Thi Vũ
2 followers · 5 following · 870 views
on the atlas · 51
- alan turing - computing machinery and intelligence5 savers
- Too much efficiency makes everything worse: overfitting and the strong version of Goodhart’s law | Jascha’s blog9 savers
- Generating music in the waveform domain – Sander Dieleman2 savers
- A Survival Guide to a PhD28 savers
- How to Email a Professor and Actually Get a Response About PhD Programs1 savers
- Philip Guo - A Five-Minute Guide to Ph.D. Program Applications1 savers
- How Might We Learn?9 savers
- Joy & Curiosity #56 - by Thorsten Ball - Register Spill1 savers
- Thank you for being annoying - by Adam Mastroianni8 savers
- notes on writing well1 savers
- An Opinionated Guide to ML Research59 savers
- Generating audio for video - Google DeepMind1 savers
- Palantir: The all-seeing tech giant | The Week1 savers
- Karp - Google Docs1 savers
- TF Representations and Masking — Open-Source Tools & Data for Music Source Separation1 savers
- Hire people who give a shit. - Rational in the Fullness of Time7 savers
- Background — Connecting Music Audio and Natural Language1 savers
- How to Choose Your Grad School — Tim Dettmers4 savers
- Sequence Modeling with CTC3 savers
- Recurrent Neural Networks Tutorial, Part 1 – Introduction to RNNs · Denny's Blog2 savers
- A Visual Guide to Mamba and State Space Models3 savers
- 'Quái kiệt' Lê Viết Quốc: Làm AI cần sáng tạo, đừng vội vã chạy theo - Báo VnExpress Công nghệ1 savers
- Systematic Learning Sucks: Why I Don't Use Anki - wrycode.com1 savers
- Lessons from Andy Matuschak1 savers
- Understanding Moments2 savers
- Features | The Mnemosyne Project1 savers
- You And Your Research · Gwern.net13 savers
- Dario Amodei — On DeepSeek and Export Controls17 savers
- Late Takes on OpenAI o12 savers
- ChatGPT Is a Blurry JPEG of the Web | The New Yorker1 savers
- AI and Stress2 savers
- Theory of Change (Aaron Swartz's Raw Thought)84 savers
- Productivity - Sam Altman58 savers
- Don't Call Yourself A Programmer, And Other Career Advice | Kalzumeus Software38 savers
- How to Do Great Work25 savers
- transformer_attention.pdf23 savers
- What are Diffusion Models? | Lil'Log16 savers
- The American Scholar: Solitude and Leadership13 savers
- Diffusion is spectral autoregression – Sander Dieleman13 savers
- Spaced Repetition for Efficient Learning · Gwern.net13 savers
- The Disadvantages of an Elite Education -9 savers
- Hard Startups - Sam Altman9 savers
- Lessons from my PhD - Austin Z. Henley7 savers
- [2412.06769] Training Large Language Models to Reason in a Continuous Latent Space7 savers
- The Art of Subtle Influence - No Small Plans - Every7 savers
- Commonplace Book - https://arvid.xyz/commonplace-book/5 savers
- What You Can't Say5 savers
- How is LLaMa.cpp possible?4 savers
- Seeking the Productive Life: Some Details of My Personal Infrastructure—Stephen Wolfram Writings4 savers
- Multimodal Neurons in Artificial Neural Networks3 savers
- Last Lecture Series: How to Live your Life at Full Power — Graham Weaver - YouTube2 savers
highlights · 302
I recommend emailing at least 6-8 weeks before the deadline
How to Email a Professor and Actually Get a Response About PhD ProgramsOne or two recent papers that seem to align with the professor’s primary interest, and then, get to writing!
How to Email a Professor and Actually Get a Response About PhD ProgramsIt’s also helpful to look at and skim what they’re working on - reverse chronological sorting of Google Scholar is a good platform.
How to Email a Professor and Actually Get a Response About PhD ProgramsKeep it lightweight and professional
How to Email a Professor and Actually Get a Response About PhD ProgramsMake sure you have a super clearly-defined primary research area that shows your greatest strengths, since you will be competing directly against other applicants IN YOUR RESEARCH AREA.
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsI'd say in general for applicants in my field it's normal to apply to 10-20 Ph.D. programs. Anything over 20 feels like too much to me.)
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsDon't just apply to the few 'top-ranked' departments in your field.
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsIt's the same amount of work for your recommendation letter writers since they'll just re-send the same letter to all the departments, so you aren't creating more work for them.
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsTip #1: Apply broadly to as many departments as you are willing to attend if you get in.
Philip Guo - A Five-Minute Guide to Ph.D. Program Applicationsprofessors will be able to tell who has done their homework sincerely and who hasn't
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsAlways always always write a polite and enthusiastic follow-up email after each interview, summarizing what you discussed and expressing your personal interest in their lab and Ph.D. program (if you really are genuinely interested).
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsAlso, explore their academic home page to see what articles or blog posts they've written or podcasts they've been on, separate from their main research publications. This could give you a more well-rounded sense of what they care about.
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsIf you can't even answer questions about what you worked on, then you're toast.
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsCite only your OWN papers, and bold your name in the authors list.
Philip Guo - A Five-Minute Guide to Ph.D. Program Applicationsit's OK to cold-email professors whose research genuinely interest you – after you submit a research-dense application
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsOf the hundreds of apps I've seen, I'm not an expert in most of their research. Thus, it's crucial for you to show me why your research projects are important, what specific role you played in each one, what the main challenges were, and what the impact was – ideally a publication or steps toward one.
