|
| 1 | +AI Innovation Corner |
| 2 | +==================== |
| 3 | + |
| 4 | +These articles are part of the *FIRST* Tech Challenge AI Innovation Corner. |
| 5 | +This is a place where we'll post custom and curated articles relevant to |
| 6 | +*FIRST* Tech Challenge as it relates to AI and its impact on our daily |
| 7 | +lives and the world around us. We would like to thank Google for their |
| 8 | +generous contributions to *FIRST* Tech Challenge to increase access to |
| 9 | +our program in underserved communities and for providing |
| 10 | +sponsorship and occasional technical direction for this content. |
| 11 | + |
| 12 | +Articles are ordered on this page chronologically, with the newest content |
| 13 | +at the top of the page expanded. Just click to expand any other articles |
| 14 | +you'd like to see. |
| 15 | + |
| 16 | +.. dropdown:: Week of 09/30/2024 "AI Competition Manual Assistant" |
| 17 | + :open: |
| 18 | + |
| 19 | + .. _competition_manual_assistant: |
| 20 | + |
| 21 | + **AI Competition Manual Assistant** |
| 22 | + |
| 23 | + In our first article, the Google AI Studio was introduced as a |
| 24 | + tool to interact with Google’s Gemini AI. Gemini is one of several flagship |
| 25 | + Large Language Models (LLM’s) that have been meticulously trained on massive |
| 26 | + amounts of text data to learn the patterns and relationships between units |
| 27 | + of language - these models have actually learned how to recognize text-based |
| 28 | + language, read and understand data, and synthesize what it learned to |
| 29 | + predict and interpret future data. This is the exact process humans make in |
| 30 | + learning and understanding the world around us! In Google AI Studio, users |
| 31 | + can interact with the Gemini AI through “prompts” to perform tasks for them. |
| 32 | + Prompts are instructions or queries given to an AI in order to generate a |
| 33 | + response - the quality of the response is often directly related to the |
| 34 | + quality of the prompt. Through these prompts, Gemini can provide responses |
| 35 | + based on the massive dataset that it has been pre-trained with, or users can |
| 36 | + also provide additional documents, text, or media that the AI has never seen |
| 37 | + before. These multimodal prompts, or prompts that include multiple types of |
| 38 | + content, can be very beneficial in interacting with an AI using content that |
| 39 | + is specific to a niche area like *FIRST* Tech Challenge. Can you think of ways |
| 40 | + to put this ability to good use in *FIRST* Tech Challenge? |
| 41 | + |
| 42 | + In *FIRST* Tech Challenge, one of the first tasks teams have to do is to read |
| 43 | + and understand the *FIRST* Tech Challenge Competition Manual. This can be a |
| 44 | + very painstaking task, and even a skilled reader can miss subtle nuances |
| 45 | + provided by the manual. However, an AI can break down and analyze the manual |
| 46 | + in a matter of seconds, usually preserving the nuance provided in the |
| 47 | + document. Users can then interact with the AI that has analyzed the |
| 48 | + Competition Manual, and prompt the AI to provide insights - these questions |
| 49 | + might involve locating specific information likely found in the Competition |
| 50 | + Manual, summarize important rules or processes, or even involve asking the |
| 51 | + AI to make a best guess. Through a process known as “role playing” the user |
| 52 | + can prompt the AI to take on a role or persona and direct the AI to follow |
| 53 | + specific rules as it interacts with the user in subsequent prompts. The |
| 54 | + remainder of this article is a tutorial on how to set up a “role playing” |
| 55 | + session with the Google Gemini AI through Google AI Studio to analyze and |
| 56 | + answer questions based on the *FIRST* Tech Challenge 2024-2025 Competition |
| 57 | + Manual for the INTO THE DEEP presented by RTX season. While some of the |
