Or Biran

[Or Biran @_orbiran]

@yuvalmarton I admire your resolve. It's obviously a trap and I would just respond with the required meta "meh, semantics"

[Or Biran @_orbiran]

@nyhabash That is awful. I am so sorry for your loss Nizar

[Or Biran @_orbiran]

If the difference between "fine tuned" and "emergent" is just that we explicitly curated the data to include something, then emergence just means we aren't rigorous in training base models. I don't think we have evidence of models learning behavior that clearly isn't in the data

[Jason Wei @_jasonwei]

Yann LeCun is obviously a legend but I found this tweet to be quite misinformed.

The whole point of "emergent abilities" such as few-shot prompting and chain-of-thought prompting, is that we clearly *did not* explicitly train or fine-tune them into the model. These abilities require scaling up to a lot of parameters, which is the reason we call them "emergent"---they are quantitive behavior shifts that come from qualitative changes. At least this is how researchers use the term.

All we did was train a language model with a lot of parameters on a big dataset. These abilities were not intentionally added into the model during training---rather, people found them via evaluations after the model was trained. A great pointer to this is Jacob Steinhardt's post: https://t.co/Y47ZuzkL00

What's also significant is that we often aren't able to predict these emergent phenomena ahead of time, and this has safety implications since risks can also be emergent.

Also, I don't know why finetuning is mentioned, because these abilities are clearly in the base model, which is should be apparent from papers before RLHF became popular... (e.g., PaLM-1, GPT-3 paper)

But even if you insist to take a finetuning as an example, ChatGPT RLHF was originally done in English, and it extended to many other languages. This multilingual ability was not "finetuned" into the model, which is a pretty good example of zero-shot generalization.

Please correct me if I'm wrong, but the technical use of the term definitely still holds.

Mustafa's tweet is also a bit misinformed, in my circles people don't use "emergence" to mean autonomy or agency. Again, what AI researchers usually mean is that it is a qualitative change that arises from quantitative changes.

[Noémie Elhadad @noemieelhadad]

My friend Drago just passed away. He left behind his wife, Axinia, and daughters, Laura and Victoria, who has a disability and requires care. We set up a GoFundMe so that Axinia can provide Victoria with the care she needs. If you can, please contribute: https://t.co/VSbLXBeCm7

[Or Biran @_orbiran]

Drago was one of the best 💔

[Harlan Krumholz @hmkyale]

The #AI community, the #computerscience community, the @YaleSEAS community, and humanity have suddenly lost a remarkable person, @dragomir_radev - kind and brilliant, devoted to his family and friends... gone too soon. A sad day @Yale @YINSedge @YaleCompsci #NLP2023

[Or Biran @_orbiran]

It's also the only one that seems crucial even with just increasingly powerful AI tools of the kind we make now

[Or Biran @_orbiran]

Even if you fully buy into the (agentful, dangerous) AGI stuff, (3) seems the hardest and least likely to happen anytime soon.

Is there any effort at all from any AI lab or AI safety org to openly push in that direction? I don't know of any.

[Sam Altman @sama]

things we need for a good AGI future:

1) the technical ability to align a superintelligence
2) sufficient coordination among most of the leading AGI efforts
3) an effective global regulatory framework including democratic governance

[Or Biran @_orbiran]

@cocoweixu Yay congrats!!

[Or Biran @_orbiran]

@GregHBurnham It means having a positive experience with something you've previously only had negative experiences with.

"Oh you visited Jamaica and hated it? But it's awesome! Let's go together and have a corrective experience!"

[Or Biran @_orbiran]

2005: "You do AI?" "Sorta, it's called ML, it's mostly just stats. 'AI' makes ppl imagine sentient evil bots"

2014: "You do ML?" "ML/AI. It's the same thing but the PR ppl say 'AI' sells better"

2023: "You do AI? I read a blog that proves it'll kill us all. How do you sleep?"

[Or Biran @_orbiran]

@jessyjli @UT_Linguistics Congratulations!!

[Or Biran @_orbiran]

When did we transition from artificial intelligence to artificial creatures?

[Yann LeCun @ylecun]

Before we reach Human-Level AI (HLAI), we will have to reach Cat-Level & Dog-Level AI.
We are nowhere near that.
We are still missing something big.
LLM's linguistic abilities notwithstanding.
A house cat has way more common sense and understanding of the world than any LLM.

[Or Biran @_orbiran]

@Plinz Agree that LLMs can improve on this significantly, but seems to me we always did lossy text compression as a side effect of text simplification (removal of certain particles, adjectives and adverbs, replacing complex specific terms with simpler approximations, etc)

[Or Biran @_orbiran]

@yoavgo Admittedly, what I proposed probably also falls short

[Or Biran @_orbiran]

@yoavgo I thought your point was "not all X are LMs", and that it might be clearer with an X that is a superset of LMs

[Or Biran @_orbiran]

@benajibayassine @paulisci It's brilliant :) we're so good at finding causes

[Or Biran @_orbiran]

@yoavgo @yuvalel 1st is the only one that doesn't have an implicit "try to..." ahead of it

[Or Biran @_orbiran]

@yoavgo @EigenHenry it's not the most common story here, but there's definitely a path where you apply for a master's, get into a research project and then continue with the same advisor to a phd. it's especially common if you have been away for more than a year after undergrad (e.g for work)