Republic Venture Fellow & Associate Investment Thesis

Taylor Silveira | Fall 2022 Cohort

Investment Thesis

The growing potential of generative AI in edtech and education has been recognized as a catalyst for investment. The onset of GPT-3 writing tools such as ChatGPT from OpenAI, a language generation model that can generate human-like text, has garnered attention and investment in the edtech industry due to its potential application in writing instruction. This technology can revolutionize writing instruction by giving students more autonomy over their education and providing teachers with tools to scaffold learning and assess performance. Leveraging the power of generative AI, schools can move beyond traditional instruction and create personalized, adaptive learning experiences to help students become confident and competent writers.

What is Generative Artificial Intelligence (AI)?

Generative AI models are trained on a dataset and use probabilistic models to generate new, previously unseen data that is statistically similar to the training data. It has two main components: a generative model and a discriminative model. The generative model produces data, while the discriminative model evaluates that data and determines whether it's valid or not.

What is Education Technology (EdTech)?

EdTech refers to the use of technology to support and enhance teaching and learning. It’s about using computers, tablets, and other digital devices to access a wide range of educational tools and resources, such as online learning platforms, educational software, and mobile apps. The goal is to improve learning outcomes and make education more accessible and efficient by leveraging the power of technology.

Background & Trends

Venture capital investment in artificial intelligence, particularly in companies working with generative AI, has seen significant growth in recent years. Data from PitchBook shows a 425% increase in investment in generative AI from 2020 to 2022, with $2.1B invested in 2022 alone. This trend suggests that investors are confident in the potential of generative AI and are willing to support its continued development. In 2022 alone, investors invested a total of at least $1.37B in venture-capital-backed companies through 78 deals. This is almost as much as was invested in the previous five years combined, according to PitchBook.

Generative AI companies, which use artificial intelligence to automate tasks such as creating content and sales emails, are attracting significant investment from seed investors due to their potential for high valuations in later funding rounds. Some recent notable investments in the sector include a $1.5B valuation for copywriting company Jasper, a $1B valuation for Stability AI, Descript, an audio and editing company valued at $575M in a deal led by the OpenAI startup fund, and a $10M Series A for Regie.ai, a startup that automates blog posts and emails, according to PitchBook.

Investment in GPT-3 and other AI technologies in edtech has been increasing in recent years. According to a report from Grand View Research, the global education technology market was valued at $106.46B in 2021 and is expected to grow at a compound annual rate of $16.5% from 2022 to 2030. This growth is driven in part by increasing investment in AI technologies such as GPT-3. Many of these AI technologies in edtech are still in the early stages, and it remains to be seen how they will be adopted and integrated into education. However, it is likely that investments in these technologies will continue to increase as they demonstrate their ability to improve education outcomes.

Generative AI Writing Tools Market Map

I’m incredibly optimistic about investors championing generative AI writing tools that make us re-conceptualize the discipline of writing itself and how it’s taught in educational settings. The recent public launch of ChatGPT to the world has only strengthened my conviction that this technology will have a transformational impact on education. Particularly, in the area of writing instruction, teachers can incorporate it into their curriculum to provide students with intelligent writing and editing tools that take into account a student’s individualized writing style and preferences.

How Some of These GPT-3 Powered Tools Can Be Used for Writing Instruction

• Scraft: a writing assistant that uses GPT-3 to suggest alternative phrases and sentences as you write. It can be used in writing instruction to help students brainstorm ideas, improve their word choice, or learn about different writing styles and techniques.

• QuillBot: a paraphrasing tool that uses GPT-3 to write sentences more clearly and concisely. It can be used in writing instruction to help students revise their writing on every front, or to learn more about different ways to express an idea.

• Descript: a tool that allows users to transcribe, edit, and manipulate audio and video using natural language. It can be used in writing instruction to help students transcribe and analyze spoken language, or to practice summarizing and synthesizing information from audio or video sources.

• Copysmith: an editor that can be used on students’ drafts that will provide suggestions for improving grammar, style, and clarity. Its features for feedback and collaboration can be used, such as the ability to leave comments and suggestions for revision. It offers analytics tools to track students’ progress and identify weak points.

Exploring Generative AI’s Role in the Quest for Education Reform

I recall the profound impression the 2015 Sundance Film documentary, “Most Likely to Succeed,” had on me when I watched the premiere of it in grade nine in school at High Tech High. It powerfully illuminated the shortcomings of the 21st American education system. The documentary featured interviews with notable figures in the education and tech industries, such as Sal Khan (founder of Khan Academy), Larry Rosenstock (founder of High Tech High Schools), and Laszlo Block (former VP of People Operations at Google), exploring the challenges that have spurred calls for reform and the implementation of new models of learning in education that better prepare and equip students to thrive in the modern economy.

