CS50
First, youโll learn how GPT-4 works and why human language turns out to play such a critical role in computing. Next, youโll see how AI-native software is being made.
Taught by Ted Benson, founder of Steamship, MIT Ph.D., & Y Combinator Alum; and Sil Hamilton, researcher of emergent AI behavior at McGill University.
Slides at: https://cdn.cs50.net/2023/spring/talks/gpt4/gpt4.pdf
***
This is CS50, Harvard University’s introduction to the intellectual enterprises of computer science and the art of programming.
***
HOW TO SUBSCRIBE
http://www.youtube.com/subscription_center?add_user=cs50tv
HOW TO TAKE CS50
edX: https://cs50.edx.org/
Harvard Extension School: https://cs50.harvard.edu/extension
Harvard Summer School: https://cs50.harvard.edu/summer
OpenCourseWare: https://cs50.harvard.edu/x
HOW TO JOIN CS50 COMMUNITIES
Discord: https://discord.gg/cs50
Ed: https://cs50.harvard.edu/x/ed
Facebook Group: https://www.facebook.com/groups/cs50/
Faceboook Page: https://www.facebook.com/cs50/
GitHub: https://github.com/cs50
Gitter: https://gitter.im/cs50/x
Instagram: https://instagram.com/cs50
LinkedIn Group: https://www.linkedin.com/groups/7437240/
LinkedIn Page: https://www.linkedin.com/school/cs50/
Medium: https://cs50.medium.com/
Quora: https://www.quora.com/topic/CS50
Reddit: https://www.reddit.com/r/cs50/
Slack: https://cs50.edx.org/slack
Snapchat: https://www.snapchat.com/add/cs50
SoundCloud: https://soundcloud.com/cs50
Stack Exchange: https://cs50.stackexchange.com/
TikTok: https://www.tiktok.com/@cs50
Twitter: https://twitter.com/cs50
YouTube: http://www.youtube.com/cs50
HOW TO FOLLOW DAVID J. MALAN
Facebook: https://www.facebook.com/dmalan
GitHub: https://github.com/dmalan
Instagram: https://www.instagram.com/davidjmalan/
LinkedIn: https://www.linkedin.com/in/malan/
Quora: https://www.quora.com/profile/David-J-Malan
TikTok: https://www.tiktok.com/@davidjmalan
Twitter: https://twitter.com/davidjmalan
***
CS50 SHOP
***
LICENSE
CC BY-NC-SA 4.0
Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Public License
https://creativecommons.org/licenses/by-nc-sa/4.0/
David J. Malan
https://cs.harvard.edu/malan
malan@harvard.edu
David "1.5x" Malan ๐
Very interesting, thanks!
Chat gpt is broken because everyone knows the only word that should follow "I am" is "groot"
The steamship/hackathon link is dead.
Detailed Summary:
03:28 ๐ง GPT-4, a large language model, is trained to predict the next word in a sequence of text. It uses a vocabulary of 50,000 words to generate new text by predicting the most likely word to follow a given sequence.
08:09 ๐ค ChatGPT evolved into a versatile tool after instruction tuning, becoming capable of answering questions, providing assistance, generating content, and more.
09:49 ๐ Building applications with ChatGPT involves wrapping it in endpoints that inject specific perspectives or goals into the conversation. This allows for personalized interactions with the language model.
14:07 ๐ฌ Companion bots can be created by customizing GPT's prompts to give it a particular personality and role. This enables interactions like language tutoring or providing personalized advice.
18:27 ๐ Question-answering apps involve segmenting documents, converting text into embedding vectors, and using these vectors to find relevant information within the documents.
20:33 ๐ค Using vector databases to store numbers for question search and retrieval.
21:00 ๐ Developing AI-native software by embedding queries and document fragments.
22:12 ๐ Using vector approximations and database fragments to answer questions.
23:10 ๐ Repeating context-specific information retrieval using software prompts.
23:51 ๐ฃ Creating question-answering systems using basic prompts and tools.
24:47 ๐ Building utility functions for automating basic language understanding tasks.
26:14 ๐ Leveraging AI to generate content suggestions based on domain knowledge.
32:09 ๐ Exploring multi-step planning AI (baby AGI) for self-directed tasks.
37:39 ๐ง Addressing hallucination issues through examples and tools.
