Today on the podcast, explores the capabilities of ’s #ChatGPT with none other than, well, ChatGPT! We explore LLMs, translation systems, “jailbreaking” ChatGPT, stable diffusion, and even attempt to discuss LamDA and sentience!
🎧twimlai.com/go/603
Amr Sharaf’s Tweets
Come join us this summer!! We are looking for exceptional interns to work on cutting edge research in various Machine Translation areas, including large-scale multilingual models, customization, adaptation, robustness and efficient training and runtime.
Super excited to announce that T-ULRv6 XXL, based on “XY-LENT”, has achieved 1st position on both the Google XTREME and GLUE leaderboards.
Blog: lnkd.in/dAWaPDAb
Paper: lnkd.in/dxqhnDH8
1
4
26
Show this thread
I'm excited to share that the Turing Universal Language Representation Model (T-ULRv6) is the first multilingual language understanding model to achieve 1st position on both the Google XTREME and GLUE leaderboards. Congrats to our Turing team on this accomplishment!
4
19
Please spread the word: MIT DeCoDE lab (decode.mit.edu) has an available Postdoc position on deep generative models and geometric deep learning for engineering design.
Apply here (Job #21852): careers.peopleclick.com/careerscp/clie
8
13
Let's talk about MONEY.
Some Ph.D. students never worry about it, some Ph.D. students are worried about it all the time.
I am sharing what my monthly looks like as a CS Ph.D. student at Stanford and how I think about my financial future.
30
116
799
Show this thread
🤙 Calling #HCAI researchers!
📃 Please submit to the #HCAI @ #NeurIPS2022 workshop!
🗓️ Deadline 09/22/2022
➕ hcai-at-neurips.github.io/site/call.html
Would love to see submissions across topics like theory, experiences, design, fairness, privacy, transparency, accountability, governance, etc.
read image description
ALT
read image description
ALT
1
42
100
AWESOME👏👏👏
In 2017, a boy asked Roger Federer:
“Can you please continue to play for 8-9 years so that I can play you when I go pro?”
411
8,825
39K
If you are an AI practitioner (broadly construed) who has used/built blocklists, allowlists, rule-based systems, classifiers, or other filtering technologies to mitigate RAI issues, please consider participating in our study!!!!
Quote Tweet
We @Microsoft are conducting an interview study on practices around using filtering technologies (e.g., blocklists) to mitigate responsible AI issues in language/multimodal systems. If you have used/built filtering tech, consider taking our short screener: forms.office.com/r/60wLp4PCDw
Show this thread
4
12
Are you a data scientist, UX researcher, product manager, academic researcher, or other person involved in building machine learning apps who has at least a basic understanding of explainable AI?
If so, please consider signing up for our interview study!
Thank you!!!!
Quote Tweet
Care about Explainable #AI (#XAI) & Responsible AI?
Want to join a *PAID* interview study with experts from @MSFTResearch & @gtcomputing to explore new ways to make AI systems explainable & increase user empowerment?
Sign-up: microsoft.qualtrics.com/jfe/form/SV_5i
Please RT & share!
1/3Show this thread
5
2
It is great to see about 200 audiences attending! Come join us this afternoon at the Structured and Unstructured Knowledge Integration (SUKI) workshop #NAACL2022!
10
30
Show this thread
Two late-breaking Microsoft Research internship opportunities for this summer: one on large-scale recommender systems and another to explore storage side acceleration to enable and improve vector search at Microsoft/worldwide scale. Apply at careers.microsoft.com/us/en/job/1187
17
36
🧐 When targeting zero-shot use, should you train a T5, a PrefixLM, or a GPT? What if you plan to leverage multitask finetuning (à la T0)?
🤩 In arxiv.org/abs/2204.05832, we explore how architectures & pretraining objectives impact zero-shot performance.
⬇️ Thread time!
3
88
319
Show this thread
We have some idea what neural nets see, but what do they *hear*? We break down a speech recognition net and sonify (rather than visualize) what each neuron responds to. You can see (err...I mean hear) the process proceed from shallow to deep features.
cs.umd.edu/~amin/apps/vis
2
15
88
invites faculty applications in all areas of computing and applied math:
applications.caltech.edu/jobs/cms
At Caltech, we are particularly proud of our emphasis on fundamentals coupled with tight integration across disciplines.
We also have a lovely campus (photos by me).
1
23
96
There's no evidence that SGD plays a fundamental role in generalization. With totally deterministic full-batch gradient descent, Resnet18 still gets >95% accuracy on CIFAR10. With data augmentation, full-batch Resnet152 gets 96.76%.
arxiv.org/pdf/2109.14119
29
217
927
Show this thread
Recent Advances in Language Model Fine-tuning
New blog post that takes a closer look at fine-tuning, the most common way large pre-trained language models are used in practice.
8
265
864
Want to learn more about data poisoning and backdoor attacks? Our survey paper (arxiv.org/abs/2012.10544) clarifies the state of the field for newbies and veterans alike!
14
62
MiniTorch v0.2 (minitorch.github.io). A code-it-yourself textbook for minimal autodiff, tensors, gpu's, and nn's.
Battletested through a semester with 5 autotested assignments.
