Discover and read the best of Twitter Threads about #LLMs
Most recents (24)
I remember when I got into the beta user list for GPT-3.0, and interacting with it became my daily hobby.
Fast forward three years and your stance on large language models can make or break your company's stock price!
Business adoption is imminent. 🧵👇🏻 #llms
Fast forward three years and your stance on large language models can make or break your company's stock price!
Business adoption is imminent. 🧵👇🏻 #llms
I’m super excited about the emerging tool stack that helps to deploy and scale real-world LLM solutions.
I made a YouTube video about a big-picture overview of four tools from the emerging LLM developer stack that lets you build advanced LLM apps 👇🏻
I made a YouTube video about a big-picture overview of four tools from the emerging LLM developer stack that lets you build advanced LLM apps 👇🏻
YouTube video link:
In addition to the wonderful personal experiences, #EAMT2023 was also a professional highlight for me that has really inspired me and reassured me that my work is useful and appreciated. A bragging-thread 👇 cc @EAMT_2023 @lt3ugent @SignONEU
🎉 I presented MATEO, a tool for effortless #MT evaluation! It received an overwhelming response from researchers, users, and industry, validating my belief that improving accessibility to #opensource #evaluation methods fills a need. Check it out here: lt3.ugent.be/mateo/
The project funding (thank you @EAMTee and @CLARINERIC!) is ending this month but I've had the pleasure to talk to many engaging people to keep the project alive in an #openscience manner for the foreseeable future. All contributions and ideas are welcome! github.com/BramVanroy/mat…
📯 Please RT!
✨ A Message To The #AICommunity
📣 I can't stress this enough: #LLMs are designed to reflect and amplify what we feed them. They're like kids, or psychedelics. (My friend @NeseLSD co-authored a superb paper on #psychedelics as non-specific amplifiers I will post… twitter.com/i/web/status/1…
✨ A Message To The #AICommunity
📣 I can't stress this enough: #LLMs are designed to reflect and amplify what we feed them. They're like kids, or psychedelics. (My friend @NeseLSD co-authored a superb paper on #psychedelics as non-specific amplifiers I will post… twitter.com/i/web/status/1…
1️⃣ Okay, so! My first point in response to all of the fear-mongering and doomsaying in the tech world right now — beyond the obvious, which is that there's an inherent conflict in hastily bringing forth tools you fear and maybe that's worth some personal reflection, because it's… twitter.com/i/web/status/1…
♻️ Side Note: I plan on writing all of this up into a more thorough and well-formatted article with an extensive bibliography, but for the time being I want to drop a couple of relevant tweets that serendipitously came up in my feed today.
Here's one from from @speakerjohnash,… twitter.com/i/web/status/1…
Here's one from from @speakerjohnash,… twitter.com/i/web/status/1…
New colab!
@LangChainAI 🤝 @elastic
LangChain is a cutting-edge library to make large language models (LLMs) more usable and versatile.
In synergy with ElasticSearch, this powerful duo opens new doors in application development and search applications.
Let's dive in 🧵
#AI
@LangChainAI 🤝 @elastic
LangChain is a cutting-edge library to make large language models (LLMs) more usable and versatile.
In synergy with ElasticSearch, this powerful duo opens new doors in application development and search applications.
Let's dive in 🧵
#AI
LangChain is a toolbox providing a standard interface for LLMs & facilitating their integration with other tools.
It makes developing applications like chatbots, decision-making agents, and more, easier by blending LLMs with other computations & knowledge bases.
It makes developing applications like chatbots, decision-making agents, and more, easier by blending LLMs with other computations & knowledge bases.
Who's it for?
@LangChainAI is perfect for developers looking to harness the potential of LLMs to build transformative applications.
From question-answering apps to intelligent agents, the scope of possibilities with LangChain is immense.
@LangChainAI is perfect for developers looking to harness the potential of LLMs to build transformative applications.
From question-answering apps to intelligent agents, the scope of possibilities with LangChain is immense.
#dts23 is about to begin! Excited to be here!
