From now on I will be spending much more time on LinkedIn, and at LinkedIn. Very excited about my next play!
Follow
Click to Follow xamat
Xavier Amatriain
@xamat
Xavier Amatriain’s Tweets
It's a "social" platform to interact with AI bots and share your interactions. It's also a platform to easily deploy different AI chatbots. At this moment it has two based on OpenAI technologies, and one based on Anthropic. More details in the blog post above by
2
1
Show this thread
Today announced a new product, Poe, that had been out in private beta for a while. I encourage you to try it out if you are interested in genAI/LLMs, or just for fun
3
6
24
Show this thread
I had missed this. The video of my invited talk last year at was published. Almost two hours that include many good questions from the audience youtube.com/watch?v=UO6fWj (Slides are missing from the video unfortunately, but are available here slideshare.net/xamat/dataai-d)
1
3
18
Ten years ago & me published a blog post describing an architectural blueprint for Recommender Systems. I'm now revisiting it by including several alternatives published since, and a new one that in some ways includes all the previous ones: amatriain.net/blog/RecsysArc
6
45
215
The LLM family tree, by ! 🤯👇
Check out the blog post: 🔗amatriain.net/blog/transform
#GenerativeAI #NLP #LaMda #GPT3 #ChatGPT
2
24
76
Most of the magic behind the recent generative AI models comes from the incredibly powerful transformer architecture and the deep neural networks that were built using it.
1/3
5
53
304
Show this thread
Wow, my Transformers Catalog has been viewed over 10k since last week's update. Many good comments too. Quick follow up includes link to github for folks who want to file a PR. Also adding a new timeline view where the Y-axis is model size amatriain.net/blog/transform
1
12
62
We're hiring at and would love to talk to any talented folks out there (particularly in engineering!).
Some of the most exciting and cutting edge work in the world: using AI+LLMs to make high quality medical care affordable to everyone.
curaihealth.com/careers
9
47
Transformer Catalog ( 2023 Edition)
Really neat timeline of Transformer models by . A great reference to track and find the latest notable Transformer models.
amatriain.net/blog/transform
15
200
648
Big shoutout for 's LLM family tree blog post! It comes with nice & concise summaries of each model (63 pages, if you export it as a PDF!)
amatriain.net/blog/transform
14
160
592
Pretty big update to my Transformer Catalog. I added ChatGPT, Sparrow, and Stable Diffusion among others. I also included a section about RLHF and Diffusion models and a new timeline view. Enjoy! amatriain.net/blog/transform
23
231
902
"Prompt Engineering 101: Introduction and resources"
Xavier Amatriain () shares a compilation of tips, tricks, examples, and resources related to Prompt Engineering in a blog post
6/18
Quote Tweet
I have compiled tips, tricks, examples, and resources related to Prompt Engineering in a blog post. Let me know if I missed something! amatriain.net/blog/PromptEng
2
2
4
Show this thread
I have compiled tips, tricks, examples, and resources related to Prompt Engineering in a blog post. Let me know if I missed something! amatriain.net/blog/PromptEng
1
6
26
When was the first time you knew search/Google will be disrupted?
Quote Tweet
AI is going to kill search as we know it. What's going to be fascinating is watching Google miss that boat.
Show this thread
1
5
4
13
I have mostly quit Twitter for the past few weeks, and, I have been fine.
1
1
Show this thread
There's absolutely no reason to be here. The information will get to you. And, you are just feeding the trolls by being here.
2
12
Show this thread
Just came here today to say that I am on mastodon. See you there (and LinkedIn of course!)
1
2
Ok. It has to be done. Will keep my account and will read tweets occasionally (don't want to miss preprints), but will #DeleteTwitter now. (Also glad I decided to buy an electric car other than Tesla).
Quote Tweet
Seriously considering deleting this app today
2
23
Seriously considering deleting this app today
7
2
22
+100 Immigration is the US' biggest advantage. We should 10x it, particularly when there's millions of jobs open.
