NotebookLM for Research: Introduction — Transcript

Full transcript

1. 0:00I'm Francesca El Bredy, and this is the
2. 0:03first video in the O R series on
3. 0:06Notebook LM for research.
4. 0:10AI, machine learning, and large language
5. 0:13models are being utilized more and more
6. 0:16within research practices.
7. 0:18When applying these tools within
8. 0:20research contexts, it is important to
9. 0:23consider the ways AI systems can be open
10. 0:26or closed.
11. 0:28You might think of it as a spectrum. As
12. 0:30this article and info vis by Angela Luna
13. 0:34from 2024 demonstrates.
14. 0:37For example, the code for the model
15. 0:40itself could be released open source,
16. 0:43where others can customize or contribute
17. 0:45to it.
18. 0:47Or it could be proprietary to a
19. 0:49particular entity,
20. 0:51meaning the code is only available to
21. 0:53those with access and is protected under
22. 0:56copyright.
23. 0:58Models can be released for public use
24. 1:00and still be closed, such as certain
25. 1:03versions of Chat GPT.
26. 1:06In these cases, users should not assume
27. 1:09that any use of the model is private or
28. 1:12protected in terms of their own
29. 1:13intellectual property.
30. 1:16Your data is being captured and
31. 1:18processed on proprietary servers.
32. 1:22Instead, you may choose to download and
33. 1:24run an open-sourced LLM locally using
34. 1:28something like Ollama,
35. 1:30so that your data is not shared.
36. 1:33The AI system itself may also be
37. 1:36designed to operate under more limited
38. 1:38conditions.
39. 1:40Imagine you just want to ask questions
40. 1:42about a single set of documents,
41. 1:45and you only want answers that can be
42. 1:46found directly in that data.
43. 1:50AI knowledge retrieval systems, often
44. 1:53using techniques like retrieval
45. 1:55augmented generation, rag,
46. 1:58probe a single knowledge base
47. 2:00data set that the user defines
48. 2:03which increases output accuracy and
49. 2:05contextual relevance.
50. 2:08Rather than drawing on many different
51. 2:10sources
52. 2:11for example, a data set that is akin to
53. 2:14the whole of the internet
54. 2:16this RAG approach limits the data scope
55. 2:20and so the system can only respond based
56. 2:22on the information that has been
57. 2:24provided in the data set.
58. 2:27And obviously, more secure systems are
59. 2:29necessary when it comes to using AI/ML
60. 2:33tools for sensitive or protected data.
61. 2:37Which is why RAG is great for research.
62. 2:41Notebook LM is a web-based RAG system
63. 2:44that sticks to the provided sources such
64. 2:47as PDFs, texts, websites, audio, or
65. 2:52video.
66. 2:53Though it can access the outside if
67. 2:56needed or prompted.
68. 2:58It's a notebook in the sense that the
69. 3:00application operates as a user's
70. 3:02personal space
71. 3:04using private data that the user
72. 3:06provides.
73. 3:08LM stands for language model.
74. 3:11The generative AI engine is currently
75. 3:13based on Gemini 3
76. 3:16with Gemini 3.1 Pro available to
77. 3:19Pro/Ultra users for advanced tasks.
78. 3:24And the Nano Banana model handling
79. 3:26specialized multimedia tasks like
80. 3:29cinematic video overviews.
81. 3:32So why use Notebook LM for research?
82. 3:35A common challenge for researchers is
83. 3:38that we often find ourselves drowning in
84. 3:40PDFs, meeting transcripts, and browser
85. 3:43tabs.
86. 3:45Unlike ChatGPT, which is pointed at the
87. 3:48whole of the internet where anyone can
88. 3:51post anything about a topic,
89. 3:53Notebook LM knows your data.
90. 3:56As source-grounded AI,
91. 3:59Notebook LM only answers based on what
92. 4:02you give it,
93. 4:03significantly cutting back on
94. 4:05hallucinations and providing instant
95. 4:08citations.
96. 4:10This makes it a far better tool for
97. 4:12assist in high-level academic tasks,
98. 4:15while maintaining output accuracy and
99. 4:18contextual relevance
100. 4:20as compared to a general LLM.
101. 4:23Notebook LM transforms how you interact
102. 4:25with information by grounding its
103. 4:28intelligence in your specific data.
104. 4:31Under the features umbrella, the
105. 4:34platform allows you to manage an
106. 4:36extensive library of up to 50 individual
107. 4:39sources, including PDFs, Google Slides,
108. 4:43and even YouTube transcripts.
