Milad Alizadeh

@y0b1byte I like it 🤩

@notmilad you'll like it))

Okay, replacing my Zotero usage with this setup: - One markdown file for inbox/all the notes. - Each note has a title, a path on my local machine (all papers are timestamped and are in a single folder), tags and notes - I press a hotkey, and the script on the left is run on the content of my clipboard. The paper is added to the list on top. Radical simplicity.

It is just so sad that the #NeurIPS2024 main conference ended with such a racist remark by a faculty when talking about ethics. How ironic! I also want to commend the Chinese student who spoke up right on spot. She was respectful, decent, and courageous. Her response was exemplary: she began by acknowledging the speaker’s efforts, then gave the speaker an opportunity to clarify (though, regrettably, the speaker’s reply only reinforced her bias), and ultimately called attention to the inappropriate racial bias and offered constructive suggestions. Thank you for speaking out!

@y0b1byte I do this for git integration and it works decently: #L22" target="_blank" rel="nofollow noopener">github.com/mil-ad/dotfile…

Played with Helix a bit. It's quite slick! But a couple of features will make it less convenient for me: not being able to show the current buffer structure (e.g. what methods are there in the class etc) and better git support (nvim/fugitive is amazing). Will give it a go again in a couple of months. Good defaults and easier setup is def a win. I'm fine with it being less customisable as I'm trying to be as barebone as possible. LSP support is there, that's great.

thoughts on alignment

@markpash بیا باهم حرف بزنیم 🤗

فکر کنم باید عمیقاً این LLM ها (چیزهای چت جیپیتی مانند) رو یاد بگیرم. به کارم میاد. دلم میخواد یه مدل رو دائم آپدیت کنم از یه دیتاست خاص. ولی همه چی برام سواله، از آماده کردن دیتا، تا ساختن مدل، تا اجرا کردن مدل توی حالت چت. چقدر این چیزا open source هستن که بتونم خودم انجام بدم؟

New from me. Wondering how “existential risks” around AI causing human "extinction" became mainstream? It’s part of a tech-billionaire backed effort that relied on colleges to grow—both as a pathway to prestige & recruiting ground for idealistic talent washingtonpost.com/technology/202…

All 7 women in this survey said "no" to "Is an AGI likely to cause civilizational disaster if we do nothing?". 6/7 warned of that narrative distracting from corporate power grabbing and other immediate risks. Only 5/15 men said "no" to this question. (p-value for χ² = 0.014)

Top AI researcher Kyunghyun Cho (@kchonyc) dismisses AI ‘extinction’ fears, challenges ‘hero scientist’ narrative “I am very aware of the fact that the EA movement is the one that is actually driving the whole thing around AGI and existential risk.” venturebeat.com/ai/top-ai-rese…

The Future of Life Institute is a problem. Being in same age-group (lower end maybe😃)/cultural background (8-bit programming📼⌨️) as these men, makes me feel uncomfortable and embarrassed for them. Here they (Musk, Hassabis, Tegmark, Bostrom etc.) are in 2017.

👏🏽👏🏽

I was planning to post this event here today, but it has sold out before I was even able to advertise it. I decided to tweet anyway to emphasise the obvious demand for these conversations and to encourage other arts organisations to do what they can... vam.ac.uk/event/gJQ5oy1P…

If my disgust for Nick Bostrom’s intellectual trash hadn’t hit peak vom with his embrace of (not so borderline) support of eugenics as part of longtermism and suckling the corrupt teat of the trashiest billionaires, it now has.

I have finally written about this. wired.com/story/effectiv…

Huge thanks to @niclane7 and @yaringal for years of great supervision and helping me actually finish this thing! Thanks also to @SongHan_Omni and Andrew for making the viva as pleasant as possible. Looking forward to much more ML @CohereAI 🤖

I spoke to some brilliant artists who are producing work in support of the current protest movement in Iran and wrote about them here: theguardian.com/artanddesign/2…

@OATML_Oxford group members and collaborators are proud to present 11 papers at the #ICML2022 main conference and workshops. Group members are also co-organizing the Workshop on Computational Biology, and the Oxford Wom*n Social oatml.cs.ox.ac.uk/blog/2022/07/1…

Check out our latest work on SSL for Human Activity Recognition using 700,000 person-days of wearable data - the largest study to date! 📃arxiv.org/abs/2206.02909 🖥github.com/OxWearables/ss… Amazing collab with @angerhang, Shing Chan, @apcreagh, David Clifton & @aiden1doherty!

@OATML_Oxford members will be presenting a number of papers and also co-organizing a workshop. See you at @iclr_conf this week! oatml.cs.ox.ac.uk/blog/2022/02/0…

@roydanroy @y0b1byte We do have some ImageNet experiments.

@y0b1byte @notmilad Well, that would be one explanation. Or they are conflicts of these authors.

Happy to share our latest work, where we use meta-gradients to prune networks *before training*. Check out the paper and code to see how we did it and how this connects to the lottery ticket hypothesis!✂️🍀#ICLR2022 📄 arxiv.org/abs/2202.08132 👨‍💻 github.com/mil-ad/prospr 🧵1/6👇

Work done with @satailor96, @luisa_zintgraf,@joost_v_amersf,@seb_far, @niclane7, @yaringal from @OATML_Oxford, @CaMLSys, @whi_rl To appear at ICLR 2022! 📄 Preprint at arxiv.org/abs/2202.08132 👨‍💻 Code at github.com/mil-ad/prospr 🧵6/6

We show empirically that Prospect Pruning achieves higher accuracy compared to existing pruning-at-init methods. Our approach is also "single shot" in the sense that the pruning is applied to the network's initial weights in a single step. 🧵5/6