Last week, we discussed techniques to speed up the training speed of large language models🔥💨
How about saving memory during inference? 🧠💾 Check out int8 & int4 quantization, which is supported in Lit-LLama 👉github.com/Lightning-AI/l…
🧵1/3
#LLMs #ML #DeepLearning
How about saving memory during inference? 🧠💾 Check out int8 & int4 quantization, which is supported in Lit-LLama 👉github.com/Lightning-AI/l…
🧵1/3
#LLMs #ML #DeepLearning
How does int8 quantization work? 🤔
It's a 2-part procedure with
1) using 8bit quantization
2) 16-bit matmuls for outlier feature dimensions
Check out the LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale paper for details arxiv.org/abs/2208.07339
🧵2/3
It's a 2-part procedure with
1) using 8bit quantization
2) 16-bit matmuls for outlier feature dimensions
Check out the LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale paper for details arxiv.org/abs/2208.07339
🧵2/3
And how about int4? 🤔
It's a one-shot weight quantization method based on approximate second-order information⚙️📉
For more details, see GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers: 📚🔍 arxiv.org/abs/2210.17323
🧵3/3
It's a one-shot weight quantization method based on approximate second-order information⚙️📉
For more details, see GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers: 📚🔍 arxiv.org/abs/2210.17323
🧵3/3
Welcome to day 2 of the 5th World Chatbots & Voice Summit, hosted the the beautiful city of #Berlin
I'll be keeping this thread going to cover all the presentations and learnings of the day.
As co-chair, I'll be looking after the chatbot track, today.
#chatbotvoice2023
I'll be keeping this thread going to cover all the presentations and learnings of the day.
As co-chair, I'll be looking after the chatbot track, today.
#chatbotvoice2023
First up, Alin from @SapienceS2P
Talking about how to deliver value and a competitive edge through customised chatbot solutions
SapienceS2P is an #SAP procurement consultancy
Talking about how to deliver value and a competitive edge through customised chatbot solutions
SapienceS2P is an #SAP procurement consultancy
Taking us through their #procurement chatbot, Sapience Satori, which can integrate with #ChatGPT, most SAP APIs and also communication channels - #teams #slack #whatsapp
Welcome to #chatbotvoice2023 from Berlin.
I'm throwing up some tidbits around #conversationalai live from the 5th World Chatbots & Voice Summit from Berlin.
I'm throwing up some tidbits around #conversationalai live from the 5th World Chatbots & Voice Summit from Berlin.
Testing is an impossible manual task and it needs to be automated. Done wrong, it can cost heavily. So it is a huge, but important challenge.
#LLMs like GPT-4 are stuck in time — they only understand the world as it exists within their training data...
This causes issues when we need up-to-date info or interaction with specific datasets — here's how we fix that
pinecone.io/learn/langchai…
This causes issues when we need up-to-date info or interaction with specific datasets — here's how we fix that
pinecone.io/learn/langchai…
🔍📚 Retrieval augmentation uses an external knowledge base as a "window" into the world — we explore how to implement this for LLMs using @LangChainAI, reducing hallucinations and keeping info up to date
📖🔧 To give LLMs relevant source knowledge, we create a knowledge base:
- We start with a dataset (e.g., a subset of Wikipedia)
- Split it into chunks
- Generate embeddings (numerical representations) of the "human meaning" behind each chunk
- We start with a dataset (e.g., a subset of Wikipedia)
- Split it into chunks
- Generate embeddings (numerical representations) of the "human meaning" behind each chunk
🧵
Czy SZTUCZNA INTELIGENCJA jest szkodliwa?
Pod względem językowym - NA PEWNO!
Czy termin “sztuczna inteligencja” faktycznie oddaje to co robi:
- uczenie maszynowe
- sieci neuronowe
- deep learning?
#ai #etyka #język #filozofia #ChatGPT #LLMs #sztuczna #inteligencja
1/40+👇
Czy SZTUCZNA INTELIGENCJA jest szkodliwa?
Pod względem językowym - NA PEWNO!
Czy termin “sztuczna inteligencja” faktycznie oddaje to co robi:
- uczenie maszynowe
- sieci neuronowe
- deep learning?
