What's New in Chemical Education?

Welcome. My name is Jason Sonnenberg and I'd like to talk to you today about new things in chemical education since the last wtc. My other presenter, Camilo, is not going to be able to join us. He's in grad school at the moment and wasn't able to get out of class today. So I'll be giving both parts of the presentation. So we'll be talking about new features in Wolfram Alpha, WolframAlpha Notebook and the language and move on since the abstract is here for archival purposes. As you might know, we have quite a few tools for chemical education and we'll talk today about wolframalpa, wolframalpa Notebook and some of the other things that are in the language. But we also have a collection of virtual chemistry labs. We have a very large collection of demonstrations related to chemistry that you could use in your classroom to build activities around or to dig deeper into a particular concept. We have a growing Wolfram function repository and Wolfram packet repository for individual and packlets of functions. Jason Biggs just spoke previously about the most recent packlet updates and so I hope you'll check out his talk if you haven't already. And then of course there's the data repository with easily accessible chemical data beyond what's in the system. So our tools are so accessible that students use them quite a lot. And from our own Wolfram summer programs, I'd like to share some of the activities that they did. We had a rather large crowd of chemistry projects this year, so we had an enzyme kinetics project that Sujith was doing working building functions for biochemical enzyme kinetics. We had Vaisha doing protein sequencing based on foldings. Lakshmi did colligative properties functions for binary solutions. We had a mapping reactivity in organic chemistry where Alexander was using the pattern reaction properties that and apply reaction properties that were new in version 13. And then Joshua is doing chemical space networks from molecular data, also using a lot of the graph functionality. So these are just a sampling of projects that can be done by students using our technology to solve chemistry problems. And I hope that if you have a student either from high school all the way up to college or graduate school, that if they are interested in spending a couple weeks during the summer working on an intensive research project, you should check out our summer programs and come do some chemistry with us. The other big development this year, as you probably are well aware of, is ChatGPT and the Wolfram plugin, and there has been quite an enormous amount of literature out on not only what ChatGPT can do, but also what it can do with the Wolfram plugin. And there's a lovely wolf from you. Talk yesterday about all the gory details. I encourage you to check that out to get more details and I'll have time to present here today. But from the chemistry literature, there's been a lot of investigations into how this could be used as a tool in chemical education. And one of my favorite papers thus far is here. ChatGPT needs a chemistry tutor too. So this is from JChem Ed. The links available in the notebook that should be available in the handouts and the links are in the notebook. If you click over to the handout section, you should be able to download the notebook. So in this particular literature article, they were taking 27 questions from a first year final first year final exam for a introductory chemistry course at the collegiate level. And they were feeding it to ChatGPT and comparing how it did. They found that when you're feeding chemistry questions, it didn't matter what the format was, whether they were formatted as free response or multiple choice questions. The scores ranged at the best 37%, at the lowest 15%. And one of the very frustrating situations that occurs is that ChatGPT would potentially be able to select the correct answer, but then when interrogated further as to explain why it selected that particular answer, it would fall flat and not be able to either have a correct answer or just be able to answer, explain its logic. Needless to say, that's not entirely great for a pedagogical situation where you're trying to train students. And so the. The author said that. The author said that ChatGPT needs a chemistry tutor too. As far as the handout, I uploaded it, we'll have to. Let's see, is it. Hopefully it's showing up now. All right, if you're using ChatGPT and turning on the Wolfram plugin, you'll need to have a ChatGPT plus account. And you'll also want to enable the plugin. And I can show that in just a moment. Let's go over. So here's the ChatGPT interface. And if I, in my case, I have access to ChatGPT 3.5 and 4, and I'm using a beta model and so I can go here and go into the plugin store and there's a whole slew of plugins that one can install. I'm going to use the Wolfram one. I already have it installed. And then once it's installed, you can choose to utilize that particular Plugin when you're doing a query tool from to excuse me, to chatgpt. So here, for example, if you haven't seen this live, let's see what does. Look like. And if the bot's moving fast, and it is, it will go and make a call to the Wolfram plugin and give us results. And this is what using ChatGPT with the Wolfram plugin to do chemistry queries. So here what it did is it made a WolframAlpha request saying caffeine structure and here's what it got back. And then it processed that text and the link information to display the 2D structure diagram. And let's see how it's doing now. Let's see. So you can continue your conversation and let's see. It'll make a choice. It'll possibly use the language to do this, or it may call out to WolframAlpha again and it looks like, yep, it asked WolframAlpha rather than using Molecule Plot 3D. But this is just a very simple model of what you can get it to do and switch back to the presentation. After months of working with ChatGPT, I can offer some advice for inputting chemical questions. You're going to possibly want to include these additions to your prompts. It may be helpful to look up the syntax for the chemistry functions that you plan to use before executing code. Don't need to report out what you learned from looking at the syntax. A lot of the chemistry functionality is very new. It's a very active area of development in the language, and so humans and AI bots may not have knowledge that certain new functionality exists. And so to get around that with ChatGPT, you need to tell it to go look things up. Otherwise it will tend to hallucinate and get syntax wrong. But that can be avoided if you remind it that the functionality is there to look up the syntax. If