The documentary showed the innovative approach to education at High Tech High in San Diego, where teachers and students engaged with pedagogy centered on a model known as project-based learning, one of the many alternative modes of education out there. Equipping people with the capacity to comprehend knowledge that goes beyond merely committing facts and figures to memory, but rather encourages them to synthesize this information to demonstrate their understanding through the creation of something tangible and enduring that illustrates their learning.

The beginning of the documentary establishes the central thesis of this film project, arguing why we must adopt new models of learning in the education system. The most compelling defense for this is how AI automation surpasses humans' capabilities in regurgitating information, knowledge, and hard facts when necessary to make its applicability of value when completing tasks. This is apparent in the story of Ken Jennings, who held the record for 74 consecutive wins on Jeopardy!, but in 2011, he was defeated by IBM’s Watson. This was a victory milestone in the development of AI and the power of modern computing, demonstrating its capacity to seamlessly perform tasks once thought to be in the exclusive domain of humans.

The Release of ChatGPT from OpenAI

Eleven years later, cutting-edge AI technology is on the cusp of powering a new learning revolution and isn’t slowing down anytime soon. As of December 1st, 2022, OpenAI released ChatGPT which surpassed 1 million users in under a week. Scrolling through a feed of non-stop posts with screenshots from people who are stunned by the high-quality output that the ChatGPT product is able to churn out in a matter of seconds when fed a prompt. Its limitations are clear, as the technology is only capable of producing words based on the initially written input but lacks the semantic meaning of those words.

The ChatGPT product is a tool that uses the GPT-3 language model and a layer of reinforcement learning from human feedback (RLHF) to create a chat interface with some degree of memory. ChatGPT’s easy-to-use and free nature have led to a surge of interest in OpenAI’s capabilities and the impact of AI on society. OpenAI’s API allows for the development of products, but the cost of fine-tuning models with RLHF can have a hefty price tag. The tool’s popularity has been a great success for OpenAI and has generated a lot of feedback for the company’s research.

Not long after this exciting development, freshly published pieces started to roll in from individuals and their take on the implications of this tool and future ones (GPT-4) that will spur great attention on the education front. The titles of these viral pieces are revealing and self-explanatory in terms of understanding how AI will impact the edtech and education sectors.

You can check out my curated reading list on Generative AI x EdTech here:

**Taylor’s Generative AI x EdTech Reading List**

Strengths and Limitations When Using Generative AI to Produce Writing

+ Can generate large volumes of unique, creative, and compelling content.

+ Trained on large amounts of data to produce content tailored toward a specific audience.

+ Produce large amounts of text efficiently without the need for manual input.

+ Produce text that is grammatically correct and free of errors.

+ Can replicate the style of a given author or text sample, allowing for creative and interesting writing.

- Lacks knowing the semantic meaning of words, leading to unnatural-sounding writing.

- It can be prone to bias, as it generates text based on data exposure.

- Doesn’t capture the nuances of human language, meaning it may be difficult to create text that is truly unique and compelling.

- Requires a significant amount of training data to produce high-quality text.

- Inability to assess the quality of written content.

The Decline of Standardized Testing in the Age of AI

Standardized tests have long been the primary way to evaluate student understanding. The reliance on these tests for measuring academic performance will decrease because AI systems will process vast amounts of data quickly, draw conclusions, and execute immaculately by providing answers to test questions more efficiently than human test-takers. Students may start to rely heavily on AI-generated solutions, leading to a need to examine the ethics of integrating this tool in schools. Valid concerns on the integrity front must be addressed when evaluating the impact this product will have on learning to avoid confusion surrounding its fair use.

In one use case of ChatGPT, a Twitter user showed how they used the product to take an entire SAT test. The results were staggering as it scored 1020/1600, which according to College Board is in the ~52% percentile. The one caveat here is that ChatGPT isn’t capable of distilling information from graphs and visual content, which likely contributed to its lower score. A question we must ask here is: What’s the future utility of standardized tests if AI systems can outperform humans in the efficacy domain?

Processing Raw Data and Crafting A Novel Idea or Concept & AI Ethics

While AI can make decisions based on data and algorithms, it can not mimic human cognitive skills in emotional intelligence, abstraction, and creativity that emerges from the interconnection of ideas, experiences, and imagination. These unique abilities enable humans to tackle problems from a different vantage point in ways that AI cannot.

Bloom’s “Taxonomy” emphasizes the process of combining and organizing pieces of information to generate a higher-level concept or idea. It highlights the role of human cognition and creativity in this process, rather than simply suggesting that information is mechanically “spit” out by the individual or system. Benjamin Bloom breaks this down in his work on the “Taxonomy,” in one section, elaborating on the “Six Levels of Cognition,” and the nature of human cognition’s role in the world.

