Karthik Rajkumar

@johnjhorton @ycombinator Congrats, @johnjhorton !

Expected Parrot just launched on @ycombinator's Launch YC! Expected Parrot: Simulate your customers with AI agents. Check them out: ycombinator.com/launches/Ol2-e…

@johnjhorton True definitely, but one thing about academia is that researchers go out of their way to make salient the institutions they work at on papers, etc. An effect this has is the lay reader consumes that info as co-equal to the authorship info itself.

"A {{ prestigious_university }} study shows.." is typically a strong indicator you should discount what follows next. Universities don't do studies---specific researchers do---and a characterization of a study that starts with a category error should put you on guard.

@johnjhorton Don’t know what I love about Claude more. The constant “you’re absolutely right!” validation, or how it totally changes its opinion 180° when I push back against something it did.

Claude laying it on so thick it sounds sarcastic / it's making fun of me:

@alz_zyd_ Because they lie all the time, and not in the way an intelligent human would. With LLMs, you can just never ever let your guard down because they lie so spectacularly and unpredictably.

I honestly don't understand how one can use LLMs meaningfully for more than a few hours and not emotionally feel like the LLM is basically thinking. Does it not _feel_ human to you? I'm convinced people have just not tried talking to it like it's an intelligent human

@RobDonnelly47 There seems to be this experimental (and unmaintained) library, which converts pandas code to SQL? github.com/AmirPupko/pand…

What's the best way to programmatically construct complex sql queries in python? I frequently see people doing it via string interpolation, which is error prone. It makes me really miss dbplyr from R. #python #R

@alz_zyd_ Cmd + space on a mac should do it for ya much faster.

does anyone ever open up R/stata/matlab to do simple additions like 26 + 374

@alz_zyd_ And Twitter, even pre Musk, is very small compared to the industry as a whole. The hiring slowdown is more due to the postpandemic hiring boom + interest rates.

@alz_zyd_ By some estimates, Twitter’s active users have dropped by a quarter and revenue by more than half, both far worse than comparable platforms, since the privatization.

How beautifully ridiculous that one dude overpays for Twitter, fires 80% of workforce, and it still works, and thus single-handedly kills the tech entry-level job market

@AnthonyLeeZhang Now, now, let's not forget canids and corvids and cetaceans in our quest to anoint a flashy text generator as intelligent

It's only consciousness if it's from the Sapiens species of the Homo genus, otherwise it's just glorified pattern matching

@AnthonyLeeZhang The central point of statistics that has to be impressed upon ML engineers over and over again is just because you crunched some numbers doesn't mean you've uncovered what you care about. "just import sklearn" helps with executional speed but when your model fails

@AnthonyLeeZhang (because of poor calibration, selection bias in training data etc.) it remains as important as ever to understand what parameters you actually care about. Not to say academic textbooks aren't full of irrelevant theoretical bloat, but that doesn't undermine thinking about modeling

Actually a reasonable point We learn all these BLUE and such properties of linear regression, how outdated is this all now for a practitioner, to what extent should we cut some of the old stats theory from applied stats teaching

@EconBerger Made me literally LOL

Are there any therapists that specialize in methodological issues with the CPS and CES?