41:28 ๐ค Considering collaboration between AI agents for better outcomes.
42:09 ๐ง Collective Intelligence: Instead of making a single AI smarter, using multiple software agents with distinct roles can solve complex problems by drawing upon their collective intelligence.
42:37 ๐ฐ Overengineering and Consensus: Drawing an analogy to space shuttles, spacecraft systems use redundant computers to achieve consensus on critical decisions, emphasizing the importance of agreement and minimizing errors.
43:21 ๐ฌ Mode of Interaction: Using specific prompts can guide the language model into different modes of interaction, adapting its responses to the desired context and role.
44:17 ๐ Narrative and Simulation: GPT-4 can simulate personalities and interactions, assuming roles and completing stories as different characters, enhancing its conversational capabilities.
46:01 ๐ค Logic and Reasoning: GPT-4's ability to pass tests like LSAT suggests some rational or logical capabilities, but it still requires experimentation to determine optimal prompts and strategies for different tasks.
47:26 ๐ผ Business Value: Startups are leveraging GPT-4 to create AI-powered products and services, emphasizing the combination of GPT-4's capabilities with domain knowledge and data for practical applications.
48:36 ๐ Evolution of Models: The trajectory of AI models like GPT-4 indicates that they will become integral to various devices, much like microprocessors, leading to widespread adoption and incorporation into everyday applications.
49:59 ๐ Reliable Interaction: Techniques for reliable interactions include providing examples, using diverse prompts, and applying post-processing to ensure successful responses.
51:11 ๐ Privacy and IP: Different deployment options exist, including relying on cloud providers, private hosting, or running models on your own machines, with varying implications for privacy and intellectual property protection.
Great talk, but it wuld have been event better if you could a) hear the questions during Q&A b) the presenters repeating the questions.
Wild Card: Yes Man
Wow, they were so happy for pizza at the end
Remind me of die hard 4, happening at a random computer unable to track
I am Batman
This guy is quite electrical!!!
๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐๐
7:28
train the model to be really large really wide and we have to train it for a really long time and as we do that the
7:33
model gets more and more better and expressive and capable and it also gets a little bit intelligent and for reasons
Damn right out the box dude look like he hit Starbucks w 5 expresso shots ๐
Miss the classes of David Malan ๐ข
These apps are in reach for everybody and not that difficult which means, someone with 5 milion dollars will make a better app that you with 500 dollars.
Getting into chat gpt startups is a bad idea.
I am still in dilemma whether GPT will replace software engineers or not. i also want to be a software engineer, but afraid of tools like gpt
I've been working on ~2000 token long conversational prompts with response formatting and decision making even with data structures in the context and it just keeps on giving, spent hours tweaking my prompts and they keep on giving, amazing tech!
Where can I get the code repo for the demo?
funny to see that ref to pizza kk
The problem is reducing or narrowing down answer or response to a particular problem, if it does have a solution than it was simple test, if it not sounding good or text is jumble of mess ,can than use percentage and text similarity between Thinking process for it check possibilities than can sound or have a meaningful process ๐ฎ alot of ways a person can think of an response
์ด ใ ใ ์์์ ์ ๋ง ์ด์ ์ ์ ๋๋ค.
์ด๋ฐ ์ ^o^ ์์ ๋ง์ด ์ฌ๋ ค์ฃผ์ธ์!
์ด ๋ถ ^-^ ์ ๊ทธ๋ฅ ๋๋จํ ๊ฒ ๊ฐ์์.
์ด ์์ (^o^) ์ ๋ณด๋ฉด ๋๋ ๋ญ๊ฐ ์๋ก์ด ๊ฒ์ ํ ์ ์๊ฒ ๋ค๋ ์๊ฐ์ด ๋ค์ด์.
์ด ์์ ~์์ฐ~ ์ ์ ๋ง ๋ฐ๋ปํ ์์์ ๋๋ค.
์ญ์ ^^ ํ๊ตญ์ธ์ ๋๋จํ๋ค์!
์ด ์ ^o^ ์์ ์ ๋ง ๋ฐ๋ปํ ์์์ ๋๋ค.
๋๋ ์ด ์์์ ๋ณด๋ฉด ๋๋์ด ์ข์์ ธ์ ๋ ์ด์ฌํ ์ผํ๊ฒ ๋ฉ๋๋ค.