In an increasingly automated world where artificial intelligence can outperform humans on complex and hard trivia questions on game shows and standardized tests given in schools, it is essential that we innovate other types of assessments that allow people to remain competitive and successful participants in the global economy. We must prioritize focusing on the irreplaceable human qualities that can’t easily be replicated by artificial intelligence. The key distinguishing capability humans possess separate from AI is the ability in deriving abstract concepts through the synthesis of diverse information outputs.

A New York Times columnist penned an essay called “What Would Plato Say About GPT?” claiming that Plato believed the future use of advanced technology would cause people to become forgetful, negatively impacting a person’s ability to think critically and retain information. He thought it would instill a false sense of knowledge, as people would rely heavily on technology to give information on a silver platter, rather than plucking from their own understanding and memory. One may use technology as a crutch to appear knowledgeable when in reality, they lack a true understanding of the ideas and concepts being presented to them.

This leads us to the issue of the ethical use of AI products in education and how to teach students the importance of rejecting the notion of succumbing to the simple copy+paste option.

The highly influential Silicon Valley-based venture capitalist Marc Andreessen suggested in a tweet that “ChatGPT plagiarism” should be the least of our concerns:

If a person isn’t striving to learn how to write better than a machine, they should not be writing at all. Writing is a complex skill that not only involves technical proficiency, creativity, critical thinking, and persuading an audience. There’s still a “human watermark” in one’s writing that is distinguishable from what machines churn out, that no matter how advanced, may never be able to be fully replicated. While these tools can be used to tweak one’s writing, the headier process of engaging in deep thinking for the purpose of articulating these ideas into cohesive strings of language is the most difficult and invigorating part.

The legacy models of delivering instruction in education are being disrupted as certain technologies such as GPT-3 prove to have inevitable utility in the education sphere, forcing administrators and educators to address the concerns associated with its use. It’s crucial to introduce complex issues into the discourse to understand the role of GPT-3 technology in education and its broader consequences for learning and the dissemination of information. Shielding these technologies is doing a major disservice to students because if you are preventing them from gaining literacy in a subject that they will most likely need to become more knowledgeable about to be productive and generous participants in an economy that is choosing to adapt and embrace new innovations.

The Future of Homework

To mitigate the negative impact of students using AI tools solely to perform the heavy lifting necessary to complete homework assignments, resulting in learning loss, we must find ways to harness this technology that enriches student learning rather than simply replacing human efforts. By integrating AI into the educational process in a purposeful manner, we can continue to support and foster student growth while also benefiting from the advanced capabilities these tools offer. Ben Thompson suggests in his piece called “AI Homework” in Stratechery that the “Zero Trust Homework” approach could be a viable solution to this problem.

He argues that in this new educational paradigm, students can use AI software to generate answers to their homework assignments, which are recorded for teachers to assess. The skill one picks up is verifying and editing AI’s answers rather than simply regurgitating information. This skill set is not only valuable in an AI-dominated world but also valuable today in combatting misinformation that’s published on the internet. Adopting a “Zero Trust” approach, where all information is treated as suspect and verified, is the most laudable goal in the search for the most truthful answers while maintaining a free society.

Generative AI’s Application in Writing Instruction in Education

There are numerous advantages to using GPT-3 technology as a teaching tool that streamlines tedious labor processes, freeing up students’ and educators’ time to work on more high-level tasks. One possible application is to help students create structured plans for their written assignments. It can serve as an engaging digital mentor, offering assistance that surpasses the capabilities of traditional search engines.

For students:

Overcoming the intimidating blank page is a common challenge for writers, especially students. This tool can be useful for generating ideas and getting the creative juices flowing. It can be used to give feedback on one’s writing and suggest improvements for grammar and syntax, allowing students to see the impact of their choices on clarity and effectiveness. This can be an efficient supplement to traditional methods of writing instruction, such as workshops and conferences with instructors. This could be particularly useful for students who may not have access to individual writing tutors in improving their writing skills.

Incorporating the ability to differentiate between various genres and writing styles can be a crucial element in a writing course. Students can provide the language model with examples of the author’s writing and ask it to generate new text in a similar vein. Students can develop their critical thinking and analytical skills, as they consider the unique attributes that factor into the creation of an author’s style and prose. It will encourage students to think about the nuances of writing style and how it enhances the piece. It will deepen their understanding of writing styles and will gain a greater appreciation for the diversity of techniques and approaches from different authors.

