Should-ing yourself harms your mental health in various ways. It makes you overlook the effort you had already put in. Instead of celebrating your hard work, you start resenting it. You do it often enough and should-ing will kill your self-esteem 💔

What is your favorite matplotlib configuration setting for beautiful scientific charts? Links welcome to open source or examples of charts you love.

How to find research opportunities? Finding opportunities to gain experience is arguably the most challenging part for students wishing to pursue grad school, particularly for those who don't have resources/connections. Some tips on approaching potential mentors. 🧵

PicoGPT - A GPT in 62 lines of Python (see photo below) and Numpy code along with a tractable article on the simplified architecture Article: jaykmody.com/blog/gpt-from-… GitHub: github.com/jaymody/picoGPT

I wonder if a new conspiracy theory will emerge where people are convinced they are actually biological LLMs, trained & controlled by some shadowy govt authority.

"semantic embeddings" are becoming increasingly popular, but "semantics" is really ill-defined. sometimes you want to search for text given a description of its content. current embedders suck at this. in this work we introduce a new embedder. Shauli Ravfogel Valentina Pyatkin ➡️ ICML Avshalom Manevich

Our new paper on RL From Human Feedback is out: arxiv.org/abs/2305.18290. In Direct Preference Optimization (DPO) we reparameterize the reward model in a suitable way without any loss in generality and optimize the EXACT RLHF objective directly with a simple classification loss.

This is freaking amazing! Jia-Bin Huang has prepared a collection of twitter threads to help out students in academia ; like how to get LOR, how to apply for a PhD program, etc. I found this really really helpful! #AcademicTwitter #AcademicChatter 🔗:github.com/jbhuang0604/aw…

There are three primary ways in which language models learn. Let’s quickly go over them and how they are different from each other… 🧵 [1/7]

linkedin.com/posts/prithivi…

📢 Thank you so much for attending our tutorial! 🙌 🔗All the materials are available online Our slide: acl2023-retrieval-lm.github.io The live Q&A on sli.do: app.sli.do/event/ok8R2jMM… If you registered for ACL, you can see the recorded Zoom video on Underline.

My son has just started calculus, and I asked him what the relationship was between the gradients of the tangent and the normal to a curve at a given point. His first reply was, "They are perpendicular." I've noticed many times that something one gains with experience ... 1/7

Mark Chen Mark, if a point of view seems to you to rely on an obviously ridiculous premise, you're almost certainly misunderstanding it. When that happens, don't dismiss it, but instead try to understand more deeply.

Peyman Milanfar Here is the free version : archive.org/details/poor-m…

Q-Learning is *probably* not the secret to unlocking AGI. But, combining synthetic data generation (RLAIF, self-instruct, etc.) and data efficient reinforcement learning algorithms is likely the key to advancing the current paradigm of AI research… TL;DR: Finetuning with

Project #2: LLM Visualization So I created a web-page to visualize a small LLM, of the sort that's behind ChatGPT. Rendered in 3D, it shows all the steps to run a single token inference. (link in bio)

# On the "hallucination problem" I always struggle a bit with I'm asked about the "hallucination problem" in LLMs. Because, in some sense, hallucination is all LLMs do. They are dream machines. We direct their dreams with prompts. The prompts start the dream, and based on the

e/ia - Intelligence Amplification - Does not seek to build superintelligent God entity that replaces humans. - Builds “bicycle for the mind” tools that empower and extend the information processing capabilities of humans. - Of all humans, not a top percentile. - Faithful to

And, finally, there's no way that someone can factcheck to cite sources for the papier-mache extruded by a text synthesis machine that does not track information provenance. >>

I agree with much of both @emilymbender.bsky.social’s #ACL2024 presidential talk and (((ل()(ل() 'yoav))))👾’s rejoinder, but I want to comment on just one aspect where I disagree with both: the definition and domain of CL vs NLP. 🧵👇