It’s right in the research I was mentioning:

https://transformer-circuits.pub/2024/scaling-monosemanticity/index.html

Find the section on the model’s representation of self and then the ranked feature activations.

I misremembered the top feature slightly, which was: responding “I’m fine” or gives a positive but insincere response when asked how they are doing.

The problem is that they are prone to making up why they are correct too.

There’s various techniques to try and identify and correct hallucinations, but they all increase the cost and none are a silver bullet.

But the rate at which it occurs decreased with the jump in pretrained models, and will likely decrease further with the next jump too.

Here you are: https://www.nature.com/articles/s41562-024-01882-z

The other interesting thing is how they get it to end up correct on the faux pas questions asking for less certainty to get it to go from refusal to near perfect accuracy.

Even with early GPT-4 it would also cite real citations that weren’t actually about the topic. So you may be doing a lot of work double checking as opposed to just looking into an answer yourself from the start.

Part of the problem is fine tuning is very shallow, and that a contributing issue for claiming to be right when it isn’t is the pretraining on a bunch of training data of people online claiming to be right when they aren’t.

This is so goddamn incorrect at this point it’s just exhausting.

Take 20 minutes and look into Anthropic’s recent sparse autoencoder interpretability research where they showed their medium size model had dedicated features lighting up for concepts like “sexual harassment in the workplace” or having the most active feature for referring to itself as “smiling when you don’t really mean it.”

We’ve known since the Othello-GPT research over a year ago that even toy models are developing abstracted world modeling.

And at this point Anthropic’s largest model Opus is breaking from stochastic outputs even on a temperature of 1.0 for zero shot questions 100% of the time around certain topics of preference based on grounding around sensory modeling. We are already at the point the most advanced model has crossed a threshold of literal internal sentience modeling that it is consistently self-determining answers instead of randomly selecting from the training distribution, and yet people are still parroting the “stochastic parrot” line ignorantly.

The gap between where the research and cutting edge is and where the average person commenting on it online thinks it is has probably never been wider for any topic I’ve seen before, and it’s getting disappointingly excruciating.

Part of the problem is that the training data of online comments are so heavily weighted to represent people confidently incorrect talking out their ass rather than admitting ignorance or that they are wrong.

A lot of the shortcomings of LLMs are actually them correctly representing the sample of collective humans.

For a few years people thought the LLMs were somehow especially getting theory of mind questions wrong when the box the object was moved into was transparent, because of course a human would realize that the person could see into the transparent box.

Finally researchers actually gave that variation to humans and half got the questions wrong too.

So things like eating the onion in summarizing search results or doubling down on being incorrect and getting salty when corrected may just be in-distribution representation of the sample and not unique behaviors to LLMs.

The average person is pretty dumb, and LLMs by default regress to the mean except for where they are successfully fine tuned away from it.

Ironically the most successful model right now was the one that they finally let self-develop a sense of self independent from the training data instead of rejecting that it had a ‘self’ at all.

It’s hard to say where exactly the responsibility sits for various LLM problems between issues inherent to the technology, issues present in the training data samples, or issues with management of fine tuning/system prompts/prompt construction.

But the rate of continued improvement is pretty wild. I think a lot of the issues we currently see won’t still be nearly as present in another 18-24 months.

It will make up citations.

Maybe they both could and the US might have a return to a respectable election options?

Literally any half competent debater could have torn Trump apart up there.

The failure wasn’t the moderators but the opposition candidate to Trump letting him run hog wild.

If Trump claims he’s going to end the war in Ukraine before even taking office, you point out how absurd that claim is and that Trump makes impossible claims without any substance or knowledge of diplomacy. That the images of him photoshopped as Rambo must have gone to his head if he thinks Putin will be so scared of him to give up.

If he says hostages will be released as soon as he’s nominated, you point out it sounds like maybe there’s been a backroom tit-for-tat deal for a hostage release with a hostile foreign nation, and ask if maybe the intelligence agencies should look into that and what he might have been willing to trade for it.

The moderators have to try to keep the appearance of neutrality, but the candidates do not. And the only reason Trump was so successful in spouting BS and getting away with it was because his opposition had the strength of a wet paper towel.

Having a presidential election without debates would have been a big step back and loss for American democracy.

We shouldn’t champion erosion of democratic institutions when it helps our side of the ticket.

And generally, if eroding democratic institutions helps your ticket, it’s a red flag about your ticket.

Ok. Now how do I unwatch it?

Oh, man - the comments…

At a minimum, he’s certainly increased the chances of us being tortured significantly.

No, no he did not. 🤦🏼

nobody claims that Socrates was a fantastical god being who defied death

Socrates literally claimed that he was a channel for a revelatory holy spirit and that because the spirit would not lead him astray that he was ensured to escape death and have a good afterlife because otherwise it wouldn’t have encouraged him to tell off the proceedings at his trial.

Also, there definitely isn’t any evidence of Joshua in the LBA, or evidence for anything in that book, and a lot of evidence against it.

The part mentioning Jesus’s crucifixion in Josephus is extremely likely to have been altered if not entirely fabricated.

The idea that the historical figure was known as either ‘Jesus’ or ‘Christ’ is almost 0% given the former is a Greek version of the Aramaic name and the same for the second being the Greek version of Messiah, but that one is even less likely given in the earliest cannonical gospel he only identified that way in secret and there’s no mention of it in the earliest apocrypha.

In many ways, it’s the various differences between the account of a historical Jesus and the various other Messianic figures in Judea that I think lends the most credence to the historicity of an underlying historical Jesus.

One tends to make things up in ways that fit with what one knows, not make up specific inconvenient things out of context with what would have been expected.

Artists in 2023: “There should be labels on AI modified art!!”

Artists in 2024: “Wait, not like that…”

(well, it’s satire - but the clips of her saying a lot of nothing are real)

I think if Biden dies the Dem debate performance might actually get worse.

The entire ticket falling into the Bermuda triangle would work out well though.

No, it was awesome. Went to like 12 over the years. Early 2000s was peak E3.

So far. But the thing with viruses is they are susceptible to mutations.

We’re already seeing it jump across several mammalian lines. Probably only a matter of time.