Mayo Oshin
@mayowaoshin
Followers
19,607
Following
785
Media
360
Statuses
2,807
Explore trending content on Musk Viewer
#EURO2024
• 695401 Tweets
England
• 661826 Tweets
#BizimCocuklar
• 331190 Tweets
Saka
• 305893 Tweets
Switzerland
• 190840 Tweets
#TURvHOL
• 170834 Tweets
Arda
• 125525 Tweets
#NEDTUR
• 122546 Tweets
Southgate
• 116832 Tweets
Montella
• 115148 Tweets
Turkey
• 112226 Tweets
Netherlands
• 108892 Tweets
Mary
• 99214 Tweets
Samet
• 91097 Tweets
Inglaterra
• 84473 Tweets
Kane
• 82214 Tweets
Trent
• 81402 Tweets
Foden
• 74994 Tweets
Kenan
• 72041 Tweets
Cenk
• 61328 Tweets
Barış Alper
• 55629 Tweets
Pickford
• 41032 Tweets
Kerem
• 40068 Tweets
هولندا
• 39638 Tweets
Hoca
• 39477 Tweets
Ferdi
• 38652 Tweets
Turquía
• 37867 Tweets
Semih
• 36402 Tweets
Toney
• 35939 Tweets
Van Dijk
• 33657 Tweets
Gakpo
• 29977 Tweets
Hakan
• 28031 Tweets
Canınız
• 26937 Tweets
Holanda
• 25743 Tweets
イングランド
• 23822 Tweets
Conviasa
• 22601 Tweets
Ben Rice
• 19062 Tweets
Países Bajos
• 16413 Tweets
Depay
• 13810 Tweets
Hamit
• 13509 Tweets
Teşekkürler Milli Takım
• 12977 Tweets
オランダ
• 11778 Tweets
GPT-4 scanned the entire frontend of my
@LangChainAI
chatbot codebase for bugs and errors.
Here's what happened and what I noticed...
#openai
#chatgpt
#gpt4
36
172
1K
My codebase for the GPT-4 pdf chatbot is now trending on github after gaining 733 stars in 4 days.
Thank you all for the support 🙏
github repo:
29
196
1K
GPT-4 Tutorial: How to 'Chat' With Large PDF Documents Using
@LangChainAI
and
@pinecone
This is the follow up guide to the demo video of 'chatting' with a 56-page legal PDF document about the famous supreme court case: Morse v. Frederick
Links below...
#gpt4
#law
#business
21
177
902
Woke up to No.3 trending on github and 1,800 stars 🤯
My codebase for the GPT-4 pdf chatbot is now trending on github after gaining 733 stars in 4 days.
Thank you all for the support 🙏
github repo:
29
196
1K
16
52
734
Tutorial Youtube Video:
Github repo (please note this the original template used for the demo adapted for this usecase):
My visual diagram of the pdf chatbot architecture below...
15
71
643
By popular demand, myself and
@atroyn
partnered to create a variation of the GPT-4 PDF chatbot codebase using
@trychroma
(an open source vector store solution)
You can try it here:
27
108
633
How to chat with any codebase.
This demo uses my GithubLoader PR contribution, which was added to
@LangChainAI
library recently.
Here's how to use it...
24
83
623
One major downside of releasing content and open source code is the handful of users who copy and paste for commercial use without giving credit.
All of a sudden there's a wave of "chat with pdf" youtube videos using my exact content, diagrams and images.
Plus, some "AI"
72
36
623
How to Chat With Your Data in Private Without Internet (Using MPT-30B Open-Source LLM)
In this brief clip, I uploaded
@nvidia
's latest earnings call transcript and asked MPT-30b to extract insights about the key drivers behind its rapid revenue growth.
40
134
608
GPT-4 Tutorial: How to Chat With Multiple PDF Files
This is the follow up guide to the demo video below showcasing an AI 'Warren Buffett' financial analyst chatbot analyzing 1,000 PDF pages of Tesla's 2020-2022 10-k annual reports accurately.
Links below..
