Every model is a dry run for the next one. Scaling takes ambition and patience.
microsoft.ai/news/frontier-…
microsoft.ai/news/frontier-…
1
7
94
5,912
We also hope being transparent about the details & learnings would be beneficial to the community.
0
0
11
961
Note we do not use distillation from 3rd-party models, and the goal here is to validate the workflow before scaling up.
0
2
10
951
Hanxiao Liu retweeted
Pi just got a huge upgrade! It’s now powered by our latest LLM: Inflection-2.5, which is neck and neck with GPT-4 on all benchmarks and used less than half the compute to train.
Pi now has world class IQ, combined with its distinctively kind and curious personality.
Give it a go at pi.ai
inflection.ai/inflection-2-5
Pi now has world class IQ, combined with its distinctively kind and curious personality.
Give it a go at pi.ai
inflection.ai/inflection-2-5
81
236
951
685,221
Hanxiao Liu retweeted
Evaluation is everything! While testing Inflection-2.5, we found that MT-Bench has a bunch of incorrect answers.
Here we share the corrections for everyone to use, and we release a new Physics GRE benchmark for people to try out.
inflection.ai/inflection-2-5
Here we share the corrections for everyone to use, and we release a new Physics GRE benchmark for people to try out.
inflection.ai/inflection-2-5
14
118
331
105,958
TL;DR: next stop would be GPT-4 and stay tuned.
🎉 Introducing Inflection-2, the 2nd best LLM in the world!
Get ready to experience the future of AI with us.
bit.ly/3TaUpcD
5
2
53
19,477
RT @mustafasuleyman: Utterly insane weekend. So sad. Wishing everyone involved the very best.
In the meantime, we finished training Infle…
In the meantime, we finished training Infle…
0
183
0
It has been incredibly fun to work at @inflectionAI despite a small team. Join us!
Excited to announce that we’ve raised $1.3B to build one of the largest clusters in the world and turbocharge the creation of Pi, your personal AI.
forbes.com/sites/alexkonr…
3
5
66
22,015
Hanxiao Liu retweeted
It’s a big week! We’ve raised $1.3 billion and are building the world’s largest AI cluster (22k H100s).
We’re grateful for our investors and new funding that will help us accelerate our mission to make personal AI available to every person in the world. inflection.ai/inflection-ai-…
We’re grateful for our investors and new funding that will help us accelerate our mission to make personal AI available to every person in the world. inflection.ai/inflection-ai-…
67
211
1,462
2,768,523
Check out our tech report (shorturl.at/ktT67) for more details. This is just a starting point so stay tuned!
We have amazing results to announce! Inflection-1 is our new best-in-class LLM powering Pi, outperforming GPT-3.5, Llama and PALM-540B on major benchmarks commonly used for comparing LLMs. inflection.ai/inflection-1
1
1
20
7,590
Hanxiao Liu retweeted
Today we are announcing the release of the first version of @heypi_ai, your personal AI.
Pi stands for “personal intelligence” because it provides infinite knowledge based on your unique needs.
Talk with Pi today at heypi.com or by texting +1 (314) 333-1111.
Pi stands for “personal intelligence” because it provides infinite knowledge based on your unique needs.
Talk with Pi today at heypi.com or by texting +1 (314) 333-1111.
121
153
650
305,031
Hanxiao Liu retweeted
New @GoogleAI paper: How do language models do in-context learning? arxiv.org/abs/2303.03846
Large language models (GPT-3.5, PaLM) can follow in-context exemplars, even if the labels are flipped or semantically unrelated. This ability wasn’t present in small language models.
1/
Large language models (GPT-3.5, PaLM) can follow in-context exemplars, even if the labels are flipped or semantically unrelated. This ability wasn’t present in small language models.
1/
12
196
915
321,814
Hanxiao Liu retweeted
At #AutoML_Conf, we care about #reproducibility. Today, we're introducing the novel concept of reproducibility reviewers, who assess to which degree results can be reproduced with the shared code. Full story: link.medium.com/1sqoRuiqInb
Sign up as a reviewer: medium.com/@fh_91350/intr…
Sign up as a reviewer: medium.com/@fh_91350/intr…
1
21
61
Check out our new work on simple & effective Transformer modifications found by evolution!
Primer: Searching for Efficient Transformers for Language Modeling
arxiv.org/abs/2109.08668
We use evolution to design a new Transformer variant, called Primer. Primer has a better scaling law, and is 3X to 4X faster for training than Transformer for language modeling.
0
1
3
Hanxiao Liu retweeted
Wish your neural networks faster and more accurate?
Check out our recent EfficientNetV2 and CoAtNet, which significantly speed up the training and inference, while achieving state-of-the-art 90.88% top-1 accuracy on ImageNet.
Check out our recent EfficientNetV2 and CoAtNet, which significantly speed up the training and inference, while achieving state-of-the-art 90.88% top-1 accuracy on ImageNet.
As #NeuralNetwork models and training data size grow, training efficiency has become more important. Today we present two families of models for image recognition that train faster and achieve state-of-the-art performance. Learn more and grab the code ↓ goo.gle/3kfIsAk
9
72
253
Hanxiao Liu retweeted
Happy to introduce CoAtNet: combining convolution and self-attention in a principled way to obtain better capacity and better generalization.
88.56% top-1 with ImageNet21K (13M imgs), matching ViT-huge with JFT (300M imgs).
Paper: arxiv.org/abs/2106.04803
88.56% top-1 with ImageNet21K (13M imgs), matching ViT-huge with JFT (300M imgs).
Paper: arxiv.org/abs/2106.04803
6
105
341
Hanxiao Liu retweeted
Today on keras.io: three MLP models for image classification, demonstrated in less than 300 lines of code in total. MLP-Mixer, FNet, and gMLP.
keras.io/examples/visio…
keras.io/examples/visio…
2
59
267