| 58 | + nuanced elements (like AI prompting) will be shallowly covered in this |
| 59 | + article, it is something we’ll cover a lot more in future articles. |
| 60 | + |
| 61 | + Creating an AI expert using Google AI Studio is fairly straightforward - the |
| 62 | + hard part is creating the proper prompt, and there we’ve got you covered. |
| 63 | + |
| 64 | + **Step 1** - First, log into `Google AI Studio |
| 65 | + <https://ai.google.dev/aistudio>`_. You can do this by clicking the “Sign in |
| 66 | + to Google AI Studio” button on the front page of the Google AI Studio home |
| 67 | + page. You will need a Google account in order to do this - getting one is |
| 68 | + left as an exercise to the reader. The Google account is used to store your |
| 69 | + Google AI Studio prompt sessions and any content you upload to the model, |
| 70 | + and to track usage of the Gemini APIs. |
| 71 | + |
| 72 | + **Step 2** - Let’s download the *FIRST* Tech Challenge Competition Manual to your |
| 73 | + local computer. You can always find the latest Competition Manual PDF at the |
| 74 | + following link: |
| 75 | + |
| 76 | + * https://ftc-resources.firstinspires.org/file/ftc/game/manual |
| 77 | + |
| 78 | + **Step 3** - In the left navigation pane towards the top of the pane, there |
| 79 | + is a circle with a plus inside it with the text “Create new prompt” next to |
| 80 | + it. Clicking on this button will start a new prompt - though if you’re |
| 81 | + using Google AI Studio for the first time it’s likely a new prompt is |
| 82 | + already open. |
| 83 | + |
| 84 | + Now that we have a new prompt, you can give the prompt a name. This will |
| 85 | + allow the prompt to be saved in your "My Library" so you can come back and |
| 86 | + interact with the prompt later without having to recreate the prompt session |
| 87 | + every time. |
| 88 | + |
| 89 | + In the bottom center of the workspace is a text field where you can enter in |
| 90 | + your prompt (it has a default prompt of “Type something”). BEFORE we enter |
| 91 | + our prompt, we want to add our Competition Manual PDF document. To add the |
| 92 | + document, click the “Plus” icon to the right of the prompt area. This will |
| 93 | + give you several options, choose “Upload to Drive”. You can either click the |
| 94 | + “Browse” button to browse for the PDF of the Competition Manual that you |
| 95 | + downloaded, or you can drag the file into the window. This adds the |
| 96 | + Competition Manual to your prompt, it may take a minute or two to upload the |
| 97 | + PDF so please be patient. |
| 98 | + |
| 99 | + .. figure:: images/new_prompt.* |
| 100 | + :align: center |
| 101 | + :alt: Google AI Studio Screen |
| 102 | + :width: 75% |
| 103 | + |
| 104 | + Creating a prompt in Google AI Studio |
| 105 | + |
| 106 | + **Step 4** - Now that we have our document uploaded, we now want to enter our |
| 107 | + prompt. This prompt directs the AI in how to manage its responses, what |
| 108 | + information to use when developing a response, and sets up the role that the |
| 109 | + AI will attempt to play. Enter the following prompt and press the “Run” |
| 110 | + button: |
| 111 | + |
| 112 | + * *You are a helpful AI assistant providing answers to questions about the |
| 113 | + provided PDF. Do not use any prior knowledge; you have everything you need |
| 114 | + to answer questions in the one PDF provided. Cite all references.* |
| 115 | + |
| 116 | + Once the AI processes the initial prompt, we can then ask questions that the |
| 117 | + AI will use the Competition Manual to answer. Depending on the question, it |
| 118 | + may take the AI between several seconds up to a couple minutes to answer - |
| 119 | + be patient! Here are several questions you can ask (remember to press the |
| 120 | + “Run” button after asking each question): |
| 121 | + |
| 122 | + Example sample questions: |
| 123 | + |
| 124 | + * How many SAMPLES is a ROBOT allowed to CONTROL at a time? |
