Uncharted AI Time 🗺️🤖

This week we will be breaking down the biggest story in AI, Meta AI, and what it means for social media.

As for the industry deep dive, by popular demand, will dive will be on the intersection of AI + Health Care/Medicine.

Big Story Alert 🚨

Llamas Everywhere

This week, the big story award goes to Meta.

It was a close thing though with Elon’s lawsuit everywhere and his AI startup X.ai in talks to raise 6 billion at an 18 billion dollar valuation. But we already covered Elon in the last edition and the lawsuit likely won’t be resolved until years from now, so we’ll have plenty of time to circle back in future editions. Yay.

So what’s going on with Meta? You may have noticed if you’re one of the 3.24 billion people who use one of their products, that a new feature has popped up in the search bars — Meta AI. It’s been integrated across Instagram, Facebook, WhatsApp, and also has its own dedicated website.

Mark says the goal for Meta AI is to be “the most intelligent AI assistant that people can freely use across the world”. This puts MetaAI directly at odds with ChatGPT, the fastest growing consumer application in the world. But what’s hundreds of millions of monthly active users when faced with the 3.24 billion people that now have Meta AI integrated directly into the apps they use every day?

This is why you never count the Zuck out. Just like the like button, stories, and other social media features he helped pioneer, Zuck has always seemed able to push new features to mass adoption- and now he’s doing the same with AI chatbots. So what does MetaAI actually enable for the average user?

For now, it seems to be a chatbot with internet search capabilities, the ability to make images, and animate images into GIFS, but it’s hard not to see that there’s so much more potential to unlock here. I’m betting soon that Meta AI will allow you to find better accounts to follow, find accounts similar to yours, and unlock a lot of value for content creators in analyzing their account performance. There’s literally so much more to explore here - what does an AI chatbot over your social media online experience mean for how people use their socials? Looks like we will be the ones finding out!

To touch on the technicals, Meta AI runs of llama 3, which is Meta’s most powerful OpenSource model to date. There are two versions of Llama 3 right now with 8 billion and 70 billion parameters respectively. These models significantly outperform earlier versions, and a larger 400 billion parameter model is under development.

That’s all on Meta. Reply with your thoughts on Zuck’s new getup, Meta AI, and how you think it will change social media! I’d love to hear from you all.

HEAR THE FULL STORY HERE

Industry Deep Dive Time 🏊‍♂️

Ok, so now for our first industry deep dive ever! We’re starting off with a big one: AI + Healthcare & Medicine. Healthcare on it’s own is a 4.3 trillion dollar industry and accounts for 20% of the US GDP.

And medicine is also very broad and has many moving parts like biotech, pharma, medical devices, CROS/CMOs, and the wholesale drug distributors that all come together. In this section, I won’t be going through each one of these in detail, but will touch on some and give a broader outlook on how AI will change medicine + healthcare.

Healthcare, as it is now, is a primarily human service driven service. Because of this fact and the low software penetration, AI is posed to seriously bring down costs and raise the efficiency of healthcare systems. Like, wouldn’t it be amazing if AI could help us achieve a future where everyone can get amazing care and medical debt is no longer the number one cause of bankruptcy in the country? Here’s how we’re going to do it and how it’s already being done.

Transforming Diagnosis

Computers models have been used to make diagnoses since the 70s. An expert system trained at Stanford called MYCIN would take manual typed inputs and spit out a list of diagnoses. It beat 5 pathologists when put to the test. However, you don’t see MYCIN in hospitals in charge of diagnosing patients. It never penetrated the healthcare system because there were many legal and ethical issues related to the use of computers in medicine, especially regarding the responsibility of the physicians in case the system gave wrong diagnosis. Furthermore, for its time, the system was incredibly complex to install for your average hospital, but software has come a long way since then.

As computers continue to get better at diagnoses, approaching higher and higher accuracy, there comes a point where it becomes stupid not to use them to help diagnose patients. AI's ability to process data beyond the scope of human capacity enables the integration of multiple data sources for a much better diagnosis. These sources can include not only medical imaging but also genetic information, patient histories, and vast medical literature.

