Dr. Zsolnai demonstrates GPT-4 diagnosing a dog's tick-borne disease, generating app designs from text, writing poetic prose with words like "caustic," and writing code so secure it produced zero invalid programs in 1,000 tests while cutting insecure apps from 20% to 4%.
Questions this video answers
April 12, 2023.
LARRUZ (featuring Dr. Zsolnai).
Out of 1,000 examples, GPT-4 wrote zero invalid programs.
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today we are going to look at four amazing examples of how we are finding new ways to harness the incredible capabilities of OpenAI's GPT4AI and its ChatGPT assistant. These are things that were not possible at all just a couple years before today and are now a reality. 1. A dog got diagnosed with a tick borne disease, got treatment from a vet but unfortunately no symptoms developed. Upon explaining the situation and adding more information about the blood test results for this good boy, GPT4 interpreted the results and made two potential diagnoses for the dog. Then a vet was asked to perform the tests and hold on to your papers fellow scholars, because the first diagnosis was correct. Yes, this good boy has made an almost full Recovery since. Wow. GPT4 and the vet may have saved this dog's life. What a time to be alive. Beyond this being an incredible success story, two things really jump out here. 1. GPT 3.5 the previous version was unable to do this, so the recent improvement to this AI is responsible for this amazing capability. With that, just imagine what GPT6 or 7 will be capable of. If you have any ideas, let me know in the comments below. 2. Of course it should not be thought of as a great doctor, but perhaps a great advisor instead. 2. Everybody knows we have text to image and some of you fellow scholars know even text to video. But get this, here is text to app design. Yes, we just write that we wish to get an idea for an onboarding screen for a dog walking app and bam. We get something quite cool immediately. And if we wish to change it, these designs are editable. And what's more, just imagine how this can be combined together with the amazing midjourney or other text to image AIs to populate your app with beautiful new icons and artworks. 3. GPT4 is a fantastic writer and it can demonstrate its skills when we ask it, something that I would not expect a machine to be able to do. Now little AI write the most beautiful sentence. It has to be rhythmically and phonetically beautiful. Ephemeral whispers of twilight danced gracefully upon glistening waves serenading the eternal stars. My goodness. I think most of us would agree that this sentence is indeed beautiful, but I wonder if this one was a fluke, so I also tried it myself. As the sun set over the horizon, the sky blazed with hues of crimson and gold, painting a masterpiece of fleeting beauty. This is incredible. And the other two are also amazing. Now let's make it a little harder. Now make sure to use the words refraction and caustic. Interestingly here it chose to play with the fact that a caustic can mean bright patterns in nature where an object concentrates light in a beautiful manner, but it can also mean sarcastic and bitter. So how do you create a beautiful sentence out of that? It wrote the the caustic words of the bitter old man were like shards of glass, but through the refraction of his pain a glimmer of hope still shone. Wow, this is a bittersweet sentence that I think takes a considerable understanding of language and even the human condition to write. Wow, this little AI can write four the improvements from the previous GPT 3.5 is huge when it comes to writing computer code. Sometimes it used to write invalid code, but the new version? What can that really be? In this particular test, out of a thousand examples, GPT4 had written zero invalid programs. And what is also pretty amazing is that about 20% of the apps from the previous version were written in an insecure manner, but for the new one it is down to only 4%. Wow. Now note that this is not a peer reviewed experiment. It and thus some caveats apply. I put a link to it in the video description if you would like to have a look. But wait. If it can write secure computer code, can it also check previously existing projects for flaws? You bet your papers it can look. It can have a look at this piece of code and tell us block by block what the potential harms are. Also it makes a final assessment and as to what it exactly does as a whole and why it has big problems. Now this is a case where we can use this AI as a shield. Loving it. And I have been thinking about that and found that this is an incredible value proposition. Imagine if when reading or hearing anything from the news to reports, we could also use it as a shield to fact check everything we hear or read, search for sources, evaluate the sources and mark falsehoods with red in any article or comment. And the incredible thing is that I would not be surprised if that is already possible. GPT4 has incredible capabilities and it is all up to our ingenuity to bring out the best from it and help humans and yes, doggies too. What a time to be alive. This episode is brought to you by Anyscale, the company behind Ray, the fastest growing open source framework for scalable AI and scalable Python. Thousands of organizations use ray, including OpenAI, Uber, Amazon, Spotify, Netflix and more. Ray lets developers iterate faster by providing common infrastructure for for scaling data, ingest and pre processing, machine learning training, deep learning, hyper parameter tuning, model serving, and more, all while integrating seamlessly with the rest of the machine learning ecosystem. Anyscale is a fully managed RAY platform that allows teams to bring products to market faster by eliminating the need to manage infrastructure and by enabling new AI capabilities. Ray and Anyscale can do recommendation systems, time series, forecasting, document understanding, image processing, industrial automation, and more. Go to anyscale.com papers and try it out today. Our thanks to Anyscale for helping us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.
