Kia Karbasi

King – Man + Woman = Queen

[Kia Karbasi @karbazhyev]

Bullet train coming soon: Tehran to Tel Aviv by way of Baghdad #MahsaAmini

[Kia Karbasi @karbazhyev]

@heamiri دزژوری

[Kia Karbasi @karbazhyev]

@heamiri @grok true?

[Andrej Karpathy @karpathy]

Don't think of LLMs as entities but as simulators. For example, when exploring a topic, don't ask:

"What do you think about xyz"?

There is no "you". Next time try:

"What would be a good group of people to explore xyz? What would they say?"

The LLM can channel/simulate many perspectives but it hasn't "thought about" xyz for a while and over time and formed its own opinions in the way we're used to. If you force it via the use of "you", it will give you something by adopting a personality embedding vector implied by the statistics of its finetuning data and then simulate that. It's fine to do, but there is a lot less mystique to it than I find people naively attribute to "asking an AI".

[GLADIA Research Lab @GladiaLab]

LLMs are injective and invertible.

In our new paper, we show that different prompts always map to different embeddings, and this property can be used to recover input tokens from individual embeddings in latent space.

(1/6)

[Kia Karbasi @karbazhyev]

@PouyaSamangouei Iran is china’s houthis

[Naval @naval]

The average man lives in a matriarchy, the average woman lives in a patriarchy, and they’re both correct.

[Kia Karbasi @karbazhyev]

@notlouisck When you do a 180, left becomes right and right becomes left

[Kia Karbasi @karbazhyev]

In life, you either want to do something but you can’t, or you don’t want to do something but you can.
Either way, it ain’t gonna happen!

[Kia Karbasi @karbazhyev]

Turned out Biden was the other Carter🤷‍♂️

[Kia Karbasi @karbazhyev]

#Trump was republican #carter

[Kia Karbasi @karbazhyev]

https://t.co/zTPY6yjvyG

[Kia Karbasi @karbazhyev]

@elonmusk Yes! We live in a republic not a direct democracy

[Pouya Samangouei @PouyaSamangouei]

Hey @GoogleDeepMind please give us Deep Research with Gemini 2.0 instead of 1.5 , and show inference time compute (thinking/reasoning) outputs 🙏 please also give access to this amazing service via API!

[Naval @naval]

Your family is broken but you’re going to fix the world.

[Kia Karbasi @karbazhyev]

United States of Earth (USE)

[Omead Pooladzandi @HessianFree]

We've been cooking.

No this is not PSGD. Yes, I have trained a 3b parameter model with it.

[Kia Karbasi @karbazhyev]

America, year zero!

[Kai-Wei Chang @kaiwei_chang]

Here's a list of NLP superstars 🌟 who are beginning their journey 🚀 in the 2023/24 academic year.
@jieyuzhao11
@GabrielSaadia
@acbuller
@Lianhuiq
@ManlingLi_
@yuntiandeng
@rajammanabrolu
@YueDongCS
@tanyaagoyal
@MinaLee__
@yuntiandeng
@alsuhr
@wellecks
@hllo_wrld
@Xinya16

[Walid Saba @sabawalid]

If the dumb parrot (ChatGPT) only knew that for all s:

reverse(reverse(s)) = s

It would not have reversed the list twice. I guess symbolic logic does count, after all 🙂

[Kevin Meng @mengk20]

How & where do large language models (LLMs) like GPT store knowledge? Can we surgically write *new* facts into them, just like we write records into databases?

Explainer 🧵 on how interpretability & model editing go hand-in-hand, and why these emerging areas are so important 👇