Mohonish Chakraborty

Senior AI Engineer. MS in Software Engineering '20 @ Carnegie Mellon University. Writing Generalist. Curious.

[Mohonish Chakraborty @mohonishc]

https://t.co/Vi7O7fDqfT is an example of web design that is so good it feels like a movie where the subtitles are distracting.

The visuals are so pleasing, I did not read any of the text. What does Pear do? Ugh, need to scroll - I mean - scrub the timeline again.

[Mohonish Chakraborty @mohonishc]

@hams Am I the only one bothered by these “gloopies”? Is this for real?

[Mohonish Chakraborty @mohonishc]

The EU AI Act now requires providers to mark generative output so it can be detected. So Anthropic and Google are embedding watermarks in the choice of words. But does it survive the well-placed emoji?

Read more: https://t.co/TtlY09xGeI

[Mohonish Chakraborty @mohonishc]

If you are curious and would like to learn more about AI-generated text and it's various detection mechanisms (sans the tech bro jargon), read more:

https://t.co/y6TlZwnaps

[Mohonish Chakraborty @mohonishc]

A promising startup in this space is @pangram - they built a classifier to detect AI-generated text. No watermarks needed.

The known limitations of it's high accuracy classifier is (funnily) people who just happen to write like LLMs.

[Mohonish Chakraborty @mohonishc]

When the LLM is working on "The capital of france is _" the word "Paris" will have 99 point something probability. Paris will win.

Compare that with "The city was _" which has many probable options and is therefore able to encode a watermark more easily.

[Mohonish Chakraborty @mohonishc]

In the simplest of terms, a secret key will split the possible next words into two groups. The probability of each word in one of those groups is multiplied by a factor, so they get chosen by the LLM more often.

[Mohonish Chakraborty @mohonishc]

Given that key, you can now place it in each position and determine how often the words from the preferential group were chosen. So now you know AI generated that passage.

But it has many limitations.

[Mohonish Chakraborty @mohonishc]

Earlier this week Anthropic explained how Claude marks AI-generated content. Of the two mechanisms, text is interesting because it hides the signature in the choice of words.

[Mohonish Chakraborty @mohonishc]

@scottbelsky you’re right. I think a common deterrence for shipping bold UX designs is the need to cater to every target consumer persona, thereby converging towards the average experience that works for all. That won’t be a problem anymore.

[Aaron Levie @levie]

The elimination of engineers was one of the wilder hypotheses out there. Just like absurdly wrong. We just gave engineers a power tool that can accelerate the development of whatever we want.

Of course their value not only remains critical in that world, but actually goes up in many domains because we can apply engineering to far more work than before.

If you’re trying to automate drug discovery, you need engineers. If you’re trying to automate manufacturing, you need engineers. If you’re taking on larger and larger software projects, you need engineers.

This will happen in domains beyond engineering too. AI causes companies to take on way more work than before, which leads to needing more experts to oversee the work.

Even as models get better and better, they can be better utilized by the experts in those fields than the novices. Great time to be an expert.

[Mohonish Chakraborty @mohonishc]

But AI agents have no nervous system, and they feel nothing. They have a different constraint - tokens.

So the website that wins is the boring one, with a markdown version for AI.

Read more: https://t.co/0C8spzqXPO

[Mohonish Chakraborty @mohonishc]

The web as we know it is changing because it has a new consumer: AI agents.

But for the last twenty years, every serious business has spent enormous amounts of time and money optimising the conversion funnel for humans.

[Mohonish Chakraborty @mohonishc]

All of that rested on an assumption that is being invalidated today:

> A human arrives on a webpage, feels something, and can be persuaded

[Mohonish Chakraborty @mohonishc]

Grateful for the opportunity to work on this incredible project and kudos to my amazing teammates!

[Master of Software Engineering Programs @MSE_CMU]

Congratulations to Mohonish Chakraborty, Jiawu Li, Dan Myers, and Vaithy Narayanan for receiving the Software Engineering Project Excellence Award!

#softwareengineering #kudos #dreamBIGGER

https://t.co/Fj8eQr3AnH

[Marques Brownlee @MKBHD]

Nobody Tells This To Beginners

https://t.co/1nONUt68cl

[a meme page to check while Xcode is indexing files @ios_memes]

🙏

[Swift Language @SwiftLang]

The Swift community just keeps building more cool stuff!

Announcing the new Swift Atomics open source package:

https://t.co/m9iS0cEeqT

[Swift Language @SwiftLang]

Introducing Swift on Windows.

New toolchain image is available for Windows, downloadable from https://t.co/5NNXraGyus. To learn more: https://t.co/D4w6d6ME1h