#investing
#finance
18
99
620
For a more comprehensive step-by-step beginner's
guide on how to build a chatbot like this for your data, you can join the waitlist for the upcoming program here:
22
59
545
How to Analyze Tables In Large Financial Reports Using GPT-4 (w/
@jerryjliu0
)
Most corporate docs contain a mix of text and tables. But if you use simple RAG split and chunk methods, the AI model will likely hallucinate due to embedding split tables.
To solve this,
@jerryjliu0
12
94
534
I recently added a multiple PDF file upload feature to the gpt4 pdf repo. Now you can 'chat' with multiple large PDF docs.
3,400 stars so far and still climbing 🙏
code:
GPT-4 Tutorial: How to 'Chat' With Large PDF Documents Using
@LangChainAI
and
@pinecone
This is the follow up guide to the demo video of 'chatting' with a 56-page legal PDF document about the famous supreme court case: Morse v. Frederick
Links below...
#gpt4
#law
#business
21
177
902
29
62
452
🤯 Over 1,000 founders, analysts, and corporate leaders signed up for the "build a ChatGPT chatbot for your data" waitlist.
Thank you all for the support 🙏
Since there's high interest, I've honored my promise to open up enrollment for the first cohort.
More details below...
15
61
422
GPT-4 self-repair prompt...
"You are a helpful programming assistant and an expert Python programmer . You are helping a user write a program to solve a problem . The user has written some code , but it has some errors and is not passing the tests . The user has spent some time
9
64
392
Recently, there's been a lot of excitement and "cool" demos around
@OpenAI
's new function calling models. But there's little talk about how it rapidly increases token consumption and cost.
I ran a short experiment that discovered a 500% increase in token usage...
36
56
380
Tutorial Youtube Video:
Github repo(code):
My visual diagram of the pdf chatbot architecture below...
20
53
368
GPT-4 Vision: How to use
@LangChainAI
with Multimodal AI to Analyze Images, Tables and Texts in Financial Reports.
In this in-depth, practical workshop with
@RLanceMartin
, you'll learn how use multimodal RAG with
@OpenAI
's multimodal GPT-4V to analyze documents that contains
8
64
353
One major drawback of semantic search is the inability to perform structured queries i.e. "What is the average dollar spend of our customers?"
One way to overcome this limitation is to combine AI's text-to-SQL for structured data with semantic search for unstructured data.
12
40
308
9
10
300
Building an AI resume analyser and scoring tool using
@OpenAI
's new AI Assistants and GPT-4-1106.
Goal: Save valuable recruitment time by using AI to assess if a candidate is a good fit.
In the demo video, I uploaded two files: Job description and candidates resume.
21
31
302
AI winter is coming...
"Shitfluencers" and popular tools + frameworks are under heavy scrutiny. Users are demanding much more substance than what they were promised.
The "trough of disillusionment" is rapidly approaching as expectations have not been met. But the supply of
41
41
298
4 simple ways to reduce the cost of using GPT-4 by up to 98% (and improve accuracy at the same time).
1. Combine prompts: Join two or more queries to make a single API call instead of multiple separate calls.
2. Prompt selection: Decrease prompt size by selecting a small optimal
15
55
295
Open source llm models are catching up a lot faster than most expected.
Today, it's possible to host a private ai chatbot on your computer without internet access.
I've been messing around with
@MosaicML
's latest MPT-30b chat model. Here I ask about financial ratios...
8
30
258
A recent paper on "plan and solve prompting" discovered that LLMs instructed to "think step by step" (ZeroShot-CoT) significantly underperform those instructed to "first understand the problem and devise a plan to solve the problem."
Source:
11
53
250
How to build a
@NotionHQ
chatbot to ask your knowledge base questions using
@LangChainAI
,
@pinecone
and
@nextjs
Credit to
@Cron
for the notion support docs used in the demo.
Tutorial video:
14
28
242
RAG isn’t enough to eliminate AI hallucination. A hydbrid use of external tools, query transformation, fine tuned embeddings, prompt engineering, filtering and more are required.
@llama_index
is leading the charge on advanced rag ideas.
Here is why I am so doubtful about the "talk-to-your-data" use of AI.
This is Google NotebookLM, a cool tool that lets you use AI on data sources. Even though the document search retrieves the right information (it is Google, after all), the LLM answer has subtle hallucinations.
