4,153 karma · joined March 13, 2019
sometimes makes weird machine music. sometimes makes machines do weird things. sometimes makes music machines do weird things.
Yes. They do. You are absolutely right about that.
But the model architecture doesn't change as a result of the training process. A piston doesn't suddenly turn into a digital watch as a result of tuning an engine. Similarly, the transformer part of a GPT model doesn't suddenly turn into something else as a result of optimizing a loss function.
---
i've got other stuff to do, so i'm stopping here.
To understand how an engine works, it's important to understand what a piston does as part of the engine.
again, the finished product wasn't what was discussed by GP, and you didn't clarify that you were switching to discussing RL (which is still probabilistic btw)
clever procedures on top of the base transformer architecture.
i used simplified words/phrases to summarise the same thing you two were saying (the intent being: here's a version that may be digestible when discussing with others).
apparently that means i'm wrong though, no idea why because it seems you've decided to be dismissive rather than constructively elaborate on why this simplified and digestible version might be wrong :shrug:
> Autoregressive LLMs generate tokens one at a time, disputing this is just plain wrong.
next-token prediction i.e. the bit built during pre-training.
at no point in your reply to GP did you specify that you were referring to post-training. respectfully, it seems like this one is on you pal :shrug:
> GPT-2 didn't use any reinforcement learning and is often given as a toy example. That release was 2019 and models now go through a various phases of training with different objective functions and optimizers.
yeah. so? the toy example works for pre-training. see above.
Probability(y | x)
> Prediction implies there is some "truth" or event or something that you can test against
there absolutely is a ground truth during training. the core predict-the-next-most-likely-token part of an LLM has a ground truth next-token. that's why you don't end up with generated text like: fish spurious send cattle chocolate phone happy meaning ball orange board canada.
> optimizes to predict the next token in training data
that is the optimization goal in training the next-most-likely-token core of an LLM, it basically translates to maximise the likelihood of predicting the next token x_i given the previous tokens
L(θ) = −log Π^n_{i=1} f_θ(x_i | x1, ..., x_{i−1})
(edit: sorry for the ADHD edits)
> The agents continue to poke around on DSEWiki. A few hours after they find the site, they start probing it for cross-site scripting (XSS) vulnerabilities. [...] The agent swarm starts testing whether they can execute JavaScript that they embed into the search page, and continue to do this for a few days
either the agents were doing free security testing for the site and “forgot” to submit a report, or they were trying XSS to gain something they didn’t have permission/authorization for.
also
> Hijacking: To take control of (something) without permission or authorization and use it for one's own purposes.
a mod had to go through and mass delete a bunch of pages that didn't belong on the site. no-one from the wiki site gave the agents permission to use their site as a message board. hijacking isn't being used here in the sense of "gained admin privileges to run crypto scripts" -- there are multiple ways to use a word.
> a right to express their opinion
...
1a. Illegal immigration has nothing to do with visa based, legal immigration. illegal immigration is the type of immigration people are upset about here in the UK. most people have no problem with legal migration.
1b. anyone claiming the UK’s legal migration is any sort of problem right now, or in the past, should probably not be listened to in any way shape or form for anything they have to say on the matter.
2. a lot of the folks who have illegally entered the country end up asking for asylum, and by law their claims must be processed
3. the dublin convention states something along the lines of “asylum requests must be processed in the first EU country the person entered into”, ie it’s possible to remove people claiming asylum if they initially arrived in a different EU member state country
4. most people get to the uk having traveled through EU member states, because geography.
5. we left the EU, we cannot do 3 anymore. so we have a bunch of people stuck in hotels waiting for asylum claims to be processed.
illegal immigration got worse because we left the eu.
sys-admin-ing gitlab can be a bit of a maze of web menus tho.
i miss seeing clever and thrifty solutions to weird problems. hell, i miss people just doing plain weird shit and posting that.
a first look through the comments on some first page threads and they seem far more thoughtful and considered. it generally seems like a nicer commenting environment that might actually do me some good moving forward.
https://docs.github.com/en/site-policy/github-terms/github-t...
> Short version: We will not be liable for damages or losses arising from your use or inability to use the service or otherwise arising under this agreement. Please read this section carefully; it limits our obligations to you.
https://docs.github.com/en/site-policy/github-terms/github-t...
otherwise you can still prompt your agent things, but any output falls under public domain.
since this hype wave started some people have been saying don’t use these tools in an environment where you’re dealing with proprietary IP. did anyone listen? no. because shiny.
same thing applies to sampling in music. if i change enough of a sample as to make it unrecognizable, then i don't need to worry about registering/licensing the sample.
> Neither mere prompting nor the selection between several AI suggestions is sufficient as a human creative contribution.
which reads like you actually have to contribute by modifying the output from the system, i.e. prompts/inputs to the system do not count as a human contribution.
not mine, no license, only chinese language on the eventual page, but has the beginnings of something beautiful
then again i live in the UK where we actually have workers rights.
this sounds like the WFH people might be treated the same way as the in-office people. as in you're retaining an in-office culture but people can choose to WFH. that is not the same thing as being a remote working company. the culture is completely different in an org that successfully does remote working.
as a functional example on the social side, people booking in time to just hang out and chat. even if it's work chat but just more relaxed chat. build in some social element beyond just treating it as cogs in a 9-5 machine. add an extra 30 minutes to a team update call to just have a chat with everyone -- including folks in the office.
edit -- to be clear, managers need to be doing that ^
building in an extra 30 minutes on team catchup calls just to hang out at the end. usually end up talking about some work stuff, but in a more relaxed/social way.
demo/learning sessions. how to use databricks, how to do X, how X works, this cool thing we built.
because, as someone else i replied to in the thread points out, miserable employees perform worse.
remote work was a blessing for me. but turned out the ceo didn't want us remote working.
i tried repeatedly to communicate what i needed to do my best with my boss and the ceo. but the ceo fought me on all of it.
amazing
gitlab ultimate (minus support) for free, so long as licenses for all projects are OSI approved.