109. 4:46This creates a closed-loop system where
110. 4:49the AI only knows what you tell it. And
111. 4:53because security is baked in, your
112. 4:55uploads stay private and are never used
113. 4:58to train Google's public models.
114. 5:01Let's take a closer look at the
115. 5:03interface layout to demonstrate how you
116. 5:06navigate between the source panel, the
117. 5:08chat, and the studio.
118. 5:11The Notebook LM interface is structured
119. 5:14as a three-column workflow that moves
120. 5:17from data ingestion to synthesis.
121. 5:20You input your raw data on the left,
122. 5:23interact with and refine that data
123. 5:26through the chat in the middle,
124. 5:28and finally, produce a polished output
125. 5:31or creative asset using the studio on
126. 5:34the right.
127. 5:36Let's start with sources, where you
128. 5:38input your data.
129. 5:40Located on the left side of the screen,
130. 5:43this panel is the foundation of your
131. 5:45notebook.
132. 5:47You can upload PDFs, websites, text,
133. 5:51videos, or audio files.
134. 5:54There is also a direct integration for
135. 5:56importing files from Google Drive.
136. 6:00In addition, it features a search bar
137. 6:02for the web and specialized options like
138. 6:06deep research for in-depth reporting and
139. 6:09fast research for quick queries.
140. 6:13These allow you to pull from the
141. 6:14internet to find additional sources.
142. 6:18Once you've added your sources, this
143. 6:20area serves as the library for all the
144. 6:23grounding materials the AI will use to
145. 6:26answer your questions.
146. 6:28It won't go beyond what is added here
147. 6:31for answering any questions you have or
148. 6:34prompts you make.
149. 6:36Next is chat, which is the area where
150. 6:39you interact with your data or sources.
151. 6:43This central panel is the primary
152. 6:45workspace where you engage with your
153. 6:47uploaded data.
154. 6:49At the bottom, there is a text entry
155. 6:51field to start the dialogue once sources
156. 6:54are added.
157. 6:56Through the conversational or chat
158. 6:58interface, you ask questions about your
159. 7:01data or prompt it to summarize specific
160. 7:04documents or find connections between
161. 7:07different sources.
162. 7:09Because the AI is grounded in your
163. 7:11sources, its responses are limited to
164. 7:14the information you provided in the left
165. 7:16panel, reducing hallucinations.
166. 7:20Finally, studio, the output, is located
167. 7:24in the right-hand panel.
168. 7:26The studio contains features dedicated
169. 7:29to transforming your sources or research
170. 7:31into different types of structured final
171. 7:34products through several preset
172. 7:36generation tools.
173. 7:38For example, there are those geared
174. 7:41towards study and analysis, such as
175. 7:43tools to generate data tables, mind
176. 7:46maps, flashcards, quizzes, and reports.
177. 7:51Others are focused on generating visuals
178. 7:54based on your data,
179. 7:55such as the infographics and slide deck
180. 7:58tools.
181. 7:59There are some focused on media creation
182. 8:02with options for audio overviews, often
183. 8:05used for the popular podcast style
184. 8:07summaries, and video overviews.
185. 8:11Also, I want to point out the add note
186. 8:14button at the bottom, which allows you
187. 8:16to save specific snippets or thoughts
188. 8:19generated during your interactive
189. 8:20sessions.
190. 8:22You can then add those back into your
191. 8:24sources and have the studio generate an
192. 8:26output based on what you produced within
193. 8:29that chat session.
194. 8:31Additionally, Notebook LM can be
195. 8:33integrated directly into the Gemini chat
196. 8:36interface.
197. 8:38This eliminates tab switching, allowing
198. 8:41for specialized research, immediate
199. 8:43summarization, and content creation
200. 8:46within a single organized workflow.
201. 8:49We will conclude this introductory
202. 8:51overview by addressing some frequently
203. 8:54asked questions. What languages can I
204. 8:56use with Notebook LM?
205. 8:59You can use Notebook LM in over 50
206. 9:01languages in the browser or with a
207. 9:04limited set of languages in mobile.
208. 9:07Check out other videos in this series to
209. 9:10learn how to change your output language
210. 9:12in Notebook LM.
211. 9:16What level of data privacy do I have
212. 9:18when using Notebook LM?
213. 9:21According to their terms of use,
214. 9:24Notebook LM uses the files you add, the
215. 9:27outputs you generate, and your chat
216. 9:29history to build your knowledge base and
217. 9:32assist with your research and tasks.
218. 9:35The content in Notebook LM will not be
219. 9:38used to directly train the foundational
220. 9:41AI models
221. 9:42unless you choose to provide feedback.
