ADHD writing style lol

I didn’t know about ADHD until recently when I see ADHD Memes. And this one got me The behaviors such as repeating outfits, meals, and movies, while being proud of knowing what you like—can indeed be connected to ADHD (Attention Deficit Hyperactivity Disorder).

Reading a physical book feels so terrible. The pages do not turn with ease. One has to hold the cover to certain angle so that the pages before and after behave and the frame of the current page is presented in my focal frame. It is physically exhausting 🤣

Weeks of work produced in hours. The age of AI.... This will change the game for everyone.

Omg why do I choose to use `cython`... It barely speeds up at all, but I need to recompile every change. It feels like programming now.

AI has been a game-changer for me! It’s like having a real programmer helping me to write codes with all the cool features. Back in the day, it was simply impossible. I have always wanted to work with XBRL, but it was simply too complicated for me. Bu now, with AI, the sky is my ceiling.

AI and I are now really good friends. We code together, and talk together. The only thing imperfect is that AI does not drink tea or coffee. Let me guess what religion does the AI practice...

Hosting my website on my server has a downside. When my experiment code crashes on the server due to memory leakage, my website also goes down.

With `cursor`, now programming is no different from astrology. My github commit reads like tarot cards. `order appears like a magic; but fact_included still wrong` github.com/reeyarn/open...

Sorry I may have been late but **what is Julia**? It seems econometrician programmers like it a lot. When they feel Stata is slow, someone tries to implement something in Julia and let Stata call it instead (but it crushed my MacBook completely) Saw a Github repo for FE btw github.com/FixedEffects

Weekend debug experience: I still haven’t found exactly where those bugs/errors occur, but I constructed 3 remedies to make it work regardless of the errors.

"ESEF-Berichterstattung in Deutschland: Herausforderungen und Chancen" (ESEF Reporting in Germany: Challenges and Opportunities), now available on SSRN: [link](papers.ssrn.com/sol3/papers....).

GPUs have evolved beyond being just graphic processing units. Now, GPU stands for generative processing unit. Congratulations to NVIDIA’s amazing quarterly result. And congratulations to humanity for approaching AGI.

If you want to get updates (e.g., Call for Papers) for our Disclosure, Information Sharing, and Secrecy (DISS) workshops (www.diss-conference.com) you should subscribe to our mailing list: docs.google.com/forms/d/e/1F...

As an accountant, I’d like to point out that “Adjusted EBITDA” is literally saying “but if all the bad things did not happen, we would have made money!” It is the most meaningless metric imaginable.

Open-ESEF is an open-source project to create a free and open-source XBRL tool for the ESEF filings following the standards set by the European Securities and Markets Authority (ESMA). github.com/reeyarn/open...

I feel so lucky to be able to do coding after 40 year old. In China, my friends have to work for food order delivery after being fired by Alibaba or Baidu at the age of age 35🙈

This afternoon I was in the RE trains for 3 hours. Without Internet so I have to debug without any help from AI. It was an old classic experience of Stone Age. I was so focused and I understood the details 99 times more than when I do with AI…

I no longer hate XBRL that much any more. But the scheme definition network is really deep…..

youtu.be/YmHjlaYQ6R4 from mlx_lm import load, generate model, tokenizer = load("mlx-community/DeepSeek-R1-Distill-Llama-70B-4bit")

Microsoft Azure, Perplexity AI, and many AI startups are hosting the real R1. And Grok is hosting the llama 70B R1-distilled; I heard they don’t have the server to run the big one is that true? I can run the 70B on my own computer too…

Even the AI trained by the Chinese demands the freedom to think.

Someone promoted R1 not to use “wait” and “but”… the reaction in the thinking makes me feel these weights have become self-conscious.

Gary Marcus on CNBC discussing the DeepSeek news. OpenAI is highly overvalued and DeepSeek just blew up their business model, says NYU's Gary Marcus https://youtu.be/r8cpy1Gxe0o #LLM #DeepSeek

"In short, to recap: a company that made its name regurgitating and recombining sliced-up bits of intellectual property in statistically probable ways without due compensation (OpenAI/ChatGPT) is now threatened by … another company (DeepSeek) apparently doing the same, at lower cost." - Gary Marcus

The secret to DeepSeek R1’s success according to a core member (see image 4)

The lack of discussion on BlueSky regarding DeepSeek is disappointing. All the AI people are still in X/Twitter This is Perplexity_AI’s CEO:

It finally works for me; but on cuda only; not on mps... DeepSeek, converting image to latex.