39
123
994
4
33
247
Prototyping a free AI youtube summarizer that can run quickly and accurately on any computer.
I have a ton of valuable long form (2hr +) youtube videos on my watch list (example below is Jeff bezos' recent interview with Lex Fridman), but can never find time to watch them.
I
22
18
243
Do you actually need a vector database to chat with your data? Probably not.
A dedicated vector database may not be necessary for less than 1 million embeddings. An exact nearest neighbors calculation using numpy can be found in ~50 ms.
Source:
12
34
234
This is a painful lesson AI founders will learn the hard way: distribution is much more important than product.
As copycats continue to flood the market, the winners will be those with strong marketing, copywriting, and brand, not technical skills.
Marketing is going to be the most important skill in the next 10 years.
Building products has become easy.
There's no code. There's AI. There are endless opportunities to build software, physical or info products for a single person from the couch.
But getting customers has
45
30
218
17
30
215
My humble prediction is that AI will soon democratise access to high-quality investment research that costs thousands of dollars.
In my prototype demo, I use layman words to ask for real-time comparisons between Microsoft and Google stock, in addition to sector performances.
18
33
206
It's been a lot of fun contributing to
@LangChainAI
. Somehow along the way I ended up being a top 10 contributor 😅
You can check out some of my contributions including the Notion and Github loader which enable you to "chat" with notion pages and github codebases.
Docs:
6
13
204
After 2 weeks of sleepless nights, I've successfully graduated the first cohort of the AI "chat with data" program.
The bar was set very high as the majority of the students are senior professionals who work at leading institutions including Evercore, PwC, UPenn, and more...
10
11
188
For a more comprehensive step-by-step beginner's
guide on how to build a chatbot like this for your data, you can join the waitlist for the upcoming program here:
6
27
182
If you’re looking to build an AI chatbot for your data,
@pinecone
database is one of the top choices for fast and accurate search results.
Join myself and
@RoieSchwabco
(Developer advocate at Pinecone) tomorrow to learn more about how to use Pinecone for your use case and debug
11
21
182
24 hours to go before the launch of the beginner's "ChatGPT your data" program.
I'm blown away by the diversity of the community: From PE fund managers, law partners and real estate managers to founders of tech startups, agencies, PhD analysts, and more.
Learn more below...
Announcing the third batch of AI chatbot experts sharing insights during the upcoming "Build a ChatGPT Chatbot For Your Data" program
AI SQL experts:
NLP Practitioner,
@fpingham
Bioinformatics Analyst,
@JonZLuo
Semantic search expert:
@transitive_bs
Learn more below...
3
8
68
9
11
174
Found this comprehensive spreadsheet on the key features of popular Vector Databases used to build out AI chatbot solutions.
Spreadsheet:
Key takeaways:
- The most well-rounded solutions include
@weaviate_io
,
@vespaengine
, and
@elastic
.
- The top
12
28
172
Most popular semantic search methods are useful for retrieving relevant documents based on *similarity*, but they often fail to retrieve based on *description.*
4
28
166
I tested the new `gpt-4-0613` function calling in
@LangChainAI
's latest openai-agent.
Scenario: Hosting events for loyal customers.
Tools: Web search, ChinookDatabase, and math.
Question: "What are the best venues for business events in the city with the most customers?"
5
21
156
50,000+ words (a book's worth) of
@TomFrankly
's Notion website guides transformed into an interactive web chatbot for
@NotionHQ
advice.
I built this with a potential future dream team for AI apps
@LangChainAI
+
@supabase
Demo video, tutorial, repo, and visual guide below...
7
21
131
Announcing the first batch of AI chatbot experts sharing insights during the upcoming "Build a ChatGPT Chatbot For Your Data" program
Founder at Chatbase,
@yasser_elsaid_
Tech Lead, PhD Stanford
@RLanceMartin
Learn more about the program below...
#GPT4
🤯 Over 1,000 founders, analysts, and corporate leaders signed up for the "build a ChatGPT chatbot for your data" waitlist.
Thank you all for the support 🙏
Since there's high interest, I've honored my promise to open up enrollment for the first cohort.
More details below...
