Learn more on how the LLMs you use daily really work linkedin.com/posts/pranaypa…

This is great advice, and, though it doesn't please me to see these or similar 100+ times in an academic work, I try my best to highlight this to help Works Under Review. Many papers make claims far too strong, but rigor is crucial for people's understanding, trust, and more.

Didn’t realize how popular my reply to Guanya Shi’s post was until someone surfaced it via… Google Search, haha. Still stand by my statement!

Redis then I reconsidered using Redis for caching tokens to save approx 80% of costs for User and (user) System Prompts. Take it as a free tip bc this caching has had such high ROI for so long... ppl, let alone orgs, not using it... anyways, New goal now: >85% avg savings,

Proud of Ayla Croft! What a long way we've both come! Keep it up!!! 😃

Happy birthday, Samir! Keep it up. Bright future ahead for this rising tennis athlete!

Great job, Alyssa and Stanford Women’s Tennis! Way to stay cool and supportive of each other (no pun intended) during the rain delays and through the match. Yes—I stopped by shortly and wished y’all good luck! Though I don’t think you needed it :)

When you have to work because 10+ years of online gaming in MMORPGs taught you never to go to host on server-1 and to always pick server-2 so that you actually can load into the server. 🫠

Let's go!! Early-pocket-reach -> Aura Farm. Get it Tiago, Timo! Cal Men's Tennis! I still remember the 50 ft high overhead that I hit at 10-10 3rd set super-tiebreaker to send Jeff Kung, Me, Clements 1 point from State. Anxious my entire life. None at contact. Calm focus. Won.

Always aiming for signal in an ocean of noise: linkedin.com/posts/pranaypa…

< 200iq take: Poster states “officially,” but isn’t an official of anything except providing posts for which the lists of Likers make for great adds to shared filter lists. < 200iq tip: don’t be silly. Optimize for E[usefulness] in addition to virality if you “need” the latter.

It’s easy to use Opus and many other vLLMs to eviscerate these dissertations unless the Author used the models to augment excellent work, not fabricate it. I encourage accelerating output of quality work. Note: applies despite attempts to avoid detection. My experience. Yours?

When looking at François tweets, one of the obvious tells between a mediocre and great tweet based on his goal isn’t whether he is even close to correct or not. His goal is engagements, not truth. He is incorrect here. He knows this, and he met his goal through this. #PyTorch

Atal says this “might” be your highest ROI-learning this month “Might” -> unfalsifiable It’s due to a lack of confidence and fear He would benefit learning the definition of learning This video is a stepping stone. Treating it as THE goal -> low ROI Stanford NLP Group ✅ videos!

Now this is the type of person to learn from, not merely be emotionally provoked or entertained by! Thank you, Anselm Levskaya! —Your neighborhood global NeurIPS 2026 {placeholder} Chair

High-signal, no noise. I stop reading because this signals the copy-and-paster Author is negative ROI. This includes expectation of their content. Essentially everything.