Convolutional Neural Nets allow for better analysis of essentially every type of medical imaging you can think of. Deep learning Artificial Neural Networks (ANNs) are being used to predict genetic diseases. And now LLMs enable better parsing and comprehension of patient histories and note-taking for patient visits (think MedGPT). With all of these technologies working in concert in the backdrop, we can really reimagine what the health care experience looks like. Here’s an excerpt from a16z on what they think a future doctor’s visit will be:

“““Once the patient is taken back to the clinician’s office, an AI physician’s assistant (PA) could listen in to the patient’s complaints as they share with the physician. That AI PA could automatically create a structured note allowing the human provider to focus on the patient during the visit, maintain eye contact and focus, and then review the note afterwards. The AI PA could also capture key documentation essential for coding and reimbursement, reducing the need for coding teams to assess the note and the back-and-forth with providers to optimize reimbursement.

Moreover, the AI PA could assist the physicians and nurses in diagnostics and treatment plan, by offering a thorough, evidence-based differential diagnosis and recommending best options for diagnostic testing that can pressure-test the clinician’s intuition. After the patient is treated, another AI PA could assist in identifying the most appropriate follow-up care for the patient, a task that, without AI, often requires manual and time-intensive phone calls.

For the patient, the AI PA could generate a report in plain English summarizing their diagnosis, treatment plan, and next steps, all using accessible language tailored to the patient’s situation.”””

Pretty cool, huh. And sounds like so much less paperwork too for the care providers at hospitals.

AI in Treatment Planning and Personalized Medicine

Personalized medicine is a method that considers an individual's unique genetic makeup, environment, and lifestyle for treating and preventing diseases. To power this kind of bespoke preventative and predictive medicine and make it accurate, we essentially need AI.

Here are some examples of personalized medicine using AI from academic literature:

In one study, researchers used AI to classify breast cancer patients into three groups based on their genetic information and found specific genes linked to how the disease progresses. This type of classification can help doctors understand which patients are at higher risk and tailor their treatment accordingly. Taking things one step further, AI models were then used to optimize the timing and dosage amounts of medications based on the patient’s risk profile. I can see this process generalizing and becoming standard across practices.

Another research project used AI to explore treatment options for ovarian cancer. By examining genetic data and cancer characteristics, the AI could suggest combinations of drugs that might be most effective for individual patients. From there, it would be up to the care provider to decide which option to proceed with. In this case, it’s interesting to think about whether the human will continue to stay in the loop for decisions like these or if computers will gain autonomy to have their recommendations deployed without further review.

Basically, the point is, AI is a huge unlock for finding more specific treatments to help every individual, rather than operating in the realm of one-size-fits-all.

Drug Discovery and Development

AI has also been able to significantly expedite drug discovery by predicting potential drug candidates and simulating their effects. Machine learning models identify compounds with therapeutic potential and forecast their interactions within the human body and surface those with potential at a much faster rate than anything we previously relied on.

You may have heard of AlphaFold, a model which allows you to predict the 3D structures of proteins, allowing for more precise computational modeling and drug design. Once you have these drug designs, other AI models can go through and rank the different designs by their bioactivity and physicochemical characteristics to narrow in on the drugs will interact with things in a biologically and chemically safe manner.

As things stand, drug development is time-consuming and labor-intensive, predicated on iterative trial-and-error experimentation and high-throughput screening processes. To get a drug to market can take an average of 12 years. With AI, that time could get shaved down by several years. For example, a company called Insilico Medicine, used AI to design and progress a drug candidate from discovery to phase I clinical trials in just 30 months. In the future, drug development can be a lot more agile, taking less time and capital and opening up the market to other, smaller players. Basically, more drugs! Hurrah!

Enhancing Patient Care With AI + Telemedicine

Remote patient monitoring (RPM) has transformed healthcare, especially during the COVID-19 pandemic, by enabling continuous health tracking outside traditional settings. This technology has proven particularly beneficial for managing chronic conditions like diabetes, where it has improved patient outcomes by enabling ongoing monitoring and timely interventions. In fact one of my friends is working on Haema, a diabetes tracker that monitors how your daily activities affect your blood sugar by integrating with glucose monitors. His app is in test flight if you want to try it out.