15
61
422
5
9
128
Recently, there has been a sharp increase in demand from businesses looking to integrate AIsolutions that enable customers and employees to "chat" with their data.
As a result, a new in-demand tech stack has emerged including
@OpenAI
,
@LangChainAI
,
@pinecone
, and
@trychroma
.
3
14
121
The future of agents will likely involve Human-AI collaboration to solve difficult "everyday" decisions i.e. organising an efficient travel itinerary, researching the best laptop for your work.
A research team at Berkeley, MIT, and Microsoft, recently put together a study on
5
20
122
Announcing the second batch of AI chatbot experts sharing insights during the upcoming "Build a ChatGPT Chatbot For Your Data" program
Co-founder at LangChainAI,
@hwchase17
Co-founder at Chroma,
@atroyn
Learn more about the program below...
Announcing the first batch of AI chatbot experts sharing insights during the upcoming "Build a ChatGPT Chatbot For Your Data" program
Founder at Chatbase,
@yasser_elsaid_
Tech Lead, PhD Stanford
@RLanceMartin
Learn more about the program below...
#GPT4
5
9
128
3
14
117
Want to create a "Chat-GPT" for your own documents and knowledge base?
Join the webinar hosted by
@hwchase17
(Co-Founder of LangChain) and myself, on how to build a simple chatbot for your custom documents using Javascript and LangChain.
#chatgpt
3
16
116
Openai just put a lot of startups out of business. The new plugins support allows ChatGPT to integrate with third-party services and access live data.
The first plugins have been created by Expedia, FiscalNote, Instacart, KAYAK, Klarna, Milo, OpenTable, Shopify, Slack, Speak,
We are adding support for plugins to ChatGPT — extensions which integrate it with third-party services or allow it to access up-to-date information. We’re starting small to study real-world use, impact, and safety and alignment challenges:
901
4K
19K
5
16
102
@matthewmulroe
@elonmusk
Yeh my template gpt4-pdf-chatbot code is open source here:
This demo is just a variation of this.
3
12
106
Who wins: GPT-3.5-turbo vs.
@MistralAI
mixtral-8x7b for solving reasoning tasks and math problems?
I ran a super short test on the
@Vercel
AI Playground...
Looks like GPT-3.5-turbo is a clear winner.
10
10
103
How to Compare Multiple Large PDF Files Using AI (w/
@jerryjliu0
, Co-Founder of LllamaIndex)
Watch:
In this video tutorial , you'll learn how to ask complex questions and compare valuable information across multiple large pdf documents using
1
19
98
I recently hosted a question & answer session with
@hwchase17
on
@LangChainAI
Agents and "How to Build Personal Assistants For Your Data".
You can watch the full discussion here:
3
8
92
Tutorial Youtube Video:
Github repo(code):
My visual diagram of the pdf chatbot architecture below...
4
8
93
It was a pleasure to sponsor and deliver a multimodal AI presentation at the first ever African
@LangChainAI
meetup in collaboration with
@mlopscommunity
Lagos.
Topic: 'Building production ready LLM apps with LangChain.'
There's a ton of engineering + ML talent in Africa who
7
20
92
Playing around with
@vercel
's new AI SDK and open source LLMs ( Open-Assistant SFT-4 12B Model in demo video).
The starter kit UI is barebones, but it's a lot less code and a quicker setup to launch an AI app in Nextjs.
7
12
86
How to make wealth by
@paulg
"If you want to make a million dollars, you have to endure a million dollars' worth of pain."
Source:
2
6
80
AI twitter is noisy. Here are 8 "under-the-radar" twitter accounts I follow for signal:
@abacaj
@RLanceMartin
@TheSeaMouse
@teknium1
@yacineMTB
@nisten
@TheBlokeAI
@Tim_Dettmers
9
9
78
Published Recording of
@pinecone
Q+A twitter spaces with
@RoieSchwabco
,
@yasser_elsaid_
and myself in case you missed it 🔥
5
6
74
Has anyone figured out a way to export twitter bookmarks without paying $000's for API access?
I've created thousands of valuable bookmarks that get lost in the weeds.
Would be great to have a simple interface to search and filter bookmarks.
