Alisa Liu emerged from a University of Washington PhD with training contributions to models such as Olmo 3 and Nemotron 3 along with thousands of citations, yet still navigated dozens of algorithmic interviews across frontier labs before securing an OpenAI offer after roughly three months of effort.
It never goes away. Doesn’t matter how many models you have trained or what you did. Someone will still try and leetcode you.
6 yr ML PhD, trained Olmo 3, trained Nemotron 3, but still forced to grind Leetcode and Neetcode 75.
Despite all the headlines saying otherwise, Leetcode is clearly not dead.
Somehow knowing dynamic programming is more important than knowing data parallelism for interviews.
Professionals who have already built large-scale models continue to encounter the same NeetCode and LeetCode sequences required of less experienced candidates, leaving open how well these exercises map to actual research impact.
Even strong publication records and model work do not appear to waive standard coding screens at top labs, though the full extent of this pattern across the industry stays unclear from public accounts so far.
Users in the replies resent LeetCode interviews for ML roles as a silly humiliating ritual that even researchers who trained major models must endure.
No Digg Deeper questions have been answered for this story yet.
@code_star I indeed got a silly ml coding interview in my “job search”. Everyone goes through it.
@code_star my leetginity remains intact... for now...
אם שאלתם את עצמכם מה מסעיר את פיד AI ביממה ההאחרונה התשובה היא סטודנטית שפירסמה בלוג על הראיונות שהיו לה בתעשייה. הפיד סוער כי למרות שהיא סיימה דוקטורט באוניברסיטה מובילה, הייתה שותפה לכמה מהמודלים הגדולים האחרונים ובעלת אלפי ציטוטים היא עדיין עברה הרבה ראיונות leecode, נכשלה בכמה ובכללי לקח לה די הרבה זמן למצוא עבודה (בסוף היא מצאה בopenai). מה הטייק שלי מהעניין? אני לא מאמין שחזרנו לדיון על ראיונות leecode כן או לא 😢
The irony is I’m actually really good at traditional leet code stuff. I’m terrible at the sillier ml coding stuff.
Like I can implement a doubly linked list faster than two layer back prop in numpy. Mostly, because I’ve never taken the time to memorize the derivatives so some functions … that’s what auto diff is for.
@code_star Talk to a retired Director who had led a set of robotics labs at NASA over his career. He interviewed at Google for a Director-level position too, and was leetcoded! :p Son of a gun got through as well, and worked at Google for a few years before founding his own company.
@code_star i got very lucky, because i cannot do leetcode, i failed almost all interviews like that and where i got hired i wasn’t asked for leetcoding
@alakazam1314 I actually love leetcode (even if I disagree with it being a good criteria). I hate the toy ml problems though. They somehow seem even less relevant to what I do professionally.
@code_star Death, taxes, and leetcode
@code_star no one can escape the humileetation ritual
@code_star what if you can use claude code or codex to solve your leetcode for you? does that count?
@natolambert @code_star Do you still think leetcode prep is relevant? I understand knowledge of DSA & problem solving skills are essential, but I know who did 750 + leetcode and can solve it in a lot optimal way than ML folks, it feels like a wrong metric especially in this era, what do you think ?
@natolambert @code_star At some point it stops though.
Like Noam or Kaparthy was not doing leetcode.
@kalomaze @code_star real
@natolambert @code_star I don’t think ML coding interview are a problem they are usually very discussion based. Leetcode on the other hand…
@kalomaze @code_star Another prime intellect W
@code_star Same for software engineering, leetcode doesn’t match what I do actually at work. Not even close.
@ziv_ravid @_kobim לצלוח מבחנים - מכל סוג - ולהיות חוקר ואולי אפילו מפתח מבריקים, לא יושבים בדיוק על אותה משבצת.
@_kobim יש כאן שני דברים - איזה סוג רעיונות צריך והשני זה, אם מישהי כל כך מוצלחת (יחסית לסטודנטית) נכשלה כל כך הרבה זמן, מה הסיכוי של כל השאר...
@paulmarin90 @code_star no they make you write the code yourself so they can see it
@_kobim @ziv_ravid אורי ואני שנינו עם דוקטורט בבינה מלאכותית ולאורך כל המסלול מאז התואר הראשון והלאה לא התכוננו לראיונות קוד. קצת מתוך עקרון של - לא באנו להיות code monkey, אם הערך המוסף שלנו לא זוהר בלי זה, אז אנחנו לא במקום הנכון ממילא.