object Object

[Carter Leffen @carterleffen]

as of 40 minutes ago, the @MistralAI team amended their Outputs ToS. thank you for listening and your leadership on this matter!

NEW -Output ownership.
Where applicable, We assign to You, all the intellectual property rights in and to the Outputs, including (i) the right to reproduce, without any limitation of number, all or part of the Outputs (ii) the right to represent, broadcast or have broadcasted, to communicate or make available the Outputs to the public (iii) the right to adapt, correct, modify, improve the Outputs, the right to make new versions of the Outputs from all or part of the existing Outputs, the right to maintain, modify, arrange, assemble, condense, transcribe, scan, mix, migrate, zip, unzip all or part of the Outputs, the right to translate them into any language whatsoever (including computer language), the right to interface them with any software, material or database (unless otherwise stated), the right to integrate them in any existing or future works, on any support and by any means, (iv) the right to define and to modify the use, the purpose of the use and the destination of the Outputs in any form, known and unknown, existing or future, and (v) more generally, the right to use, exploit, distribute the Output. You are allowed to exercise these rights on any existing or future, known or unknown (i) medium including but not limited to any printed, electronic, digital, numerical, analog, magnetic media, (ii) mean of communication, telecommunication, or other network including but not limited to radio, video, broadcast, satellite, Internet, electronic. Unless otherwise stated, these rights can be exercised for any purpose (public exploitation, private use, free, paid for, subscription, etc.). This assignment is granted worldwide and for the entire legal term of protection of the Outputs by the intellectual property rights applicable as provided for by the applicable law. However, You are expressly prohibited to use the Outputs and/or any modified or derived version of the Outputs to (directly or indirectly) to reverse engineer the Services

[object Object @SteinsZeit]

By far the best, most thorough study on this topic I’ve seen so far.

There are many LLM inference solutions out there and especially beginners might feel overwhelmed by the different choices.

[Hamel Husain @HamelHusain]

Here is how you can obtain a massive speedup with llama-v2 models, much faster than anything else I tried.

It's so fast that its unreal. I made some additional notes on how to avoid a temporary foot gun as well

https://t.co/HXIY5e0h2Y

[Tom Goldstein @tomgoldsteincs]

#OpenAI is planning to stop #ChatGPT users from making social media bots and cheating on homework by "watermarking" outputs. How well could this really work? Here's just 23 words from a 1.3B parameter watermarked LLM. We detected it with 99.999999999994% confidence. Here's how 🧵

[Preetum Nakkiran @PreetumNakkiran]

every 3 years my friend and I need to re-learn kernels/RKHS for some theory project, and we rediscover these excellent 700 slides ("CS-friendly" but rigorous): https://t.co/OtqxMDvVNC

[Tom Goldstein @tomgoldsteincs]

Diffusion models like #DALLE and #StableDiffusion are state of the art for image generation, yet our understanding of them is in its infancy. This thread introduces the basics of how diffusion models work, how we understand them, and why I think this understanding is broken.🧵

[@[email protected] @Dennis_Kipker]

Ganz toll:

Some top 100,000 websites collect everything you type—before you hit submit
A number of websites include keyloggers that covertly snag your keyboard inputs.
https://t.co/n5R6jMChgy

[Royal Statistical Society @RoyalStatSoc]

Today we release the findings of our survey of MPs, testing their statistical skills (a thread)

https://t.co/rcyb6gc7F0

[Michael Black @Michael_J_Black]

I get a lot of reviews that say my work is not novel and I bet I'm not alone. It's always frustrating because I see novelty where the reviewer doesn't. Rather than rebut every critique, I've written a blog post to help reviewers think about novelty. https://t.co/UXLabOkYcn

[Shannon Pileggi... @[email protected] @PipingHotData]

👋why hello .drop = FALSE
🥺where have you been my whole life?
🤗welcome to the family
#RStats

[Frank Nielsen @FrnkNlsn]

Johnson-Lindenstrauss lemma: embed any n-point set of dim d into a k=O((1/ε )**2 log n) dim space with 1+ε distortion. Linear transform matrix A can be obtained as a random projection matrix.
→ Allows to perform k-means in very high dimensions!
https://t.co/rj13F82cHy

[Moxie Marlinspike @moxie]

Wrote some notes summarizing my first impressions of web3: https://t.co/16MoPROOGc

[Women in Statistics and Data Science @WomenInStat]

You know how excited @daniela_witten gets about SVD? I have about the same thing with kernels. Except that I'm not sure I explain them as well as she does SVD. Still, you're getting a thread on kernels!

[Bernoulli Society @BernoulliSoc]

2022 Wolfgang Doeblin @BernoulliSoc Prize is awarded for outstanding research in the field of probability theory. Nominations for the 2022 edition of the award can be submitted by 30 December 2021.

[ecoprof @ecotheory]

Excited to release our free e-book: Building Skills in Quantitative Biology https://t.co/SE7WvqUmpw. Quick intros to GitHub, R Markdown, multivariate stats, machine learning & optimization. For grad students & professional biologists. pls RT #Rstats

[Max Roser @MaxCRoser]

Very happy that we received a prize for our work on the pandemic!

Here is the video of Oxfordshire's High Sheriff @ImamMonawar awarding me that prize and I answer some questions about how we work: https://t.co/ZCpsd8zMlB

[Karsten Hahn @struppigel]

COVID-19 map for Germany looks pretty bad.
Except for Ludwigslust-Parchim.
It was hit by ransomware.

[Stanford Online @StanfordOnline]

Stanford Online and Stanford NLP are happy to share the latest lectures from Prof Chris Manning's course CS224N: Natural Language Processing with Deep Learning. Many thanks to @chrmanning and @StanfordNLP for making this content publicly available. https://t.co/Om7omc4rhq

[Fabio Viggiani @fabio_viggiani]

🚨ONGOING: we are investigating systems infected with a malicious version of the npm package UAParser.js (7 million weekly downloads).
The hijacked package delivers a malware loader and a cryptominer.
IOCs below:

[object Object @SteinsZeit]

@JackRhysider You’d also have more images than there a particles in the universe. Even if you put an entire universe into every particle of the universe, you’d have way more images than there are particles in this nested universe.

In case you’re interested in the math:
https://t.co/DKmGvBhWEq

[object Object @SteinsZeit]

Take 32x32 images for example. You know, those really bad quality blurry images in which you can hardly make out any details. Any such image has 32x32=1024 pixels.
Every pixel has three color channels with 256 possible values each. 2/

[Warwick Statistics @warwickstats]

Martine Barons, Director of the Applied Statistics & Risk Unit, together with colleagues from the AU4DM network have produced a toolkit for communicating climate risk which is set be presented at COP26.
See https://t.co/LrDjfmTII1