you're doing equilibrium problems and want to see ICE tables, then you want to explicitly say that you want an equilibrium table in the output. And if you're doing chemical conversions, you should remind it that WolframAlpha knows what to do to start with Wolfram Alpha before going to Wolfram language, because the Wolfram language chemical convert function is relatively new and chemical instance function is also relatively new. I won't go through all the examples. There's chemical structure, molecule properties, element composition examples all here. Since this is a coding conference, I'm going to skip down to the code assistant where here as as this prompt says, we're going to build Molecule pattern objects. These are the Wolfram language representation of functional groups, for example, and they're used in pattern reactions and they can be a little tedious to build, but the bot does a really great job. So here in these boxes, just to clarify what you're getting, here in the gray box is the prompt that I used and then the link to the actual saved chat. So even if you don't have chatgpt plus, you can see what's going on. We can swing back over here to take a look at this. So here I told it, I said I want to go look up molecule pattern and bond again because they're new functions. Then write code to create an association of functional groups and then build a corresponding molecule pattern for each. And I remind it not to use Smart Pattern Mises, a very semi cryptic text form of chemical information. And it often screws up when trying to do that. So I tell it not to and it does just what you what we said. It goes and looks up the information about molecule pattern and it builds goes and looks up the information about bond because I didn't tell it to not report back what it read. It gave me a bit of a summary, but then it built beautiful code. So it brought in the knowledge that hydroxyl is an oh. It's an OH group with a bond between the oxygen and hydrogen, along with information for carbonyl, thioamine and phosphate. So one take home message about ChatGPT is that it does have some chemical knowledge. Just don't trust it to do math or to pull up numeric chemical data. It will often screw up with that. And so it can be very handy to have the Wolfram plugin pulled in so that it's not hallucinating about numbers and the chemical data that we might need. So beyond using our tools for student work and AI chatbots, we've also been improving the offerings that we have in Wolfram Alpha. We're building on last year's efforts to improve the thermochemistry for chemical reactions. We've updated and aligned our symbology and the extra information that we give with standard ISO notation and clarified what we think folks will need when solving these sorts of problems or investigating the thermochemistry for a particular reaction here between nitrogen tetroxide and nitrogen dioxide. So that is updated and be rolling out in around Christmas, actually the middle of December. We've also been improving our step by step content. So Camilo had worked very hard on the support for ionic reactions and his latest effort has been the disassociation reaction. Here the example is barium chloride where we show how to go from this chemical formula to the ions and then build the disassociation reaction. So here we've Camilo has coded up the crisscross method that's commonly taught in the United States. Doesn't see as much use outside of the United States, but is a nice graphical way of moving through the problem solving and then it builds up the generic form of the reaction starting with cations and anions and then filling in, excuse me, stoichiometric coefficients and then finally coming back to the the answer. We've also added support for computing atomic mass where we go and pull up the various masses of the corresponding isotopes and their fractional abundance. And here we're actually using around objects under the hood to represent these values and we're doing the around math. So this is one of the first places in the step by step framework where this sort of information is being utilized. And then we get down to the end and we make a comparison to what our data was used compared to the published atomic masses and we explain why there's a slight difference. So with the addition of step by step support for atomic masses that brings step by step support for formula masses for atoms, molecules, ions and isotopes to to completion. This was a big effort and and we will be moving on all obviously to other things. But all of our formula masses can be can now have full chemical step by step support. Additionally, the quantity times last year we had quantity arithmetic, now we have quantity times. So multiplying quantities in chemistry here, grams per mole times, you know, moles. And we show we have step by step support for that too. That'll be showing up. Well it actually is live on the web currently but we'll be showing up inside as further details in existing chemistry step by step. So we're all very excited about about those efforts. We have added to the calculators. There's now a percent yield for doing percent yield calculations for for running reactions in the laboratory. We've also expanded our acid base reactions. We have, we for a while had a PKA calculator. Now we have a PKB calculator and the corresponding Henderson HasselBalch formulation for POH. Those will also be rolling out the PKB and the Henderson Hasselbalch will be rolling out in December. Percent yield is live at the moment. One of the new features that's coming out last year we previewed the the support for uncertainty calculations and this year I'm previewing our Newest feature, we now have support for chemical formula enhanced support for chemical formulas so you can put in ions. Here I did a urine complex with solvated urine complex. And you'll see that we, with the syntax here in bold, we immediately then parse and have support for the usual mass based properties of molar mass. The formula elemental composition and the step by steps are there for that as well. We also can now handle isotopologues where we, we have support for the syntax where the isotope is in parentheses. So in this case we want a tritium, so it's parenthesis 3h close parenthesis. And here I've put in a rather odd version of hydronium where each of the hydrogens is a different isotope. But again, as you can see, it parses. And we have full elemental composition support built right in from efforts in the previous years. As I said last year we previewed the air propagation, but that is all now live. So just to remind everybody, we have chemical quantity support where you can put in the error bars on in this