#ai #etyka #język #filozofia #ChatGPT #LLMs #sztuczna #inteligencja
1/40+👇
2/ CO AUTOR MIAŁ NA MYŚLI
Termin “sztuczna inteligencja” powstał w 1956 podczas konferencji w Darmouth - założycielskiego zlotu dla tej gałęzi nauki. Autorem był John McCarthy, wówczas 29-letni wykładowca.
Termin “sztuczna inteligencja” powstał w 1956 podczas konferencji w Darmouth - założycielskiego zlotu dla tej gałęzi nauki. Autorem był John McCarthy, wówczas 29-letni wykładowca.
3/ Już w trakcie konferencji pojawił się pierwszy sprzeciw.
Dwóch uczestników, A.Newell i H.Simon uważali, że bardziej precyzyjne będzie określenie "systemy złożonego przetwarzania informacji" (complex information processing).
Wygrała “sztuczna inteligencja”
Dwóch uczestników, A.Newell i H.Simon uważali, że bardziej precyzyjne będzie określenie "systemy złożonego przetwarzania informacji" (complex information processing).
Wygrała “sztuczna inteligencja”
Here are seven of the best #AI podcasts you can find online, many of which have recently covered #ChatGPT and #LLMs!
A thread 🧵 ↓
A thread 🧵 ↓
#1 `Tech Won’t Save Us` (@techwontsaveus)
A podcast hosted by Paris Marx (@parismarx) that explores the intersection of technology, capitalism, and the environment.
🎙️ techwontsave.us
A podcast hosted by Paris Marx (@parismarx) that explores the intersection of technology, capitalism, and the environment.
🎙️ techwontsave.us
#2 `The Data Exchange Podcast`
A podcast series that explores the latest trends, techniques, and tools in the world of data science, machine learning, and AI.
The podcast is hosted by Ben Lorica (@bigdata), Chief Data Scientist at O'Reilly Media.
🎙️ thedataexchange.media
A podcast series that explores the latest trends, techniques, and tools in the world of data science, machine learning, and AI.
The podcast is hosted by Ben Lorica (@bigdata), Chief Data Scientist at O'Reilly Media.
🎙️ thedataexchange.media
A recent paper titled "Sparks of Artificial General Intelligence: Early experiments with GPT-4", published by Microsoft Research makes for very interesting reading. I have gone through the entire report and it IS TL;DR. Here is my summary of the same #GPT4 #AGI #LLMs 1/n
But before that, let's take a brief detour into what intelligence actually means. One of the definitions is that intelligence is a multifaceted and complex cognitive ability that involves the capacity to understand, learn, reason, solve problems, adapt to new situations, 2/n
think critically, and apply knowledge to different domains. But what does it mean to say that an artificial intelligence system is intelligent?
From the paper:
3/n
From the paper:
3/n
So @TheEconomist tells me now that #LLMs can do planning and reasoning after all. Obviously our own dismal experience of their planning performance (c.f. the 🧵 at ) must be a clear outlier.. 🙄 Thank goodness I pay big bucks for my subscription.. 1/
Interestingly, I was just telling someone today how several of the papers on "LLMs for Task Planning by Prompting" are rife with the Clever Hans effect (c.f. en.wikipedia.org/wiki/Clever_Ha… ). I guess I will have to do a thread.. 2/
(While we should all be used to #LLM hype-expertise in the press by now, this particular case was prickly as it is my cocky son who airily pointed this article out to me at dinner with barely concealed delight.. 😡 ) 3/
I used @predictionguard (paired with @LangChainAI)to evaluate 27 Large Language Models (LLMs) for text generation and automatically select a best/fallback model for use in some generative #AI applications.
A thread 🧵
A thread 🧵
Some general thoughts:
-> No surprise in the top performer (@OpenAI GPT-3.5), but...
-> 3 of @CohereAI models show up, including the runner up.
-> Surprised to see XLNet, a model that I can pull down from @huggingface, performs better than some of the popular #LLMs
-> No surprise in the top performer (@OpenAI GPT-3.5), but...
