Pratik Mehta

I’m probably screaming into a void but a common failure mode of LLMs is that they try to stuff nullable variables that don’t accept nullables (like lists)

Feed is straight up trash now

The paradox of wanting to avoid conforming but seeking vindication in external validation

I feel like my general life strategy of "ignore the trendy thing, focus on what you believe in" was predicated on the trendy thing only being, like, 10x bigger than whatever I was doing, and not a world-historically large technological revolution.

Sure, there are deep research and reasoning LLMs - they’re good for quite a lot of current use cases but not robust enough. It’s hard for me to tell if experts agree on this.

Isn’t it weird that we have the perfect chat interfaces for advanced AI before the substrate for robust reasoning is ready?

i love how people are saying "if we write a sufficiently detailed specification, the agent can write all our code" do you know what writing a sufficiently detailed specification that deterministically maps to what a computer's actions is? it's coding

@pmehta94 Today we call this three-dimensional space "world model"

For all the brilliance of LLMs, these words from @yudapearl 's Book of Why still ring true

sigh. alright agents, here we go

My experience with translated Japanese posts has been great for the most part. I guess you get what you curate, no matter the language

Just got around to listening to the acoustic tracks on The Life of a Showgirl. @taylorswift13 hits the brief so well!

@harnidhish That is some heart-warming concern for the safety and wellbeing of your fund

Once in a while the spam folder makes me lol.

The history of key scientific discoveries should be its own subject of study. Such fascinating stories!

@sarahookr Gotta end the creative pleading phase of software engineering

this is honestly why the most important measure of progress in building true intelligence is our ability to eliminate prompt engineering.

crocodile: hi fish, how are you? fish: subah fever tha, magarmachchh better now

My unfortunate read on Indian ambition is that it's just preparation for ambition, and we've lost the ability to tell the difference. Like the credential was supposed to be the door, right? You get a credential and then you use it. But imo it has become the house and now everyone's just redecorating every few years with NO idea what they actually want their home to look like. The sequence is: get the degree, get the job, get the credibility, THEN take the risk. Except the credibility threshold keeps moving and by the time you've cleared it you have EMIs and your parents finally stopped worrying and the risk feels a lot more scary than it did at 22. So you don't do it. You become a mentor instead and point someone younger at the same treadmill. What's incredibly telling is what we do with people who just...started without waiting for permission. We don't celebrate them, we absorb them. Guest lectures. Fellowships. Honorary titles. The institution needs to stamp you before it can admit you were right. Anyway what I'm actually saying is: credentials are a way of not finding out if you're good. Every new credential is another year before the terrifying question of if you're actually good needs a real answer. Completely rational if you're trying to stay undefeated in a game you never quite enter. We're a country that worships risk-takers and systematically raises risk-avoiders and then acts confused when the builders don't come from the pipeline. Kind of insane, I think.

The real sign of AI writing is not superficial stuff like “It’s not X—it’s Y”. It’s the hollowness. Polished writing but relatively mundane ideas. The giveaway is that you’re less impressed when you read it the second time. With good writing, it should be the other way around. I’m not sure this is inherently about AI. It’s more about the fact that people tend to turn to AI when they don’t have much to say. Reading text that has the syntactic smell of AI is mildly annoying, but when I read hollow writing I feel the writer is wasting my time, which is much more frustrating. So don’t do it. People are unlikely to respond to your email or subscribe to your newsletter or whatever you’re trying to get them to do. And they’ll probably remember that you betrayed their trust as a reader.

@chrisalbon Spectral clustering FTW

Still my favorite concept in ML