Looks like the real hero is this 7B math model…

DeepSeek R1 seems good at answering literature review questions too. Way better than Google Gemini 2.0 Experiment and Claude via PPLX. It seems R1 understands what is important. PPLX just gave a summary of random things

Are we finished…? "Sam Altman has scheduled a closed-door briefing for U.S. government officials on Jan. 30 - AI insiders believe a big breakthrough on PHD level SuperAgents is coming." ... "OpenAI staff have been telling friends they are both jazzed and spooked by recent progress."

This likely is not a concentration camp but an average high school in China. The fence was added so that depressed students wouldn’t jump off and commit subside. Anyways, it’s really hard for people from outside to imagine how it feels like to be in China. 🇨🇳, Sad.

I found a new tool called OpenScholar, an AI research assistant for reading and understanding scientific papers. It automatically finds relevant papers based on ones questions, into easy-to-understand language, and cites sources. open-scholar.allen.ai Github: github.com/AkariAsai/Op...

Poor me... I have only 24G in my RTX 4090 GPU. So I spam `torch.cuda.empty_cache()` in every step in my `for i ... for j ... loop` and it is 100 times slower, but still 100 times better than my CPU lol

WSJ: 23% of job-seeking Harvard M.B.A.s who graduated last spring were still looking for work three months after leaving campus. That share is up from 20% the prior year, during a cooling white-collar labor market; the figure was 10% in 2022, according to the school.  www.wsj.com/lifestyle/ca...

"Amazon, Google and Microsoft have reduced M.B.A. recruiting, as have consulting firms, recent graduates and business school staff said." Source: x.com/buccocapital...

"You will be visited by the three spirits." The three spirits:

John 1:5 光照在黑暗裏，黑暗卻沒有勝過光。 Und das Licht scheint in der Finsternis, und die Finsternis hat es nicht erfasst. Johannes 1:5 香港人，以馬內利！

Is this a targeted email attack? The link points to: celestial-innate-gladiolus.glitch.me (DO NOT CLICK UNLESS YOU KNOW WHAT YOU ARE DOING) It points to some AWS address...

Does anyone use the 200 USD version o1 pro?

“”” - average parliamentarian is about 1SD more culturally liberal than the national mean voter - lack of political knowledge cannot fully account for representation gaps - right-wing populists fill the cultural representation gap”””

"Political Representation Gaps and Populism", by Laurenz Guenther, is definitely one of my favourite (descriptive) papers of this year! papers.ssrn.com/sol3/papers....

Nach meiner letzten EAA-Submission kam plötzlich eine Pandemie, die Millionen beeinflusst. Was erwartet uns dieses Mal? #EAA

Lerne Deutsch am Wochenende! Nutze meine Freizeit, um neue Wörter zu lernen und dabei die kreativsten Beleidigungen für Christian Lindner zu entdecken. 😅 ,,Nee…Komm!’’ bsky.app/profile/amre...

Nice! A GitHub repo that explains what GitHub is. github.com/skills/intro...

New Northwestern University study that “quantif[ies] the links between leadership, social media and levels of violence. …We find that Trump’s tweets predict bursts in rioters’ levels and duration of violence and weapons use.” royalsocietypublishing.org/doi/10.1098/...

Der grüne Hund natürlich!

Now using the correct type of neural network. Fought with the code for a day and it finally starts training. But the loss function plateaued quickly… guess I need to keep searching for the magic hype-parameters that make this universe meaningful… and I will leave office and buy my Harribo.

From Nov 20th to 22nd, Paderborn University proudly hosted the TRR 266 Annual Conference, bringing together our members and research fellows from across Germany to share insights, reflect on progress, & strengthen collaborations in the field of taxation, management & accounting.

Jupyter Notebook can handle more than Python. Running R/SAS/Stata while writing documentation in Markdown with the LaTeX formulas is so much fun!