@redscroll@hachyderm.io
@redscroll@hachyderm.io avatar

redscroll

@redscroll@hachyderm.io

(he/him)

Data Scientist |

I work on LLMs (I promise I'm not obnoxious about it) |

#AuDHD |

Trying to learn more about #smallweb and #smalltech |

Currently learning Golang |

Bookwyrm: @redscroll

Lemmy: https://lemmy.ml/u/redscroll

This profile is from a federated server and may be incomplete. Browse more on the original instance.

f1337, to mentalhealth
@f1337@hachyderm.io avatar

Things I learned at “ camp”:

  • is a wound to the body, mind, & spirit.
  • Trauma, while caused by past events, actually lives in the body, in the present.
  • Healing from trauma does not require re-experiencing the event. We can heal from trauma that we don’t remember.
  • A multimodal approach to trauma recovery is ideal. Not every mode works for every person.
  • The behavioral model for mental health is fundamentally broken pseudoscience.

@actuallyautistic

redscroll,
@redscroll@hachyderm.io avatar

@f1337 @actuallyautistic can you elaborate on the last point a little? I think I know what it means, but not entirely sure

kellogh, to llm
@kellogh@hachyderm.io avatar

i low key don't want to see a big jump in #LLM or #AI capabilities anytime soon. rn they're capable enough that my mom wants to use them, but bad enough that even she has an intuitive sense for when they're wrong

that's how you build "AIQ", the skill of using it. Lots of people toying with them, to feel out their capabilities and limitations

redscroll,
@redscroll@hachyderm.io avatar

@kellogh I'd like to see how the 15T tokens looks like on a 3B model.

kellogh, to llm
@kellogh@hachyderm.io avatar

i missed this develoment — the top on the open LLM leaderboard right now is a 10.7B model. All the top spots are variations of it. that would fit into most commodity GPUs easily. It's beating out all the 34B & 70B models https://arxiv.org/abs/2312.15166

redscroll,
@redscroll@hachyderm.io avatar

@kellogh I'll have to try it out tomorrow in prod lol

c0dec0dec0de, to LLMs
@c0dec0dec0de@hachyderm.io avatar

There’s an LLM called Grok? What the fuck. Seriously. Tech bros understand the absolute most obvious thing from your favorite piece of science fiction, please.
The meaning of grok is in the plain text! We’re not even talking about obvious metaphor, it’s just a made-up word for truly deep understanding and you’re going to name your idiot, fabulist, word generator after it?!

redscroll,
@redscroll@hachyderm.io avatar

@c0dec0dec0de it came from the compromised-land peoples, don't put that on us

redscroll,
@redscroll@hachyderm.io avatar

@c0dec0dec0de like it was built by Musk's team. "Compromised land" is a play on promised land, which is what Twitter was expected to be until the whole EM thing. All to say that the people who built it [and named it] are not representative at all of the ppl working in the LLM space and their work. I would just suggest to not let that model influence how you see this field

mariyadelano, to internet
@mariyadelano@hachyderm.io avatar

Friends who are older than me: do you have any good stories of encountering spam on the early internet? (PLEASE BOOST FOR REACH)

90s or pre-2005 please!

I might end up quoting your example in an article 😊

#AskFedi #AskMastodon #internet #request #recommendations

redscroll,
@redscroll@hachyderm.io avatar

@mariyadelano I remember writing a paper about it and the first spam email was so chaotic it like shutdown the servers of a few companies

redscroll,
@redscroll@hachyderm.io avatar

@mariyadelano I do, let me look it up!

redscroll,
@redscroll@hachyderm.io avatar

@mariyadelano I have it! How would you like for me to send it to you?

cyberlyra, to linux
@cyberlyra@hachyderm.io avatar

@protonmail Hello! When will your tools (i.e. cloud storage, calendar, etc) be available for clients, not just Windoze? Thank you:D

redscroll,
@redscroll@hachyderm.io avatar

@cyberlyra @protonmail the only one that I could get on Linux is the VPN and even that is a bit buggy

kellogh, to LLMs
@kellogh@hachyderm.io avatar

Thought:

  1. aren't trained on data after 2021

  2. Most writing becomes done by LLMs

Does that mean our language will start becoming frozen in time? Slower change?

redscroll,
@redscroll@hachyderm.io avatar

@kellogh do you.. do you think LLMs just stopped being trained after 2021?

redscroll,
@redscroll@hachyderm.io avatar

@kellogh data quality is a concern as more synthetic data is being generated by the day, but humans can't stop providing "input". We need to write, to talk, to create. We can't stop doing that anymore than we can stop breathing.

cyberlyra, to random
@cyberlyra@hachyderm.io avatar

Everyone is talking about Copilot on GitHub like this is new, but I figured it was pretty obvious from the beginning that’s why they acquired it. So what did people think of the acquisition if they didn’t think it was to use all the data to generate an assistive AI?is this honestly coming out of left field? Or is this just the moment where people are begrudgingly acknowledging the situation?

redscroll,
@redscroll@hachyderm.io avatar

@cyberlyra I feel like this meme is relevant here. It's going to be painful how many times people will get screwed over again and again until we learn

kellogh, to LLMs
@kellogh@hachyderm.io avatar

i wish i knew more about comparing . anyone have resources? one thing i’ve wondered is how to convert an embedding from a “point” to an “area” or “volume”. e.g. an embedding of a 5 paragraph essay will occupy a single point in embedding space, but if you broke it down (e.g. by paragraph), there would be several points and the whole would presumably be at the center. is there a way to trace the full space a text occupies in space?

redscroll,
@redscroll@hachyderm.io avatar

@kellogh usually embedding models are going to have a max token length like 500 tokens, unless you’re using longformer or something like that.

kellogh, to LLMs
@kellogh@hachyderm.io avatar

are just compilers of databases

redscroll,
@redscroll@hachyderm.io avatar

@kellogh mmm nah, a degree too simplistic, but I get the vibe

mariyadelano, to random
@mariyadelano@hachyderm.io avatar

Finalizing a pretty scary piece of content. This is weeks and months overdue, and in many ways the culmination of a lot of my work so far in 2023.

I’m about to lay my heart and story bare, backed by data and a lot of receipts.

Stay tuned. Trying not to get too paralyzed by fear of what I’m about to publish 😅

redscroll,
@redscroll@hachyderm.io avatar

@mariyadelano excited to finish reading it, but just wanted to give a quick heads up first:

emocoder, to random
@emocoder@hachyderm.io avatar

I love so much I even created a account to highlight quotes. Let’s be friends there: Check out my profile on Goodreads!
https://www.goodreads.com/user/show/168959653

redscroll,
@redscroll@hachyderm.io avatar

@emocoder check out Bookwyrm.social!

geekruthie, to Matrix

Great big open-source projects

I was talking to a friend today, and suggested using for a community we're both in, instead of Slack. , you know, and open source, and actually has tools for enforcing standards of conduct. They asked, "Is there a server written in ?"

There is not. The reference implementation is in Python, and it's massive. Others have been written in Go and Rust and even C++. But there's no reason it shouldn't be doable in @Perl ...

redscroll,
@redscroll@hachyderm.io avatar
  • All
  • Subscribed
  • Moderated
  • Favorites
  • JUstTest
  • Leos
  • DreamBathrooms
  • thenastyranch
  • magazineikmin
  • osvaldo12
  • ethstaker
  • Youngstown
  • mdbf
  • slotface
  • rosin
  • ngwrru68w68
  • kavyap
  • cubers
  • anitta
  • modclub
  • InstantRegret
  • khanakhh
  • GTA5RPClips
  • everett
  • Durango
  • tacticalgear
  • provamag3
  • tester
  • cisconetworking
  • normalnudes
  • megavids
  • lostlight
  • All magazines