-> 3 of @CohereAI models show up, including the runner up.
-> Surprised to see XLNet, a model that I can pull down from @huggingface, performs better than some of the popular #LLMs
Regarding the evaluation, the process works as follows:
1. You upload some examples of the model input/ output behavior that you expect.
2. @predictionguard concurrently runs these examples through a bunch of SOTA models on the backend
1. You upload some examples of the model input/ output behavior that you expect.
2. @predictionguard concurrently runs these examples through a bunch of SOTA models on the backend
Slides to to engage clinical + informatics communities across multiple forums.
For the two people who have not already heard about and tried using #ChatGPT, over 100 million other people already have. The fastest growing internet application in history. #MedAI
For the two people who have not already heard about and tried using #ChatGPT, over 100 million other people already have. The fastest growing internet application in history. #MedAI
Good or bad, ready or not, these tools are out there and are being used in all imaginable and some unimaginable ways.
Even LAST quarter, Stanford students are using ChatGPT on the class assignments, including straight up submitting ChatGPT generated answers without any edits.
Even LAST quarter, Stanford students are using ChatGPT on the class assignments, including straight up submitting ChatGPT generated answers without any edits.
Overly simplified breakdown on how these systems work. Auto-complete on steroids. How to guess the next word you enter? Learning parameters on how often those words have appeared together in prior examples.
⚡️Breaking news from the EU & #GDPR land: the Italian Data Protection Authority #GarantePrivacy issued an order today against OpenAI, effectively blocking #ChatGPT in Italy (ordering it not to use personal data of Italians). Here is a deep dive into the short order - 1/
What did the Italian DPA #GarantePrivacy found problematic with ChatGPT4?
- Lack of transparency
- Absence of a lawful ground for processing
- Not respecting accuracy
- Lack of age verification
- Overall breach of Data Protection by Design
Why for each? A short explainer:
2/
- Lack of transparency
- Absence of a lawful ground for processing
- Not respecting accuracy
- Lack of age verification
- Overall breach of Data Protection by Design
Why for each? A short explainer:
2/
A) Lack of transparency
Pursuant verification, the #GarantePrivacy found that no information is provided to users of the service, nor generally to people whose data have been collected by OpenAI & processed through the ChatGPT service.
=> a breach of Art. 13 #GDPR
3/
Pursuant verification, the #GarantePrivacy found that no information is provided to users of the service, nor generally to people whose data have been collected by OpenAI & processed through the ChatGPT service.
=> a breach of Art. 13 #GDPR
3/
#ChatGPT #AI performance on *open-ended, free-response,* clinical reasoning exams intended for human medical trainees. #LLMs are passing medical licensing exams, but artificial multiple choice designs do not reflect realistic clinical reasoning.
medrxiv.org/content/10.110…
medrxiv.org/content/10.110…
From describing diagnostic schema, generating differential diagnoses and problem lists, to suggesting and interpreting tests, ChatGPT is already demonstrating the surprising ability to often reach a 70% passing threshold on multiple cases.
For simple recall style questions ("describe the typical symptoms of a patient with heart failure,") it will knock those questions out of the park, while showing wide variability and struggling with deeper analytical questions.
INTRODUCTION
The pursuit of knowledge and understanding has been a driving force in human history, shaping cultures, societies, and civilizations. Throughout the ages, people have sought to uncover the mysteries of the universe and their place within it.
🧵 "#Possibility Architectures: Exploring Human #Communication with Generative #AI"
Today's SFI Seminar with ExFac Simon DeDeo @LaboratoryMinds (@CarnegieMellon), streaming now:
Today's SFI Seminar with ExFac Simon DeDeo @LaboratoryMinds (@CarnegieMellon), streaming now:
"A key feature of this is talk is that we make sense of what each other are saying IN PART by what they say, but ALSO by what we expect of them."
"Language transmits info against a background of expectations – syntactic, semantic, and this larger cultural spectrum. It's not just the choices of make but [how] we set ourselves up to make later choices."
