Rutu Mulkar

[Rutu Mulkar @rutumulkar]

@GaryMarcus You said it first though!

[Pedro Domingos @pmddomingos]

LLMs do more than pattern recognition and less than reasoning. We don’t have a name for it yet.

[Rutu Mulkar @rutumulkar]

It is so true that I the skeptics can fill buckets with their tears of laughter from this one!

[Gary Marcus @GaryMarcus]

Fans of LLMs: AGI has been achieved; anything a human can do LLMs can do

Skeptics: Um, they can’t do X, Y, and Z

Fans: But they aren’t *supposed* to do X, Y, or Z. But they can play chess!

Skeptics: Hmm, they still make up moves…

Fans: Who said they should play chess?

[(((ل()(ل() 'yoav))))👾 @yoavgo]

no the transformer architecture is not turing complete. its a very bounded computation.

[Daily Loud @DailyLoud]

A cop tried to pull over a self-driving car 🤣🚔

[Gary Marcus @GaryMarcus]

Revealing example! Humans can reason about why cases like these are sexist; LLMs cannot.

[Rutu Mulkar @rutumulkar]

There is just so much information you can feed into a machine that models human language, before you reach the long tail of commonsense reasoning which is prohibitively expensive to encode into a machine to model/mimic.

https://t.co/2X6p23sTIb

[Yann LeCun @ylecun]

LLMs do *not* capture much of human thought, because most of human thought and all of animal thought is entirely non verbal.

The factual, logical, and physical reasoning mistakes that current LLMs make clearly show that they have *not* captured much of human thought.

[Rutu Mulkar @rutumulkar]

@RichardSocher Showing the long tail results. The results on the first page are usually boosted by keywords and gamified to show up on page 1. Most of the gold nuggets are in the long tail.

[Rutu Mulkar @rutumulkar]

@mindykaling Is that Barfi I see in the lunch? Yum! https://t.co/XFxUg0Rbtm

[Rutu Mulkar @rutumulkar]

😂

[@emilymbender.bsky.social @emilymbender]

For more on why language models aren't generating meaningful text even when they seem to be, let alone learning about the world, see https://t.co/z1F7fESFOn and the papers we cite.

[Rutu Mulkar @rutumulkar]

AI is trained on data, and it emulates the examples it has seen to repeat it in a meaningful way. Creating new art, or fundamentally new pieces of data has been outside the scope of AI as it stands currently. Today, AI is a mathematical optimization of data.

[immad @immad]

3/

- Creates jokes, poetry and other art
- Companionship bot - doesn't have to be like Her movie, but could just be someone to talk to.
- Compile personalized trip itineraries.
- Synthesize and compare legal contracts
- Personalized recommendations of things to buy

[Rutu Mulkar @rutumulkar]

@pabloduboue Congratulations Pablo! :-)

[Rutu Mulkar @rutumulkar]

I am curious what was the quality of the content that was removed by each of the companies. And that is the definition of "hate" that is followed by each company?

[Yann LeCun @ylecun]

The EU periodically evaluates how social network companies take down illegal hateful content.

Facebook took down 87.6% of flagged hateful content, YouTube 79.7%, Instagram 42%, and Twitter 35.9%.

EU report: https://t.co/0onGHMrc59

[Rutu Mulkar @rutumulkar]

@zelandiya This reminds me of the Pixar short film - bound and rebound: https://t.co/D7Rg7qkpwB

[Rutu Mulkar @rutumulkar]

@zelandiya Yes we are well. Hunkered down at home and trying to make the most of the lockdown.
Aarini (6 yo) is a published writer now: https://t.co/2PavQgV9O7

[Rutu Mulkar @rutumulkar]

My 6 yo wrote her second book! It is funny and silly and actually adventurous. Check it out!
My Lovely Year by Aarini Kyra Mehta https://t.co/Z2UGWIuMrb via @amazon

[Venture Out @GoBigtoSmall]

Today we opened rolling admissions into VO's Launch program! 3-5 founders accepted every other month to work with a curated list of Seattle's best mentors, founders and investors to launch your startup!

You don't have to quit your day job to participate! https://t.co/Wahoc7U7za