Christopher Marshall

Data Scientist - Machine Learning 🐬 | NYC 🇺🇸

[Christopher Marshall @marshallchris]

@DearGozde How is it possible to get this wrong?

[Christopher Marshall @marshallchris]

@elonmusk So you want to fund billionaires vanity projects and defund civil liberties? Nice populist take.

[Yann LeCun @ylecun]

🥁 Llama3 is out 🥁
8B and 70B models available today.
8k context length.
Trained with 15 trillion tokens on a custom-built 24k GPU cluster.
Great performance on various benchmarks, with Llam3-8B doing better than Llama2-70B in some cases.
More versions are coming over the next few months.

https://t.co/EkU9aIHdZE

[Christopher Marshall @marshallchris]

@ylecun You may be right, but aren't you biased to side with your employer?

[Google DeepMind @GoogleDeepMind]

Introducing Gemini 1.5: our next-generation model with dramatically enhanced performance. It also achieves a breakthrough in long-context understanding.

The first release is 1.5 Pro, capable of processing up to 1 million tokens of information. 🧵 https://t.co/qT0aXdFL0n

[Christopher Marshall @marshallchris]

@gdb Interesting that the president of OpenAI still can spend all day coding. It’s important that staff in all levels of software and data science leadership rotate in and out of development or research work. Leaders too removed from hands on work become out of touch and ineffective.

[Christopher Marshall @marshallchris]

#NewProfilePic

[Christopher Marshall @marshallchris]

https://t.co/6uRgwACtuX

[Christopher Marshall @marshallchris]

Can LLMs discover new knowledge? Google Deepmind seem to be takin a step towards that

[DeepAI @DeepAI]

🤩 Lowkey Goated When #Classifier Is The Vibe! 🔥 Check out this amazing paper by Wenyang Liu et al. including @wangyi8848 to learn more about CNN for File Fragment Classification Using Bit Shift and n-Gram Embeddings https://t.co/K1lehrZZni

[Christopher Marshall @marshallchris]

via @NYTimes https://t.co/MAvb9zFz7K

[Mira Murati @miramurati]

Unveiling GPT-4 -- our large multimodal model that exhibits human-level performance on various professional and academic benchmarks. With iterative alignment and adversarial testing, it's our best-ever model on factuality, steerability, and safety.
https://t.co/rjsIYWTN3Y

[DeepAI @DeepAI]

Lowkey Goated When Counting Objects Without Labels is the Vibe! Check out this groundbreaking paper by @nguyentienvu, Jingyi Xu et al. #AI #ObjectDetection #ZeroShotLearning 🤯🤩 Link: https://t.co/Y1J0IHOL9p

[Christopher Marshall @marshallchris]

@get_back_jo @ericweinstein How is it a public, published record of user sentiment when his followers are over 70% bots?

[Christopher Marshall @marshallchris]

@ericweinstein You know he bought Twitter for the political power it gives him. Next step was to get rid of anyone with a backbone, who may oppose any kind of morally dubious manipulation of the public. Then control of the narrative can begin. Why assume the poll was even real..?

[Christopher Marshall @marshallchris]

@anothercohen I remember Andreessen/a16z getting all their AI investments to pivot to web3 about a year ago. Wonder how that panned out for them..

[Hao Liu @haoliuhl]

Can language model pretraining be even better?

Our paper shows that by randomly masking input tokens during pretraining, the zero-shot, few-shot, and fine-tuning performance can be significantly improved.

https://t.co/Jw2gTMLKPK

🧵

[Khoa Vu @KhoaVuUmn]

Have you ever wondered whether higher education can be an important driver of growth in developing countries?

My job market paper tries to answer this question by looking at a national expansion of higher ed in Vietnam.

Over 100 new universities were opened during 2006-2013!

[Micah Goldblum @micahgoldblum]

How much data are augmentations worth? We show that augmentations can actually be worth more than extra data and invariance! They increase variance across batches, and this extra stochasticity finds flatter minima. https://t.co/wn1pL7RNbY 1/8

[Christopher Marshall @marshallchris]

Some nice features in UL2. Public checkpoints, good for both fine-tuning & few-shot learning. Also works well with chain-of-thought reasoning.

[Google AI @GoogleAI]

Introducing UL2, a novel language pre-training paradigm that improves performance of language models across datasets and setups by using a mixture of training objectives, each with different configurations. Read more and grab model checkpoints at https://t.co/A7ZAFNMCY6