222. 9:46Data that Notebook LM shares with other
223. 9:48Google services such as the Gemini app
224. 9:52is used per the Google privacy policy
225. 9:55and any service specific notices.
226. 9:59Be sure to review the terms and policies
227. 10:01closely
228. 10:03especially if you're working with any
229. 10:05kind of protected data.
230. 10:07Links are provided in the note section
231. 10:09of the slides for this session.
232. 10:13Are there limits for my use of Notebook
233. 10:15LM?
234. 10:17With Notebook LM, you get 100 notebooks
235. 10:21with up to 50 sources each and 500,000
236. 10:25words each or up to 200 megabytes for
237. 10:28local uploads.
238. 10:30There's no page limit.
239. 10:32Additionally you have daily limits of 50
240. 10:35chat queries and three media
241. 10:38generations.
242. 10:40You may also experience a source import
243. 10:42fail if your original PDF file is
244. 10:45copyright protected.
245. 10:48At the paid tier Google One AI Premium
246. 10:51or Pro
247. 10:53you can create up to 500 notebooks with
248. 10:56up to 600 sources per notebook.
249. 10:59Daily chat limits are increased
250. 11:01significantly as well as media generated
251. 11:05from the Notebook LM studio.
252. 11:07This is subject to change so do check
253. 11:10back in on this especially if you plan
254. 11:13to purchase a particular tier.
255. 11:17What happens if I delete notebooks or
256. 11:19notes generated within a session?
257. 11:22Is there any way to recover them?
258. 11:25Unfortunately there's currently no way
259. 11:28to recover deleted notes or notebooks.
260. 11:32Save to notes frequently during your
261. 11:34sessions and export out frequently to
262. 11:37make sure you have a backup of your
263. 11:39work.
264. 11:41Is it possible to duplicate a notebook?
265. 11:45Currently, this feature isn't supported.
266. 11:49Is it possible to share notebooks?
267. 11:52Yes. Notebooks can be shared for access
268. 11:55and collaboration.
269. 11:56However, our UCLA accounts do not allow
270. 12:00sharing with anyone outside our network.
271. 12:03While the core technology remains the
272. 12:05same, your choice of account impacts
273. 12:08your workflow and collaboration.
274. 12:11Using a UCLA enterprise account offers a
275. 12:14more restricted environment designed for
276. 12:17academic integrity and institutional
277. 12:20security.
278. 12:21This means you cannot share notebooks
279. 12:24with non-UCLA domain users,
280. 12:27keeping your research strictly within
281. 12:29the university ecosystem.
282. 12:31Choosing between a personal or UCLA
283. 12:34account ultimately comes down to whether
284. 12:37you prioritize the security of the UCLA
285. 12:40walled garden or the cross-domain
286. 12:43flexibility of a personal account.
287. 12:46As we look at the current landscape of
288. 12:48LLMs being used for research in early
289. 12:512026,
290. 12:53Notebook LM, now powered by Gemini 3,
291. 12:57really sets itself apart through its
292. 12:59sheer scale and precision.
293. 13:02While competitors like Claude 4 and
294. 13:04ChatGPT 5 offer impressive reasoning
295. 13:07capabilities, Notebook LM's 1 million
296. 13:10token context window allows it to ingest
297. 13:14far more data simultaneously.
298. 13:17We're looking at a capacity of up to 50
299. 13:19sources for free users and 300 for plus,
300. 13:24with a generous 200 megabytes limit per
301. 13:27file.
302. 13:28What truly distinguishes it for academic
303. 13:30and professional workflows, however, is
304. 13:33the source grounding.
305. 13:35Unlike the general responses or simple
306. 13:37footnotes you might see elsewhere,
307. 13:40Notebook LM provides direct links.
308. 13:43Meaning, when you click a citation, it
309. 13:45scrolls you to the exact paragraph in
310. 13:48the PDF.
311. 13:50Combined with its ability to generate
312. 13:52everything from audio and video
313. 13:54overviews to mind maps,
314. 13:56it's less of a chatbot and more of a
315. 13:59comprehensive research ecosystem.
316. 14:02Thank you for watching this Oark
317. 14:04workshop video.
318. 14:06We value your input, so please take a
319. 14:09moment to share your feedback on our
320. 14:10workshops by scanning the QR code on the
321. 14:13screen or following the short link
322. 14:16provided.
323. 14:17We hope this session was helpful and
324. 14:20look forward to seeing you in our next
325. 14:21tutorial,
326. 14:23where we'll dive even deeper into
327. 14:25advanced research workflows with
328. 14:26Notebook LM.