[HN Gopher] Time is encoded in the weights of finetuned language...
___________________________________________________________________
Time is encoded in the weights of finetuned language models
Author : convexstrictly
Score : 105 points
Date : 2023-12-24 21:53 UTC (1 days ago)
(HTM) web link (arxiv.org)
(TXT) w3m dump (arxiv.org)
| convexstrictly wrote:
| Twitter summary:
| https://twitter.com/ssgrn/status/1738256456250470853
|
| Github: https://github.com/KaiNylund/lm-weights-encode-time
| fnordpiglet wrote:
| Usable non musky version:
|
| https://nitter.net/ssgrn/status/1738256456250470853
| behnamoh wrote:
| the X version worked fine for me. I don't know what you want
| to achieve by posting a link to a third party website.
| jimberlage wrote:
| I'm not on Twitter, and found it valuable to see the
| replies!
| solardev wrote:
| Twitter's not very usable these days.
| CamperBob2 wrote:
| He _said_ he would sink the company...
| fnordpiglet wrote:
| I don't have a twitter account, and many don't. Twitter is
| slow, nitter is fast. Twitter has never reliably displayed
| threads for me. Nitter does.
| newZWhoDis wrote:
| >Twitter is slow
|
| It's faster than it's ever been, and seemingly without
| 85% of its staff. Says a lot
| fnordpiglet wrote:
| And yet nitter is still faster with only 47 contributors
| total. Says more.
| bobmaxup wrote:
| ... because it is essentially a caching proxy to twitter?
| Zambyte wrote:
| So is twitter.com. Unless they aren't using a CDN lol
| pests wrote:
| X's reduced user base might make X faster than ever.
| schaefer wrote:
| I think the advertiser boycott is a major contributor.
| Advertisements are slow...
| KTibow wrote:
| Twitter requires JS to work though.
| calamari4065 wrote:
| No, no it is not.
| hightrix wrote:
| Twitter doesn't show replies if you are not logged in. As
| others have said, I also don't have an account. So this
| link provides the full context. The twitter link only shows
| the post and no replies.
| bobmaxup wrote:
| Twitter doesn't even show most recent tweets from profiles
| unless you are logged in now. They show a summary of the
| profile's activity. Nitter is great if you don't have a
| Twitter account.
| electrondood wrote:
| All of the links were to a third party website.
| shzhdbi09gv8ioi wrote:
| x.com links requires being logged in to even read the
| thread.
| ParetoOptimal wrote:
| Allowing people without twitter accounts to view it.
|
| Allowing those who would otherwise avoid twitter to view
| the content.
| solardev wrote:
| Thanks!
| alephnan wrote:
| By time, they're talking about the writing style of a specific
| time period.
|
| Feels like a click bait title. Of course language model weights
| encode different writing styles. The fact that you can lift out a
| vector to stylize writing is also more interesting, but that's
| also nothing newly discovered here. It should be obvious that
| this is possible given that you can prompt ChatGPT to change its
| writing style.
| cush wrote:
| Why would it pertain only to writing style?
| n2d4 wrote:
| Besides what the sibling comment said, what's most interesting
| (imo) is that you can manipulate the vectors like that. The
| fact that you can average the vectors for January and March,
| and get better results for February, is pretty surprising to
| me.
| macleginn wrote:
| This also generalises: https://arxiv.org/abs/2302.04863
| jimbobthrowawy wrote:
| Generalizing vectors in generative models seems like an
| incredibly useful thing to know about, if you want to use them
| more effectively. Blew my mind when I saw someone demonstrate
| doing vector math on a GAN a couple years back to move an
| "input image" around the space of outputs.
|
| Maybe this could be useful for singling out post-LLM text and
| generating output that excludes it.
| k__ wrote:
| Interesting that writing style works, but other reflective
| actions don't.
|
| Like, "only use the the 2000 most common words of the English
| language" or "the response should be 500 words long".
| n2d4 wrote:
| It does work on other reflective actions, parent is just
| wrong; in the paper, they specifically run the experiment on
| a dataset of political affiliation over time
| mycall wrote:
| From the title, I was thinking "of course the neural network of
| the LLM is a [cause-effect] sequence of words" thus time is
| encoded in each connection.
| jiggawatts wrote:
| Sooo... if I'm reading this right, it's possible to force an AI
| into extrapolating into the future. As in, it'll answer as-if its
| training was based on data from future years.
|
| Obviously this isn't time travel, but more of a zeitgeist
| extrapolation.
|
| I would expect that if an AI was made to answer like it's from
| December 2024 it would talk a lot about the US election but it
| wouldn't know who won -- just that a "race is on."
|
| This could have actual utility: predicting trends, fads, new
| market opportunities, etc...
| behnamoh wrote:
| nah, this is not what they're talking about.
| dartos wrote:
| I don't think it'd be nearly as accurate as purpose built
| future predictors.
|
| LLMs aren't a silver bullet for everything.
| gmerc wrote:
| Ah, the bitter lesson teams it's ugly head
| electrondood wrote:
| > LLMs aren't a silver bullet for everything.
|
| Please explain this to my Product org.
| spacecadet wrote:
| lolol
| n2d4 wrote:
| Kind of. You still need some data from the "future" to
| extrapolate it: In the paper, they take an LLM finetuned on
| 2015 political affiliation data, and add to it the difference
| between 2020 and 2015 Twitter data, and show that the
| performance is better when the new model is asked about 2020
| political affiliation.
|
| So, the LLM still needs to know about 2020 from somewhere. In a
| way, you teach it about the task, then separately you teach it
| about 2020, and this method can combine that to make it solve
| the task for year 2020.
| habitue wrote:
| Maybe less zeitgeist, but it would be really interesting to see
| what extrapolating future writing styles are like.
| spacecadet wrote:
| Here ya go: lorizzle.nl
| throwaway81523 wrote:
| What about helixes of semi-precious stones?
| cwmoore wrote:
| I think I like time. Though spectral, indeterminate, presently a
| fixture, essential moments last forever but occur daily. Why
| would any network encode time if it were all just a crystal vase?
| ackbar03 wrote:
| what are you on?
| phito wrote:
| Crystal vase
| Vecr wrote:
| Don't worry about the crystal vase.
| tnecniv wrote:
| I gotta get me some of that
| haltist wrote:
| Because people have to publish papers, that's why.
| mjvmroz wrote:
| I think the general case is far more interesting than time
| specifically. There are cool functor/analogy ideas here.
| bkfh wrote:
| Can someone ELI5 this?
| airocker wrote:
| I think a more general way to think about it would be to add any
| data and reduce weight. For eg, if we want to create geography
| vectors, we would add all geography data to fine tune and then
| take a difference. Now add this to any other model with same
| architecture, and you have a geography capable llm.
| lproven wrote:
| I thought it was encoded as a helix of semi-precious stones, but
| perhaps I am misremembering.
___________________________________________________________________
(page generated 2023-12-25 23:02 UTC)