Tonight, @Penn President Liz Magill signaled that one of our nation's most prestigious institutions is willing to abandon its commitment to freedom of expression. “For decades, under multiple Penn presidents and consistent with most universities, Penn’s policies have been guided by the Constitution and the law,” explained Magill in a video posted to X. But now, she continued, the university “must initiate a serious and careful look at our policies,” a process to start “immediately.” This is a deeply troubling, profoundly counterproductive response to yesterday's congressional hearing on “Holding Campus Leaders Accountable and Confronting Antisemitism.” Were Penn to retreat from the robust protection of expressive rights, university administrators would make inevitably political decisions about who may speak and what may be said on campus. Such a result would undoubtedly compromise the knowledge-generating process free expression enables and for which universities exist. To be clear: Universities will not enforce a rule against "calls for genocide" in the way elected officials calling for President Magill's resignation think they will. Dissenting and unpopular speech — whether pro-Israeli or pro-Palestinian, conservative or liberal — will be silenced. Conservatives like Rep. Elise Stefanik should ask themselves: Do you honestly believe this rule won’t be weaponized to ban an Israeli cabinet official from speaking at Penn? An Israeli Defense Force soldier? The power to censor always invites abuse and never stays cabined. FIRE was founded in the wake of the infamous 1993 “Water Buffalo” incident at Penn. In that case, Israeli-born Jewish student Eden Jacobowitz was charged with harassment for shouting “Shut up, you water buffalo” at a group of rowdy sorority students outside his dorm room window. The sorority students were black, and the argument was that “water buffalo” was a racial epithet. But it was not. Jacobowitz, who speaks Hebrew, explained that water buffalo is a rough English translation of “behema,” a Hebrew slang term for a loud, rowdy person. The story captured headlines, and Penn was widely condemned for its persecution of Jacobowitz. FIRE co-founder Alan Charles Kors, a history professor at Penn, helped advise Jacobowitz. The charges were eventually dropped and the story would go on to serve as the opening chapter of “The Shadow University: The Betrayal Of Liberty On America's Campuses” — the book that launched FIRE. Over the years, Kors and FIRE helped Penn get past the water buffalo debacle. The school reformed all of its speech codes and was one of the first universities to earn FIRE’s highest, “green light” rating for speech-protective policies. But in recent years, Penn has backtracked. It’s no longer a green light school. It adopted new harassment policies that are ripe for abuse. And what free speech and academic freedom protections remain, it doesn’t consistently follow. Now President Magill suggests an institutional willingness to abandon free expression altogether. This will not end well. Vesting administrators like Magill with more power to police speech will result in more Jacobowitzes. The intended targets for these codes will not be the actual casualties — and Penn students, faculty, alumni, and donors will come to regret the day they ever entrusted campus bureaucrats with the power to police speech on campus.

As a professor who favors free speech on campus, I can sympathize with the "nuanced" answers given by U. presidents yesterday, about whether calls to attack or wipe out Israel violate campus speech policies. What offends me is that since 2015, universities have been so quick to punish "microaggressions," including statements intended to be kind, if even one person from a favored group took offense. The presidents are now saying: "Jews are not a favored group, so offending or threatening Jews is not so bad. For Jews, it all depends on context." We might call this double standard "institutional anti-semitism." University presidents: If you're not going to punish students for calling for the elimination of Israel and Israelis, it's OK with me, but ONLY if you also immediately dismantle the speech policing apparatus and norms you created in 2015-2016. Please read The Coddling of the American Mind. @glukianoff and I laid out exactly where the oppressor/victim frame came from (ch. 3), how it spread out of a few departments to gain power over administrators and campus culture (chapters 4 and 5), and how it drove the creation of the bureaucratic structures and processes that now have us all teaching and learning on eggshells (ch. 10). In chapter 13 we offer advice to leaders on how to to return universities to their academic mission and regain public trust.

@JonasMtzgr 2) We have a highly intelligent species right with us,—dogs—and non-specialists are no closer to having a decent conversation with them today. Now canine communication is body language, audio and smell, and not text, but we can collect all the data we want. Are LLMs helping here?

@JonasMtzgr Two points come to mind. 1) Humans already have dozens of scripts from ancient civilizations that remain undeciphered. I don't know if it's because we might not have trillions of samples for them, but are LLMs bringing us closer to deciphering them?

Aliens land tomorrow and give us a sequence of a couple trillion tokens downloaded from their version of the internet. We include it in the pretraining data of one of our LLMs. Will it be able to translate well enough to have a basic conversation about history and technology?

@AnthonyLeeZhang When I took real analysis at Stanford, the professor explicitly discouraged us from this terse style to a more narrative/prosaic style, while still maintaining rigor. There was a high emphasis on readability and it was actually a big learning moment for me.

Mathematicians have convinced themselves that a certain extremely terse style of textbook writing is good pedagogy. It is not, it is terrible pedagogy from the perspective of learning content, but ironically probably training for research because so much is left to the reader

I ❤️ this story! Both because academia is the better for having someone as strong and sharp as Allison, and also because mentors like Susan have actually cared to invest in all kinds of students to ensure brilliance everywhere is rewarded. My kinda happy ending 🥰

💯 💯 Totally what I’ve also been saying twitter.com/k__rajkumar/st…