Quote Tweet
This country should bring in 10x more H-1Bs and also make the pathway from H-1B to permanent residency 10x easier
3
2
14
Our new paper, Token Turing Machines is out on ArXiv! Inspired by Neural Turing Machines, we introduce an external read/write memory to discretized models like modern transformers, in a way that is end to end differentiable and improves long horizon context retention.
7
88
502
Show this thread
I wasn't aware of all of this despite growing up in Catalonia. Good to hear that this will be added to high school history curriculum.
Quote Tweet
Catalonia was a center of witchcraft trials in Europe for more than 300 years. bit.ly/3gUcQ42
1
4
Good post. A lot of interest atm in wiring up LLMs to a wider compute infrastructure via text I/O (e.g. calculator, python interpreter, google search, scratchpads, databases, ...). The LLM becomes the "cognitive engine" orchestrating resources, its thought stack trace in raw text
Quote Tweet
New post just dropped: The future of AI is action-driven jmcdonnell.substack.com/p/the-near-fut
Show this thread
31
95
738
Show this thread
We should replace the term AGI with AHI (Artificial Human Intelligence). Human intelligence is not general.
3
4
28
A Large Language Model trained on scientific papers.
Type a text and galactica.ai will generate a paper with relevant references, formulas, and everything.
Amazing work by /
Quote Tweet
Introducing Galactica. A large language model for science.
Can summarize academic literature, solve math problems, generate Wiki articles, write scientific code, annotate molecules and proteins, and more.
Explore and get weights: galactica.orgShow this thread
0:43
439.9K views
94
587
2,436
Show this thread
Is it the number of examples that matters or the number of presentations to the model during training? E.g. humans used spaced repetition to memorize facts but there are no equivalents of similar techniques in LLMs where the typical training regime is uniform random.
18
75
716
Show this thread
Please take a look at his project, and encourage him to continue in any way you can 🙏 (5/5)
2
Show this thread
I introduced him to Github and encouraged to share his project. I also learned how hard it can be for disabled people to get around things like captchas and MFA. Again, very humbling and rewarding experience. (4/5)
1
1
Show this thread
A few weeks back he reached out to me. Why out of all people he would reach out to his university professor of 20+ years beats me and humbles me. After years of falling to a dark place (in all senses), he had decided to go back to coding and needed a bit of help (3/5)
1
Show this thread
He was an undergrad student 20 years ago+ when I was teaching software engineering at in Barcelona. His disease quickly worsened over the last few years of his undergrad to the point that he became completely blind. He quit coding. I did not hear from him until... (2/5)
1
Show this thread
Friends/devs, I need you to welcome/show some love (aka github stars) to a blind coder I have been helping lately (github.com/ismaelmosquera). Keep in mind this is his first coding project in over 20 years, so, be gentle 🤗. A bit more about Ismael... (1/5)
1
4
30
Show this thread
Important to understand that many influencers don't have an opinion, they're just paid to have one.
Quote Tweet
2. Go after the YouTubers pressuring consumers to sign up.
FTX was paying YouTubers $50k+/month.
We are working on gathering people who signed up using referral links for a YouTuber class action.
Alernatively, sue the influencer that referred you to FTX in small claims court.
Show this thread
2
7
Things that matter in life
Quote Tweet
Robert Lewandowski won the 'best beer pouring' challenge that Barça players played yesterday at Estrella Damm's factory.
— @mundodeportivo
7
InstructGPT fans who missed a good Open Source alternative, Here's BLOOMZ for you!
📀 youtube.com/watch?v=Jjn39K
🤗huggingface.co/bigscience/blo.
PS: 👀 FLAN is looking at me!
4
22
91
As language models increase in scale they sometimes exhibit various useful emergent capabilities. In today’s post, we explore this phenomenon to better understand its dependence on model properties and its potential impact in applications. Read more at goo.gle/3hvJMAb
read image description
ALT
5
109
343
One of the less talked about issues AFAIK is that this will likely also kill SEO and SCM as we know it! How are new search approaches going to monetize? 🤔 Lots of innovation to come in the space!
3
Show this thread