(Aiming for a v1 by early January. DM for teacher's key)
5
90
487
Thanks everyone for attending our #NeurIPS2020 tutorial! :)
Slides: docs.google.com/presentation/d
Video: slideslive.com/38935801/pract
and I will continue answering questions on rocketchat and we also have a follow-up zoom Q/A session on Wednesday.
Quote Tweet
Tomorrow @latentjasper @balajiln and I present a #NeurIPS2020 tutorial on "Practical Uncertainty Estimation and Out-of-Distribution Robustness in Deep Learning". Whether you're new to the area or an expert, there is critically useful info! 8-10:30a PT nips.cc/virtual/2020/t
Show this thread
10
84
415
To help better understand the theoretical foundations of batch offline RL, M.Wang, S. Meyn & I are organizing a Simons workshop this week with wonderful speakers!
Schedule: simons.berkeley.edu/workshops/sche
Webinar link:
1
31
169
I succumbed to threats and wrote my 2019–20 faculty report. Hot papers last year: Electra: Pre-training text encoders as discriminators openreview.net/pdf?id=r1xMH1B, Stanza: A Python toolkit for many languages aclweb.org/anthology/2020 & Universal Dependencies v2 aclweb.org/anthology/2020
1
37
195
Trivia friends: please share to folks interested in combining trivia with PhD studies. I'm looking for a good writer interested in looking at the symbiosis between trivia enthusiasts and the machine learning community (hopefully contributing to both):
users.umiacs.umd.edu/~jbg/static/op
7
9
Many open Reinforcement Learning related positions are here: aka.ms/rl_hiring . Please consider applying if you have RL-complimentary expertise and want to discover and build the future.
3
44
158
Recognize any of these people? From Facebook? Twitter?
These images are not real — they’re from the mind of a computer, and they’re infiltrating the internet.
We set up our own AI system to understand how this technology works.
1,204
10.9K
14.9K
Show this thread
Several internship positions available this summer at MSR in NLP and related areas. careers.microsoft.com/us/en/job/9279
19
54
Training on test points is a no-no. But training on test points with fake labels turns out to be a good idea. Aurick's blog post on amortized conditional normalized likelihood (ACNML) shows how NML provides a principled view of "training on test points": bair.berkeley.edu/blog/2020/11/1
3
27
123
Show this thread
As EMNLP 2020 #emnlp2020 is close to the end, it's time to review the progress in #NLProc powered by #KnowledgeGraph ⚡️ ! About 30 papers on language models, KG representation learning, NLG, extraction, ConvAI, datasets, and more!
46
123
These tutorial slides on "High Perf NLP" are really impressive. Every slide is current to the minute. Amazing set of diagrams.
gabrielilharco.com/publications/E
( @IuliaTurc Felipe Ferreira Cesar Ilharco)
9
489
1,695
Show this thread
I've written out a collection of tips for crafting your faculty application package (mainly for computer science). Distilled from advice I've been giving during the current application cycle. Hope others find it useful, and comments are welcome.
1
63
312
The **Algorithms** group Redmond is hiring! It is the sister group to our ML Foundations group, and has already strong presence in differential privacy and coding!
Post Doc: careers.microsoft.com/us/en/job/9332
FTE: careers.microsoft.com/us/en/job/9332
25
99
Excited to share our #TACL paper with : “EDITOR: an Edit-Based Transformer with Repositioning for Neural Machine Translation with Soft Lexical Constraints”
EDITOR generates text flexibly by seamlessly allowing users to specify lexical preferences.
Link below👇 (1/6)
1
13
31
Show this thread
#EMNLP2020 paper with : Even high-similarity wikimatrix parallel segments can diverge semantically. We predict these divergences with mBERT + synthetic supervision, and annotate them reliably via rationales. Today at Gather Session 1C [UTC-05:00]
Quote Tweet
New #emnlp2020 paper w/ @MarineCarpuat at @umdclip on "Detecting Fine-grained Cross-lingual Semantic Divergences without supervision by Learning to Rank" is now on arxiv: arxiv.org/abs/2010.03662
Code and data available: github.com/Elbria/xling-S
Show this thread
1
6
14
Show this thread
Vowpal Wabbit 8.9 is out: vowpalwabbit.org/blog/vowpalwab
Many people are working on VW now, so in addition to new stuff (exploration algorithms, continuous actions, probabilistic label tree, ...), there is much better python support and easy ways to install. Enjoy.
1
18
141
Show this thread
"Domain generalization" studies predictions on domains (distributions) that are different from those used in training (also known as out-of-distribution generalization). Here is my lecture on this topic:
1
5
51
Alekh Agarwal, Akshay Krishnamurthy, and I finished a major tutorial on "Theoretical Foundations of Reinforcement Learning" (hunch.net/~tforl/) for FOCS (focs2020.cs.duke.edu/program/), but potentially of much broader interest. We'll be doing Q&A Friday.
1
95
378
My lectures on "Domain Adaptation":
Part 1: youtu.be/El760ZzsXN8
Part 2: youtu.be/wwgt_ErD3vA
Pardon the noise in the middle of the second lecture. Neighbor decided to dig a hole in their floor, aka my ceiling!
1
14
71
Interested in advanced NLP&CV&ML research? I am looking for multiple motivated PhD students to join my lab @UCSC_BSOE in Fall 2021! Particularly, we strive to build embodied AI agents that communicate with humans to do their jobs. Please help RT!
eric-xw.github.io/hiring.html
7
78
213
Show this thread