case a mass measurement of S8 and you see that that's carried down through the rest of the pods for the properties our chemical conversions are also aware. So here you can put in the air bars on both the aliquot and the molarity and when you do the conversion to milligrams of solute that is carried all the way through. And we have step by step support for that because we can take, we can parse uncertainties. Now we can do calculations where we're propagating the air in the Wolfram alpha input and here so we have an uncertainty in our molar mass and then we're multiplying by three millimoles and you see how the error is again propagated through. This is really quite powerful and allows educators to take a different approach to significant digits. Historically, significant digits were a model that approximated doing the error propagation that we see here. And now with the tools that we have available students at any level, from middle school on up, if you take that can take the time in your curriculum to explain the error bars in a measuring device and how to enter it into a notebook cell, they can do full air propagation in their work. And here's an example for a titration laboratory experiment where you have collected data so you have the molarity of the base that you're going to use to titrate. And again the syntax is just a little different, but not terribly hard to get down. You set a variable in this case and you need to remember to put parentheses around your uncertainty. So in this case 0.215 plus minus and then the error bar and then close parenthesis followed by the units. And so when you put that in, you can it evaluates rather easily and we get back that value stored. And then you can enter the rest of your data for your aliquots and you can have that automatically converted. And here the conversion is again carrying through the uncertainty. You can have your initial readings, store those away and then have your final readings. And now that data from your experiment is all stored into the notebook here, a Wolfram Alpha notebook edition notebook. And now we can start doing data analysis quite easily. You set the base volume is just the final minus initial and we get the air propagated through. You can compute the base amount times, you know, multiplying the molarity times the volume and again carrying it through. And then you can compute the acid concentration for your three different runs of the experiment carrying the values through. And finally you can average those three values to get your average molarity of your acid that you were titrating. And here we see again now the propagated error has carried all the way through and you can start asking questions about how many significant digits are in your final answer based on the output. And since the first uncertain digit is, the uncertainty is in the third digit. So we have two decimals that are significant third depending on if you want to carry that one around as well. So it's a whole. This really opens up a new potential for chemical education and possibly moving us beyond significant digits to something a little bit more connected to the laboratory experience, which is generally preferable for lifelong skills in chemistry. So with that I'll post some links. So this first link is to the core areas. So broad overview of all of our chemistry tools. For those of you that are wanting to get up to speed on our chemistry tools or want to get your students up to speed, we have a fast introduction textbook and a series of notebooks. So check that out. And then our full chemistry resources, which have links to both of the things above, along with all of our talks from past years and the full list of all of our calculators. Sort of one stop shopping for all things chemistry education and chemistry related in the Wolfram universe. So check that out. And with that I'd be happy to take any questions if there are. See. So the question, there was a question. Since chat GPT is constantly evolving, is there a problem that a prompt that worked yesterday doesn't work today? It's not so much that it doesn't work as so much. I found that you can be, your prompts can become shorter. So in, in past, say back in March or even early June, I would have to be more explicit about what, what giving instructions to the chatbot. But now I'm finding that with the September version of the ChatGPT which is currently available, it does much better. For example, it will automatically format chemical formulas with proper subscripts. So it's doing things like that where we had to, I had to explicitly say, hey, I want things formatted. And I wanted to, let's see, I wanted it to always, you know, show the chemical formulas or stuff. So it's a little bit better about how it's presenting the chemical information which has allowed the prompts to be shorter. But yes, it does change and there's always the probabilistic chance that it won't interpret the prompt that you gave it today as well as it did yesterday. But one of the take home messages from yesterday's talk is that the more detail you can give the chatbot, you know, what we would say, scaffolding of what you want it to do, the better off there. I will give that with one caveat. There's a lovely example, let's see, Of using the bot to help generate a selection of test questions. So if you're doing homework, grading or whatnot in a learning management system and you need 30 versions of a particular homework problem, the chatbot does very well to solve that sort of, sort of problem. But what I found is that you want to do it in a chain of thought sort of interaction. So you'll do one prompt and then here, take a break, then give it the question you want it to make multiple copies of and wait for a response and then ask it for the, the answers to those new problems that it generated. If you do it all in one fell swoop, then it gets a little confused. And so, you know, you sort of guided a little bit to have it only make the permutations of the question that you need. Following the guidelines here, I wanted the significant digits to not change and I wanted different digits so that you can look over at your neighbor's work. But those are just a little structural things of, you know, not sometimes ask. Giving it too much to do at one time isn't necessarily good approach to getting it to do helpful work. Okay, if there are no other questions, I thank you again for your attention and interaction and would encourage you to come to the rest of the chemistry talks. There are Stutech is presenting on Friday, and there are some other presentations from Bob Nakbar and Jan Magdalen yet to come. So check those out and have a great rest of your WTC.