Philip Guo - A Five-Minute Guide to Ph.D. Program Applicationsif you've done the most significant undergrad research in sub-field X, mark down sub-field X in your app if you get to choose
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsI'm more busy now and we get more applications than ever ... so on my first pass I read only the letters and reject if they're not research-dense. Then if letters are good, I'll read Statement of Purpose. If all of that is good, then I'll more seriously consider the app.
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsHow much effort did they put into differentiating you from the hundreds of other applicants this year?
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsDon't fill up your C.V. with irrelevant details.
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsStatement of purpose is more about your high-level observations of the research field in general terms but not about your specific projects
Philip Guo - A Five-Minute Guide to Ph.D. Program Applicationsyou developed an interest in topic X rather than talking about how you've done concrete research in topic X
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsThis idea of research density also means that you need to find people who can write research-dense recommendation letters for you.
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsStrive to write the shortest and more research-dense statement that you can.
Philip Guo - A Five-Minute Guide to Ph.D. Program Applicationsin general, avoid mentioning anything before college.
Philip Guo - A Five-Minute Guide to Ph.D. Program Applicationsthe majority of your statement of purpose should be about research
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsI'll spend 3 to 5 minutes per app and place each into one of three piles: {good, ok, bad}
Philip Guo - A Five-Minute Guide to Ph.D. Program ApplicationsYou want to just dive in, and you want learning to actually work. To make that happen, we need to infuse your authentic projects with guided support, where necessary, inspired by the best ideas from cognitive psychology.
How Might We Learn?Instead, I suggest: we should take both views seriously, and find a way to synthesize the two. You really do want to make doing-the-thing the primary activity. But the realities of cognitive psychology mean that in many cases, you really do need explicit guidance, scaffolding, practice, and attention to memory support.
How Might We Learn?on’t lie to yourself
Joy & Curiosity #56 - by Thorsten Ball - Register SpillThe problem is that nobody ever tells you what it feels like to love something. Everybody thinks love feels like perpetual bliss. It doesn’t. It mainly feels annoying.
Thank you for being annoying - by Adam MastroianniIf you’re looking for the person who will never annoy you, you’ll never stop looking. But if you find someone who annoys you juuust right, you’ll never stop loving them
Thank you for being annoying - by Adam MastroianniCloseness is annoying because it’s about the surrender of control.
Thank you for being annoying - by Adam MastroianniBesides textbooks, I recommend reading PhD theses of researchers whose work interests you
An Opinionated Guide to ML Researchyou should allocate some fraction of your time towards improving your general knowledge of ML as opposed to working on your current project.
An Opinionated Guide to ML ResearchIn my experience, switching problems too frequently (and giving up on promising ideas) is a more common failure mode than not switching enough.
An Opinionated Guide to ML ResearchGo back and look at the list of machine learning achievements you admire the most. Does your long-term research plan have the potential to reach the level of those achievements? If you can’t see a path to something that you’d be proud of, then you should revise your plan so it does have that potential.
An Opinionated Guide to ML ResearchYou should constrain your search to solutions that seem general and can be applied to other problems.
An Opinionated Guide to ML Researchchoosing a different problem from the rest of the community can lead you to explore different ideas.
An Opinionated Guide to ML ResearchI personally recommend goal-driven research for most people, and I’ve mostly followed this strategy myself.
An Opinionated Guide to ML ResearchYour ability to choose the right problems to work on is even more important than your raw technical skill
An Opinionated Guide to ML ResearchBefore continuing, it’s useful to spend a few minutes about which findings and achievements in ML have been most interesting and informative to you.
An Opinionated Guide to ML ResearchThe keys to success are working on the right problems, making continual progress on them, and achieving continual personal growth.
An Opinionated Guide to ML ResearchWe experimented with autoregressive and diffusion approaches to discover the most scalable AI architecture, and the diffusion-based approach for audio generation gave the most realistic and compelling results for synchronizing video and audio information.
Generating audio for video - Google DeepMindacknowledges that using AI-driven algorithms to aid killing is "morally complex."
Palantir: The all-seeing tech giant | The WeekPrivacy advocates say that this library of information could be weaponized by the government. Officials could access and release embarrassing details about critics' finances and health conditions, for example, or terminate benefits they receive.
Palantir: The all-seeing tech giant | The WeekHe insists he wants "less war," and the way to achieve that is to "scare the f--- out of our enemies."
Palantir: The all-seeing tech giant | The WeekThe federal government has copious data on U.S. citizens, ranging from incomes and student debts to criminal histories, charitable contributions, and medical claims. Such data is currently siloed across agencies. But the Times reported that Trump, who in March signed an executive order directing agencies to remove "unnecessary barriers" to data consolidation, had drafted Palantir to build a centralized database that pulls all that disparate information together.
Palantir: The all-seeing tech giant | The WeekSome recent waveform-based approaches for speech separation have surpassed the performance of Ideal Masks
TF Representations and Masking — Open-Source Tools & Data for Music Source SeparationFor a mixture with 𝑁 𝑁 sources, the element-wise sum of every mask should equal a matrix of all ones,
TF Representations and Masking — Open-Source Tools & Data for Music Source Separation