Education Updates at Major Developer Conferences
Original Post on Medium: Education Updates at Major Developer Conferences | by Sam Bobo | Jun, 2024 | Medium
Technology conferences are the center point of excitement for large corporations outside of the Silicon Valley start-up hub. Specifically, developer conferences provide a flood of new announcements of newly available features with audience members with keyboards on computers eagerly awaiting the aperture to open for access to the new capabilities to start exploring and playing on the “playground.” Developers are the heart of innovation and maintain the rhythmic cadence of progress. Its no wonder why open-source communities to developer ecosystems around major platforms flourish — it allows platform companies to innovate and build new tooling capabilities, from SDKs to low-code interfaces, to developers and developers to build extensions and new solutions with those features as the puzzle pieces.
Well, technology conference season is here! Over the past few weeks, Google held its annual I/O event and Microsoft is hosted Build (OpenAI decided to throw in a surprise announcement prior to Google to announce GPT-4o). Now, I’ve long applauded Satya Nadella on transforming Microsoft into an AI focused company and am extremely proud to be a part of a company leading the charge in AI but a particular topic arose from both events that I wanted to highlight this week — Education.
Readers of my blog certainly know that Education Technology is a passion area of mine and I’ve long shared how Generative AI is finally giving rise to AI within the Education sector. I recently published a joint blog post sharing my vision of AI in Education with Nick Potalitsky, which I highly recommend the read.
During I/O and Build, both organizations focused a section on EduTech, but the approaches seem a bit different from one another. Google unveiled LearnLM, a Large Language Model tailored towards Education and natural extending into YouTube, Google Classroom, and Andriod alike while Microsoft unveiled Phi-3, a Small Language Model, that had a pre-trained variant for math tutoring with natural integration into Teams for Education as well as a partnership with Khan Academy. In short, Google is taking an integrated approach while Microsoft is remaining as a platform and partner play.
Google LearnLM
Google unveiled LearnLM, a family of models fine-tuned for learning and built upon Gemini. These fine-tuned models were created alongside Google’s DeepMind division with a corresponding research paper transparently detailing its methodologies, hypotheses, use cases, and outputs. The value that LearnLM provides to learners includes (taken from the announcement page):
Students can simplify Google results into more consumable diction to simplify complex topics, to getting quizzed from YouTube videos, and using optical character recognition (OCR) on Android to answer questions. Furthermore, Google announced the concept of “Gems” that are effectively skills for Gemini that can focus on specific topics. This follows a common pattern seen within the AI industry more broadly taking the form of a platform — OpenAI GPTs, and custom bot-like extensions that have been around since the start of the AI revolution.
What this is lacking is narrative around an agent-like experience that can follow the Council of Expert Leader-Agent model and help breakdown complex tasks. Maybe that will take some engineering effort but within the education space, that seems most appropriate approach to helping with complex topics than span a humanities curriculum. I’ve outlined my minimally viable product proposal for an Education Technology AI solution, from “an EdTech Minimally Viable Product:”
For any Artificial Intelligence solution within Education, there exists a number of prerequisites that underpin the “Minimal” aspect of the MVP definition, including:
In my opinion, Google should leveraging LearnLM as a differentiator point that forms a virtuous cycle of gleaning material from Google Classroom learning management system (safely and trusted of course) combined with Google Research to have differentiated features in Classroom powered by LearnLM as a single place for students and educators.
What inspired me the most was the new solutions Google announced as part of its Education lineup coming via Google Labs: Illuminate and Learn About. Illuminate takes research papers and creates a short podcast-like summarization of the core concepts, changing the modality of interaction and lowering the complexity of understanding by reshaping the vocabulary. Learn About, however, I am most excited about as it models the Disruptive Design that ArcBrowser promised when first introduced to disrupt search:
This conversational search interface reminds me about Endless Academy created by Vik, the founder of Dataquest, a Codecademy-equivilant to data science. Endless Academy takes in input prompt and creates a lesson plan (modifiable) and then tutorial where you can add modular components to extend your learning. Note: I have yet to cover Endless Academy as the solution was launched and abandoned. The concept holds valuable but Google’s magical implementation on such an experience points the former vision into a convergence of dominant design.