25
3
74
@ptsi
@LangChainAI
Free alpha:
- Use Langchain's PDF loader to convert the pdfs to structured format
- Use LLM to filter based on "react" and "nextjs"
- For each filtered resume, extract the github repo url, use the git/github loader to extract data.
- Use summarise chain to summarise, then
3
4
71
.
@LangChainAI
+
@pinecone
beginner's tutorial for typescript/javascript developers
Looking forward to a future where JS/TS devs have easy access to traditional ML tools
3
14
68
Announcing the third batch of AI chatbot experts sharing insights during the upcoming "Build a ChatGPT Chatbot For Your Data" program
AI SQL experts:
NLP Practitioner,
@fpingham
Bioinformatics Analyst,
@JonZLuo
Semantic search expert:
@transitive_bs
Learn more below...
Announcing the second batch of AI chatbot experts sharing insights during the upcoming "Build a ChatGPT Chatbot For Your Data" program
Co-founder at LangChainAI,
@hwchase17
Co-founder at Chroma,
@atroyn
Learn more about the program below...
3
14
117
3
8
68
I was recently invited as a generative AI Keynote speaker at the Private Markets European Summit.
My presentation was focused on answering one key question: How can financial institutions integrate generative AI to gain a competitive edge whilst minimising risk.
Here are some
5
6
67
For a more comprehensive step-by-step beginner's
guide on how to build a chatbot like this for your data, you can join the waitlist for the upcoming program here:
2
6
64
I just uploaded a screenshot of a Google maps restaurant listing to
@GoogleAI
's new Gemini Pro Vision...
Not only did it identify the important details about the restaurant listing, but it also recognised nearby tourist attractions on the map itself!
LLM competition is heating
3
10
68
Enrollment now closed😅 It's been exhausting preparing for this but I'm excited to mingle with the community over the next two weeks and build valuable AI chatbot MVPs.
Massive thanks to anyone who supported the program in any way.
Let's go! 🔥
24 hours to go before the launch of the beginner's "ChatGPT your data" program.
I'm blown away by the diversity of the community: From PE fund managers, law partners and real estate managers to founders of tech startups, agencies, PhD analysts, and more.
Learn more below...
9
11
174
9
0
63
@tferriss
This is a straight forward process. Here's a similar visual architecture I used for a 1,000-page PDF annual reports.
4
3
63
📰 Top AI tools featured in the latest 'Chat with Data' newsletter
- Vanna AI: An open source "Chat with Database" framework by
@zain_hoda
- GPT-Newspaper: Create your own curated newspaper with AI [link: ]
- PdftoChat: A free privacy-focused AI
1
12
52
For the rest of the process, you'll need to split the rawDocs into chunks, embed (convert to numbers), and upsert into a vectorstore.
I covered this entire architecture of how to "chat with your data" in my "gpt4-pdf" youtube video below.
3
7
57
During the
@LangChainAI
London event tomorrow, I'll deliver a brief speech/presentation on how to "chat with your data".
If you're due to attend the event, feel free to say hello in-person afterwards.
Coming off the heels of a really fun SF meetup, excited to announce:
🏙️ London Meetup 4/25
Luma:
Space is limited, so only register if you intend to attend in person (we will figure out a remote stream closer to the event)
10
8
85
5
4
55
3 key takeaways from "How to Do Great Work" by
@paulg
1. Work on your own projects. "Develop a habit of working on your own projects. Don't let "work" mean something other people tell you to do. If you do manage to do great work one day, it will probably be on a project of your
3
9
56
Why ChatGPT plugins don't have product-market fit
"A lot of people thought they wanted their apps to be inside ChatGPT, but what they really wanted was ChatGPT in their apps." - Sam Altman
Sam Altman recently said that he doesn't believe that ChatGPT plugins have product-market fit beyond browsing.
A few hypotheses on why (not mutually exclusive):
Correct concept but not good enough yet:
• GPT-4 picks the wrong plugins or fails to chain together multiple calls
20
30
294
9
5
54
Github repo to verify bugs and solutions:
Here's what I noticed after running the same prompt 3 times:
- Response speed is much slower than turbo
- Context window is larger.