3.171 nhz what frequency for?
bmi?
reason I like the job
omg my mind is blowing up with happiness!
This is financial advice and I never give financial advice: DONT LEAVE DURING THE BEAR. If you donโt want to investโฆlearn. If you donโt want to learnโฆbuild. If you donโt want to build observe. DO SOMETHINGโฆother than leave. There is so much opportunity here. Take advantage!
The more I listen to this videos the more I get motivated to build AI products. However the more I realize about how less I know about this AI field. My CV says 3 yrs experience AI/ ML engineer though ๐ ๐๐
Thanks!
๐ฏ Key Takeaways for quick navigation:
00:00 ๐ง Understanding GPT: Introduction to GPT and its various descriptors.
10:43 ๐ Expanding GPT's Abilities: GPT's role in question-answering and how it becomes more than just a language model.
16:59 ๐ค Companionship Bots: Creating personalized AI companions.
19:09 ๐ก Question Answering with GPT: Leveraging GPT for question-answering.
19:52 ๐ How vector databases work
21:00 ๐ค Building question-answering bots
25:01 ๐ ๏ธ Building utility function apps
28:06 ๐ Leveraging creativity and domain knowledge
32:36 ๐ Exploring baby AGI and self-directed AI
40:31 ๐ง How GPT-4 works and addressing hallucinations
43:21 ๐ฃ๏ธ Influencing GPT-4's behavior through language
45:03 ๐ผ Use cases and business value of AI apps
48:36 ๐ The evolution of AI models like GPT-4
51:11 ๐ Privacy implications of GPT-4 prompts and IP
Made with HARPA AI
This is so basic but is so necessary, really good to be able to watch this. Thank you.
My dad went to Harvord and he works for the internet
I did this with The Queen's Gambit and what GPT knew about Beth Harmon and Alma Wheatley was just uncanny. They really came alive and I asked very detailed questions. And I was just using the public 2021 interface. GPT even made up the sequels with its limited knowledge. Soon LLMs will be able to watch entire films and hold discussions about them. I can barely wait for that day.
https://www.steamship.com/hackathon 404
Respect ๐
38:29 Dunning-Kruger effect?
๐ฏ Key Takeaways for quick navigation:
00:00 ๐๏ธ Introduction and Interest in AI
03:02 ๐ค Understanding GPT as a Language Model
05:46 ๐ก GPT's Role in Question Answering
09:49 ๐ค Building Companionship Bots
14:07 โ Building Question Answering Apps
19:52 ๐ How GPT-4 Works:
21:00 ๐ ๏ธ Building AI Applications:
25:01 ๐ Utility Functions:
28:06 ๐๏ธ Creativity with AI:
34:57 ๐ค Baby AGI and Auto-GPT:
37:39 ๐ญ Mitigating Hallucinations:
40:31 ๐ GPT-4 Capabilities and Hallucinations
41:14 ๐ Collective Intelligence in Software Development
43:21 ๐ฌ Influencing GPT-4's Behavior
46:15 ๐ง GPT-4's Ability to Reason
48:36 ๐ป The Evolution of AI Models
Made with HARPA AI
We needed 50 mins to answer that question with a Harvard audience? ๐
Somehow I just don't trust these people and their ultimate addenda to make money at the expense of those who are willing to buy into it. Our human brains do just fine and will continue to do so if we continue to use them. If we depend on Chat to think for us our brains will weaken and dumb us down to the point where we can be controlled.
it seems reiterative of the old pattern or established already by a human and compilation and classification can mislead that AI get some thing higher, and yet it is mundane knowledge GIGOโnothing you do not know before. It is recursive of the already learned and recessive or even current error backward like Hinton does not improve the quality of knowledgeโit is a sheer Machine not Intelligenceโ-advise not to use I word like John MCCathy. did, Be honest and go back like G Hinton to learn how to resolve the BB, the black boxes before a babbling nonsense with the Tool or Toy box without intellectual foundation and DISCIPLINES???
Pizza ๐ pizza ๐ pizzaโฆ ๐
Is a shame, programmers will NEVER be as smart as 10-20 years ago, never again, when u start using less your potential, then you arenโt going anywhere