is upside down and turn it right side up. An AI on the other hand, unless it has been specifically trained on that upside down imagery, and that for it to be proper it must be turned 180 degrees, will fail every time.

In other words, AIs can’t reason or do novel problem solving the way people can, so their use is limited to repeating that which has already been done. Again, this is fine for the 5th grade book report on Moby Dick, but when it comes to finding novel solutions or answering questions that haven’t already been answered and documented, AI is neither better nor cheaper than a human being.

Where does AI fit?

Having listed a number of ways AI isn’t the end-all-be-all technology, I should note where AI does have its uses.

As I mentioned above, AI can’t create anything that hasn’t been created before, but it is very good at (re)processing things that already exist to present them in a new way. For example, AI is very good at re-stating and summarizing longform written content. It’s also pretty good at identifying similarities and relationships among different content different pieces. And if you want to know what the pope would look like in a puffy Balenciaga jacket, AI is the way to go.

At Siensmetrica, we’re using AI to quickly deliver summaries of research papers, identify the points of consensus among distributed findings, and even help filter results among well-documented metrics like the h-index.

But when it comes down to diving deep into materials to identify novel or difficult-to-articulate measures of what makes the content of one thing more likely to be reliable than another, we’re relying on a whole series of inputs, multi-vector data analysis, and fuzzy weighting factors like trust, explainability and significance to do it.

Maybe one day AI will have the capabilities and the cost-benefit ratio that make it a viable all-in-one solution for the technology we’re developing. But until then, we’re not going all in on it.

Open access and preprint articles are overwhelming the industry with research; we need a more trustworthy, explainable and significant way to separate the high quality data from low quality data

I have been in the healthcare field for a couple of decades now; long enough to see a lot of changes. Faxes replaced by email, paper replaced by electronic healthcare records, office consults replaced by smartphone apps. But nowhere has there been as much change as in medical and scientific research publishing.

As recently as 2009 research publishing was mostly the purview of a relatively small group of academic publishers controlling the majority of journals. As gatekeepers of information, the lifeblood of science, these journals were able to generate substantial revenues through both fees to subscribe to more prestigious publications, and fees to publish (known as author publishing charges or APCs) in lower impact partner publications.

The Rise of Open Access

Things began to change in 2004. The U.S. National Institutes of Health (NIH) Public Access Policy mandated open access to research they funded, and Europe’s Plan S soon followed suit. Digital publishing and Internet access made widespread electronic distribution possible; open access and preprint publishing platforms soon followed, opening the research access floodgates.

In The Changing Medical Publishing Industry: Economics, Expansion, and Equity, a paper published by a group of physicians in the Journal of General Internal Medicine (JGIM) in July of 2023, the authors note the number of open access journals over the past 20 years has risen from near zero to over 16,000.

The authors note that we are only beginning to understand the opportunities, and consequences, today.

…open access will result in wider dissemination, and preprint platforms with faster communication of science… However, consumers of research may have more difficulty assessing the validity of conclusions from the increased number of research reports… Rapid expansion of published research brings with it the risks of prioritizing quantity over quality of medical research. This poses risks to patients if clinical decisions are influenced by shoddy science.

Identifying Quality Among Quantity

Of course, quantity over quality has always been an issue in scientific research, which is why there are checks like conflict disclosures, statements of limitations, the h-index, and peer review. But even with these, sheer volume and velocity of newly published research, and the opaque nature of peer review makes the task of identifying valuable knowledge in a rising sea of information extremely difficult. The JGIM paper’s authors note that current methods are better than nothing, but still a far from a complete solution:

“…peer review remains an important part of the scientific method. We acknowledge that the empiric benefits of peer review remain debatable, and it may not filter high-quality from low-quality research as well as we like to believe. However, we believe it is better than an alternative system in which there is no formal process for peer input and constructive critique.”

The writers go on to suggest that newer open models including engagement via comments on preprint platforms and community peer review via systems like PubPeer have potential. However, my own experience with healthcare providers, researchers and other consumers of this information, indicates that inputs like these will probably fall short. They’re labor-intensive and, more importantly, as we’ve seen in other open discussion platforms, subject to manipulation by bad actors and robots. As a result, for the Open Access research model to succeed, new models for vetting and valuing research will also be required.

Developing New Metrics for Research Quality

One of the biggest problems with open access and preprint literature is also one of the oldest: “information overload.” Researchers at Oxford University define this as:

…the situation that arises when there is so much relevant and potentially useful information available that it becomes a hindrance rather than a help.

They note that this overload creates a condition creatively called data smog, which is a cloudy, unhealthy, perception of what information is actually available and whether it is of value. Anyone who has ever done a web search on a topic that seems fairly specific and straightforward only to be presented with a million results knows the data smog effect and the skepticism in any results it leads to.

Commercial search engines like Google have developed sophisticated models that do a pretty good job at identifying and presenting relevant information in search. Unfortunately, people have developed techniques in “search engine optimization” (SEO), to game search results in such a way that unless you are a sophisticated search engine user familiar with the commands and parameters for creating highly specific search queries, you are still likely to be showered with mostly irrelevant information.

The question is then, if a near-trillion-dollar search engine giant can’t create a reliable method for identifying and presenting high quality, trustworthy information, how is a frontier community like open access academic publishing supposed to?

Trust, Explainability, and Significance

Fortunately, we have examples of how this might be accomplished from applications in other scientific fields. Researchers are leveraging advances in data science, neural networking, machine learning, and to a lesser degree, artificial intelligence to successfully identify solutions for extremely complex problems. Rather than relying on Boolean parameters like AND, IF and NOT, they are using multi-modal analysis of vast quantities of information to develop models that identify and present the most promising data. Computer scientists working in these areas often refer to three very human terms—Trust, Explainability, and Significance (TES)—as the underpinnings of how they assist human beings in identifying high quality information in an otherwise limitless ocean of data.

TES is already being used to explore new avenues of research and help make incredible advances in a range of highly complex fields from molecular chemistry to financial market behavior analysis. Significantly, these fields have an open access history that is more agile and mature, but no less complex, than that of medical research publishing. As a result, leveraging TES-based technologies for both the discovery of high-quality information, and assisting human beings in confident decision-making based on it, offers considerable promise in healthcare research and clinical decision-making.

Traditional methods for establishing the validity and value of research will still have value in an open access, open distribution publishing model. But, without some means of separating the high quality from the low in an industry that is doubling the amount of available information every three months, information overload will grow, and confidence will fall. Incorporating Trust, Explainability, and Significance concepts into medical research exploration and search technologies offers a way to confidently advance and accelerate medical knowledge and improve healthcare delivery and outcomes.