Learn About, Endless Academy, and ArcBrowser are all falling into a dominant design defined by the “Fluid Phase” quickly moving to “Static Phase” as I explained in “The Fork of Design Innovation — The Pressures of Conforming to the Status Quo”
领英推荐
This design pattern is starting to emerge more broadly and I find the generative immersive experience to be compelling in a way that its simplistic and delivers information in an intuitive and consumable manner.
Learn About seems less like a classroom assistant and more for intellectually curious minds. I applaud Learn About and have signed up to be a beta tester, however, it goes not fit the current slate of announcements that are covered in this blog. Regardless, I am excited for its potential as it encapsulates Googles organization of the world’s information for curious minds, an educational focus for a search engine!
Microsoft Phi-3
Microsoft took a different approach and focused of building a Small Language Model specifically for math tutoring, as opposed to an LLM with plug-ins. Microsoft’s approach took data from Accelerate Learning and UPchieve to fine-tune an SLM to help generate math problems and provide feedback on existing ones. While Microsoft did not throw immense research behind a dedicated LLM like Google, they did partner with Khan Academy to gather learning data (non-personal) from the platform to ingest a corpus of tutorials, problems, feedback, and more to deepen the SML to be more attune to the problem space.
From the press release:
Today at Microsoft Build, Microsoft and Khan Academy announced a new partnership that aims to bring these time-saving and lesson-enhancing AI tools to millions of educators . By donating access to Azure AI-optimized infrastructure, Microsoft is enabling Khan Academy to offer all K-12 educators in the U.S. free access to the pilot of Khanmigo for Teachers, which will now be powered by Azure OpenAI Service.
This is a highly visible and strategic move by Microsoft to penetrate more of the Education market. First off, most public schools are highly strapped for cash to provide essential needs for students. By offering Khanmigo to educators and students free of change, this will lower the barrier to entry and create an anchoring point in which to evaluate other AI based solutions in Education and tackle a part of the market not associated with Google Classroom. Furthermore, with a brand name like Khan Academy, its trusted by educators and used by students. Finally, Microsoft is able to amortize Azure compute through the use of Khanmigo on Azure. This also enabled Microsoft to remain a platform play and use Khan’s expertise to bolster Phi for Education without native access to one such as Classroom. Its also a manner to pay for high quality training content as the pattern has shown throughout the media industry, but instead of dollars its trading compute, a similar paradigm for OpenAI.
Khan also has a strong value statement here:
teachers can generate custom lesson plans, suggest student groupings, or “level” up or down text passages for learners who are either struggling or need more of a challenge. Used together, Khan Academy estimates these tools can save teachers an average of five working hours each week.
Again, a lot of the features mentioned within Khanmigo follow a common pattern within EduTech of use cases. The true value comes within the brand name, user experience, and overall network of learners. I will comment, however, that a lot of what Khanmigo boasts as feature sets, including a Socratic method of “teaching” (i.e responding to students) follows a RLHF training of the LLM to respond back with a more guided question-like approach. I am very excited about Khanmigo and would love to get testimony from teachers using the tool to truly embrace its impact!
Fast Foward — on June 18th, Microsoft announced additional capabilities within the Microsoft for Education portfolio of products. Specifically Microsoft has made significant improvements for the Education-specific variants of Office 365 suite of tools for a more education based experience including:
Summary
Both Google and Microsoft are making strides in the Education technology sector through a slew of new announcements coming out of their respective developer conferences. Pulling details from the Microsoft AI in Education special report , the common uses of AI for student are in brainstorming and comprehension while educators are worried about cheating and security, meanwhile, cheating is not a high concern that has been proven, rather, remained constant.
The new announcements, however, help continue to set a level playing field by employing common use cases in Education already solidified and understood by the AI community based on the capabilities of Generative AI (as an aside, I enumerated many of these back in early 2023). What I do believe is that whether an integrated play such as Google or a platform play such as Microsoft, we all continue to democratize access to these tools to empower sutdents and educators and raise the floor of education in the world. Furthermore, we should all continue to draw upon progressing data privacy and ingestion practices to help power more capable models that further solve the needs of our educators and students. I continue to remain highly inspired by the EduTech space and look forward to joining it one say (as opposed to simply writing about it).