Students can use the tool for collaborative writing projects, for example, they can input their research findings into GPT-3 and it can draft an outline as a springboard for when they tackle the first draft of the paper. This can streamline the writing process by allowing them to focus on editing and refining to get closer to reaching the final product right away. Once students have written their first draft, using the GPT-3 generated outline to guide them in the process, they can use the tool to proofread their draft to help with language, style, phrasing, word choice, syntax, improving the clarity of this messaging and sharpening their arguments to support this. It can be used to analyze a specific source or topic for students to better identify key points and themes that emerge in their research.

For educators:

The tool can be used by educators to help create personalized lesson plans and materials based on the needs and interests of students. Educators can input information about their students’ learning levels, goals, and preferences into GPT-3, and the model can output customized lesson plans that align with these factors. It can save educators valuable time by automating the process of creating lesson plans. When students' individual needs and interests are taken into account, they are more likely to be motivated and engaged in their learning. Educators can ensure that their students are learning about what is meaningful and relevant to their lives.

Educators can assess students’ learning progress and identify additional areas where they may need additional support or reinforcement. By inputting sample responses from students into GPT-3 and asking the model to evaluate the quality of work and suggest areas for improvement. I can help educators to track their student’s progress over time and make adjustments to their teaching methods when necessary. If a student is struggling with a particular subject, the model can generate practice questions that focus on the topic, helping the student to solidify their understanding.

It can be used to create multimodal learning materials by providing GPT-3 with a set of instructions and content and then using its generated text to create a script or interactive elements. Some potential applications for using this tool for multimodal purpose includes; generating writing prompts and challenges for students, writing exercises that can be used as part of a blended learning approach, personalized suggestions and feedback, designing a rubric, formulating directions for activities, writing emails to students, write letters to parents, and other higher-order thinking activities where this AI tool can be made applicable.

Generative AI Writing Tool Investments In Recent Years:

1) Jasper

Description: A content creation service that provides users with AI-generated outputs for their content needs. It offers over 50 content templates that are designed to help users save time and money by creating a high-quality copy with fewer mistakes. It also provides users with SEO-optimized content that is free of plagiarism, so they can generate content quickly and efficiently.

Founded: 2021

Stage: Series A

Amount raised to date: $125 million

Why worth investing: Jasper has democratized access to powerful AI-based tools, making it easy for anyone to create personalized content with their intuitive, workflow-specific templates in 25+ languages. Their customer-centric approach has resulted in over 70,000 paying subscribers in 18 months, ranging from individual creators to large teams. The AI can be leveraged to read/analyze text, identify key points, suggest relevant content, guide essay structure/word choice, and generate long-form content.

2) Sudowrite

Description: An AI assistant designed to help creative writers take their own writing to the next level. It works like a sparring partner, providing feedback and suggestions that help you to improve the quality of your work. With Sudowrite, you can quickly generate plot ideas, create new scenes, and edit your book. It can also provide you with the language and vocabulary to add more life and depth to one’s story.

Founded: 2020

Stage: Seed

Amount raised to date: $3 million

Why worth investing: The founders raised $3 million in the beginning from VCs such as Medium and Twitter founder Ev Williams, Gumroad founder Sahil Lavingia, WordPress founder Matt Mullenweg, and more impressive names on the list. It goes beyond working as a thesaurus because it has a degree of understanding that can pierce the veil of your story’s content. Sudowrite can help with writer’s block and enhance one’s writing to fill in the gaps.

3) Writesonic

Description: Meant for copywriting and paraphrasing through revolutionizing content creation and writing by making artificial intelligence accessible to all. It offers a range of tools to help produce SE-optimized articles, blog posts, ads, landing pages, social media posts, eCommerce descriptions, and much more. Empowering everyone to create any form of content 10x faster.

Founded: 2021

Stage: Seed

Amount raised to date: $2.5 million

Why worth investing: Backed by the Y Combinator and other top investors in Silicon Valley. The model uses deep learning to search the web and match with top-quality content from renowned brands to craft text that sounds like it was written by a human. It also ensures that the copy is engaging, informative, and optimized for search engine algorithms.

Final Thoughts

Generative AI is on the cusp of revolutionizing writing instruction in the education sector by empowering students and equipping teachers with new tools to guide learning and evaluate performance. Leveraging GPT in schools can help to modernize pedagogy for writing instruction to create more personalized and adaptive learning experiences for students to become confident and competent writers.

As AI-generated content becomes more sophisticated and prevalent, it is essential to consider its ethical implications in education and how we can embrace a newfound approach to its use. If the parameters are clearly defined, AI-writing tools can be an incredible asset in the classroom that improves overall educational outcomes.

There’s more practicality in teaching students to become literate in new technologies such as GPT-3, which has the potential to enrich student learning and prepare them to thrive in the competitive modern-day economy. For these reasons, GPT-3 writing tools are a worthy investment in our next generation’s future.