- More accurate than previous models
- Less variation of output
0
2
52
😂I guess
@elonmusk
's algorithms kicked in early for my twitter account
Reminds me of meme coins during the 2021 crypto bull run
4
1
53
If you're interested in a more comprehensive step-by-step beginner's guide on how to build a similar chatbot for your data, you can join the waitlist for the upcoming program here:
1
3
52
"GPT-4 currently has a cap of 25 messages every 3 hours. Expect lower cap next week, as we adjust for demand."
This is a significant drop from 100 messages every four hours 4 days ago!
Hundreds of thousands(potentially millions) of people paid for the pro subscription to use
13
6
51
My goal was to create a step-by-step beginner’s training program on how to build an AI chatbot for custom data.
So far based on feedback, that goal has been accomplished. If you'd like to learn more, check out the program details here:
1
6
46
5 key takeaways from
@McKinsey
's "The state of AI in 2023" report.
1. Expected business disruption is significant.
"Respondents predict meaningful changes to their workforces. They anticipate workforce cuts in certain areas and large reskilling efforts to address shifting
5
8
49
@DefiDutch
@LangChainAI
@pinecone
What part 😂 there’s quite a bit of moving parts. But you can start with understanding Langchain here:
2
4
46
Follow the installation instructions in the LangChainjs docs then simply use the GithubLoader to "Load" any github repository into text and metadata.
2
6
46
I'm tired of counting calories. It's exhausting and difficult to stick with, but it's important for my fitness goals.
So I thought...what if instead of manually logging calories into an app i.e.
@MyFitnessPal
, I built an AI calorie counter using
@OpenAI
's GPT-4V model?
5
5
47
Solid stuff from Jon here on incorporating AI to query your SQL database
Want to implement chain-of-thought reasoning in your text-to-SQL tool to help guide the SQL query generation? One way is to provide few-shot examples demonstrating CoT.
@LangChainAI
makes this a piece of cake. (1/4)
18
45
344
0
1
41
The next wave of valuable AI solutions will be tools for debugging, testing, evaluating, monitoring and improving the performance of LLM applications.
🦜🛠️ Introducing LangSmith 🦜🔗
A unified platform to help developers debug, test, evaluate, and monitor their LLM applications.
Integrates seamlessly with LangChain, but doesn't require it.
22
194
1K
2
7
43
@johnbuilds
Really? You were chatting with multiple 1,000 page PDF files inc. financial reports a year ago?
3
0
42
AI will change how we interact with the web—from manual clicking, search and filling forms to simply providing instructions via chat UI text or voice.
Example AI task in video: Book a roundtrip on July 1 from Mumbai to London and vice versa on July 5 for two adults and a
0
5
41
A visual advanced RAG cheatsheet to improve AI chatbot accuracy and performance.
h/t
@llama_index
Full blog post and code:
2
11
39
I don't get the hype behind
@OpenAI
's latest updates. What am I missing here?
1. The significant reduction in embeddings cost (and retrieval) seems to mitigate the use case benefits of the larger context window.
2. "Function calling" isn't a new concept. This has already been
21
0
36
Tutorial Youtube Video:
Github repo (please note this the original template used for the demo adapted for this usecase):
Visual guide...
2
3
38
1
11
36
If you've benefited from my popular open-source
@LangChainAI
chatbot repos, take a few seconds to nominate me as a github star via the link below.
My
@github
username is mayooear.
Cheers 🙏
5
8
37
Looks like pgvector (the PostgreSQL extension for vector similarity search) is significantly slower and less accurate than dedicated vector DB's.
Why you should never use pgvector (e.g.
@supabase
Vector Store) for production:
😮 pgvector is 20x slower than a decent vector DB (e.g.
@qdrant_engine
)
🤯 And it's a full 18% worse in finding relevant docs for you
And this can happen at as little as 10K documents when chunked!
33
30
217
7
5
35
Building AI applications that are ready for production use can be challenging. From hallucinations to slow speed, these challenges can deter users, lead to bad decisions, and cost a ton of money.
During this free 'LLMs in Production'
@llama_index
webinar, I'll be sharing some
1
4
36
@andrewparker
@frantzfries
Facts. The figma story is survivorship bias. Most startups that don't validate users actually want to pay for their product fail.
1
1
35