| 125 | + * What are the different ways to score points? |
| 126 | + * How large can a ROBOT be in its STARTING CONFIGURATION? |
| 127 | + * Which awards are best for advancement? |
| 128 | + * How do I write a strong engineering portfolio? |
| 129 | + |
| 130 | + Some prompts that require a lot of complex understanding or strategy can yield |
| 131 | + results that are not correct, especially if there is information “understood |
| 132 | + but not supplied.” For example, the following prompts provide some correct and |
| 133 | + some incorrect information: |
| 134 | + |
| 135 | + Examples of difficult questions: |
| 136 | + |
| 137 | + * What is the maximum score for an alliance? |
| 138 | + * Can ROBOTS pick up an opposing ALLIANCE'S SAMPLES? |
| 139 | + * How many matches does a team play at an event? |
| 140 | + |
| 141 | + This example was specific to FIRST Tech Challenge, but this process can be |
| 142 | + used for virtually any documents or media. Using AI as an analysis assistant |
| 143 | + can help you summarize news articles, find specific instructions in user |
| 144 | + manuals, review books, and more! Remember that the quality of the responses |
| 145 | + the AI provides is directly related to the quality of the prompt provided - |
| 146 | + even so, the AI isn’t always going to be able to provide correct answers so |
| 147 | + it’s up to you to verify the correctness of all answers provided by an AI. |
| 148 | + |
| 149 | +.. dropdown:: Week of 09/09/2024 "AI Innovation Corner - Google AI Studio" |
| 150 | + |
| 151 | + .. _googleAIstudio: |
| 152 | + |
| 153 | + **AI Innovation Corner - Google AI Studio** |
| 154 | + |
| 155 | + This first article launched as part of the *Tech Tips of the Week*, but is |
| 156 | + the official first article for the AI Innovation Corner. |
| 157 | + |
| 158 | + This week’s Tech Tip of the Week launches a new initiative in *FIRST* Tech |
| 159 | + Challenge, an AI Innovation Corner. Generative AI has taken the world by |
| 160 | + storm, becoming commonplace now in everything from personal assistants, |
| 161 | + search engines, recipe curation, music innovation, and vehicle maintenance! |
| 162 | + Machine Learning AI has been a part of *FIRST* Tech Challenge in some way for |
| 163 | + the past six years, and we’re now transitioning to help teams learn how to |
| 164 | + use and incorporate Generative AI in their *FIRST* Tech Challenge experience |
| 165 | + (while we’re learning ourselves!). |
| 166 | + |
| 167 | + The first step (or *FIRST* step?) to getting the most out of AI is choosing a |
| 168 | + model. What do I mean by model? Every AI is a neural network that has been |
| 169 | + trained with specific knowledge with the ability to do specific things based |
| 170 | + on that knowledge. Each version of this neural network is stored in a “model”. |
| 171 | + Each different company has different models available for different purposes, |
| 172 | + though most models are variations on their flagship model (Gemini from Google, |
| 173 | + ChatGPT 4-o from OpenAI, Claude from Anthropic, and so on). Each company has |
| 174 | + different web-based and API interfaces for interacting with their models, and |
| 175 | + everyone has their favorite. In *FIRST* Tech Challenge, the standard tool we use |
| 176 | + is `Google AI Studio <https://ai.google.dev/aistudio>`__ to interact with Gemini. |
| 177 | + |
| 178 | + Google AI Studio is free to use, but requires a Google account to access - |
| 179 | + virtually all models require a login or API token of some kind to use. Google |
| 180 | + AI Studio is our favorite for its list of examples (Prompt Gallery) and its |
| 181 | + easy to use interface to save prompt sessions and resume them later. With |
| 182 | + Google AI Studio, you also can select the specific model you want to use, and |
| 183 | + when available you can choose to use preview versions of up and coming models. |
| 184 | + |
| 185 | + |
0 commit comments