@LaboratoryMinds re: work led by @clairebergey:
@LaboratoryMinds re: work led by @clairebergey:
#LLMs are powerful, but can they make existing GUIs interactable with language? Last summer at @GoogleAI, we found that LLMs can perform diverse language-based mobile UI tasks using few-shot prompting. Exciting implications for future interaction design! #chi2023 Thread 🧵
🧠 Key Takeaway: Using LLMs, designers/researchers can quickly implement and test *various* language-based UI interactions. In contrast, traditional ML pipelines require expensive data collection and model training for a *single* interaction capability.
Learn more about it👇
Learn more about it👇
To adapt LLMs to mobile UIs, we designed prompting techniques and an algorithm to convert the view hierarchy data in Android to the HTML syntax, which is well-represented in LLMs’ training data.
Wanna build the next ChatGPT?
Here's a 6 month learning plan for AI large language models (LLMs) from scratch.
(thread)
Here's a 6 month learning plan for AI large language models (LLMs) from scratch.
(thread)
1. Familiarize yourself with basic machine learning concepts and Python programming.
Books:
1.1. Introduction to Machine Learning with Python by Andreas Müller and Sarah Guido
1.2. Python Machine Learning by @rasbt
Books:
1.1. Introduction to Machine Learning with Python by Andreas Müller and Sarah Guido
1.2. Python Machine Learning by @rasbt
2. Learn about natural language processing (NLP) and its applications.
Books:
2.1. Speech and Language Processing by Daniel Jurafsky and James H. Martin
2.2. Natural Language Processing with Python by Steven Bird, Ewan Klein, and Edward Loper
Books:
2.1. Speech and Language Processing by Daniel Jurafsky and James H. Martin
2.2. Natural Language Processing with Python by Steven Bird, Ewan Klein, and Edward Loper
40 Deep #SEO Insights for 2023:
-In 2022, I told to focus on Natural Language Generation, and it happened.
-In 2023, F-O-C-U-S on "Information Density, Richness, and Unique Added Value" with Microsemantics.
I call the collection of these, "Information Responsiveness".
1/40 🧵.
-In 2022, I told to focus on Natural Language Generation, and it happened.
-In 2023, F-O-C-U-S on "Information Density, Richness, and Unique Added Value" with Microsemantics.
I call the collection of these, "Information Responsiveness".
1/40 🧵.
1. PageRank Increases its Prominence for Weighting Sources
Reason: #AI and automation will bloat the web, and the real authority signals will come from PageRank, and Exogenous Factors.
The expert-like AI content and real expertise are differentiated with historical consistency.
Reason: #AI and automation will bloat the web, and the real authority signals will come from PageRank, and Exogenous Factors.
The expert-like AI content and real expertise are differentiated with historical consistency.
2. Indexing and relevance thresholds will increase.
Reason: A bloated web creates the need for unique value to be added to the web with real-world expertise and organizational signals. The knowledge domain terms, or #PageRank, will be important in the future of a web source.
Reason: A bloated web creates the need for unique value to be added to the web with real-world expertise and organizational signals. The knowledge domain terms, or #PageRank, will be important in the future of a web source.
1/ #ChatGPT is closing out 2022 with a bang, but what’s next? 💥
@OpenAI’s #GPT4 is set to be the first big #AI thing in 2023.
So here are some bold, optimistic, yet sensible predictions from me, @vivek7ue and @rajhans_samdani ... 👀
@OpenAI’s #GPT4 is set to be the first big #AI thing in 2023.
So here are some bold, optimistic, yet sensible predictions from me, @vivek7ue and @rajhans_samdani ... 👀
2/ Biggest model size for GPT-4 will be 1T parameters. Up 6x.
Not 100T parameters like some AI hypers are claiming ().2/ Biggest model size for GPT-4 will be 1T parameters. Up 6x
Not 100T parameters like some AI hypers are claiming ().2/ Biggest model size for GPT-4 will be 1T parameters. Up 6x
3/ The reason is simple: instruction fine tuning achieves same quality with 100x smaller models.
arxiv.org/pdf/2203.02155…
arxiv.org/pdf/2203.02155…