Widespread usage of AI health wearables are almost a guarantee in the future. Consumers are already adopting devices like the Oura ring for sleep tracking and other health tech wearables, like Whoop and other watches. As for higher risk patients, wearables can run AI models that can predict critical events, such as cardiac arrests, more accurately than ever before, significantly improving emergency care.

I can’t wait until a general purpose wearable comes out that can help me monitor my health across my sleep, physical activity, recovery, and more.

Challenges and Ethical Considerations

Large language models (which may begin to power the note-taking, patient record updates, and various other back office tasks) are prone to hallucination. In a healthcare setting this can be disastrous. It can be as simple as recording the wrong blood type and then when it comes time to get a transfusion everything can go wrong because of a single letter. The reliability of LLMs needs to improve so that they can work side by side with PAs and nurses in a hospital setting, where lives are on the line.

Another challenge is that these models, as they are, can often be a bit of black boxes. Their explainability is not as high as a doctor, who can clearly articulate why they are making one decision versus another. The ability to explain a treatment is very necessary, both for the sake of the patient and for the sake of amassing interpretable data, which will go into furthering our understanding of medicine and biology. This lack of explainability is a problem across the field of ML and as such there is a lot of work going into “glassbox“ AI models which focus on interpretability of results. If these new glassbox models can be used for medicine, perhaps that can help solve this specific problem.

One last major concern is the every-present problem of privacy and data protection. These models will be trained on sensitive patient data and as such need to be rigorously handled and safeguarded, so that the models and data are not misused.

Well that’s it for the deep dive on AI and healthcare and medicine. Let me know how you liked it. Was it too long? Too short? All feedback is welcome.

Thrill of The Shill 🎢

I made something the other week I thought people might like!

I basically took Marques Brownlee’s Youtube channel (he’s a famous youtuber and tech reviewer) and added a shopping assistant over it. I hooked it up to a website askmkbhd.com and voila! It’s like you can talk to him for shopping advice.

So if you want to shopping advice from your favourite influencers, tell me their channels and I’ll make new versions. And let me know if you like it. Thanks!

Parting is Such Sweet Sorrow

Well that's a wrap on the newsletter. Before I launch into my ending monologue, please take a moment to drag this email to your primary, so that you never miss it each week! Thank you!

Let me know how I did by filling out this feedback form or just reply to the email like you would any other. If I did well and you're in the giving spirit please support the newsletter by buying me a coffee or donating to my gumroad page. Most of the money you donate to me will go back into purchasing AI products to try out and review occasionally on the newsletter or help with industry research. Bernie says it better:

Also if you have a product that you would like to be featured on the newsletter remember to add it here!

Lastly, if you'd like to refer a friend just tell them to visit chatgptnewsletter.com to subscribe! Or you could always just forward them the email without any subject header. I'm sure they they would love that.

Cool Things You Should Try/Buy From Every Past Newsletter 🤯

(Newsletter 13) ==> Replace Siri with ChatGPT— HeyGPT!

(Newsletter 14) ==> Autonomous AI Agents in Your Browser— AgentGPT!

(Newsletter 15) ==> Create your own AI music with musicfy

(Newsletter 16) ==> Detect AI written content more reliably with GPTZero’s browser extension.

(Newsletter 17) ==> Another AI model playground, but more than just openAI models

(Newsletter 18) ==> Digitize your mind with You.ai 😵‍💫

(Newsletter 19) ==> Learn how to effectively use AI in your work and life with Maven

(Newsletter 19) ==> Super stylish generative AI website builder Dora. Explora whole new way to make websites.

(Newsletter 20) ==> A TODO list that does itself? Sign me Spellpage!

(Newsletter 21) ==> Making presentations just got easier again with Gamma

(Newsletter 23) ==> Making news concise with AI

(Newsletter 24) ==> Perplexity an AI to replace Googling!