@kellogh@hachyderm.io
@kellogh@hachyderm.io avatar

kellogh

@kellogh@hachyderm.io

I'm a software engineer and sometimes manager. Currently #Raleigh but also #Seattle. Building ML platform for a healthcare startup. Previously, built an IoT platform for one of "those" companies.

Open source: dura, fossil, Jump-Location, Moq.AutoMock, others

Do I have other interests? No, but I do have kids and they have interests. I think that counts for something. I can braid hair and hunt unicorns!

I put the #rust in frustrate

He/Him

#metal #science #python

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

kellogh, to Metal
@kellogh@hachyderm.io avatar

Someone needs to do a modern cover of Weezer’s Say It Ain’t So

kellogh,
@kellogh@hachyderm.io avatar

honestly, it’s not far off as it is

kellogh, to random
@kellogh@hachyderm.io avatar

having a lazy day, slouching in a chair by a campfire fiddling on the ukelele

kellogh, to guitar
@kellogh@hachyderm.io avatar

A is just a viola on autotune

kellogh, to random
@kellogh@hachyderm.io avatar

One of my first jobs as a software engineer was supporting electical and mechanical engineers. I remember one of my coworkers refer to them as “real engineers”. I think about that a lot

kellogh,
@kellogh@hachyderm.io avatar

@troberts is that how you got into metal?

mnl, to random
@mnl@hachyderm.io avatar

6 months in, have you felt an increase of LLM-generated misinformation or bullshit in your day to day life? If so, where and what do you think were the incentives of doing so?

For me personally, while I am bit more sensitive to formulations like "On the other hand..." I don't feel like my bubble has changed. I see it in reviews, but I haven't trusted reviews in quite a while.

kellogh,
@kellogh@hachyderm.io avatar

@mnl I’ve become a lot more sensitive to how bad google is. I attribute it to having used ChatGPT and not being able to go back. But maybe it really is getting worse 🤷‍♂️

h_thoreson, to random
@h_thoreson@mastodon.world avatar

where is everyone getting weather info now that weather.com is fully enshittified? just trying to figure out the weekend outdoor concert situation and seems to be beyond the capability of what's on there now

kellogh,
@kellogh@hachyderm.io avatar

@h_thoreson the default iOS app is fast and clean, I usually use that, but I go to WillyWeather when I want more information, like tides

kellogh, to random
@kellogh@hachyderm.io avatar

hot take: i think it's a mistake to say that training on the output of other LLMs will lead to bad models (model collapse). no doubt it can and has led to model collapse, but i think it can also lead to superior models (and it has! e.g. orca)

i think of it like how extremely smooth plates are created by grinding together less smooth plates. if you train using several LLMs, you have the opportunity to take the strengths w/o the weaknesses of each

kellogh,
@kellogh@hachyderm.io avatar

i wrote a piece about unit testing last year that uses the same line of thinking. basically, why does unit testing even work? given that you're using imperfect code to test imperfect code... (well, you "grind" the tests against the prod code until they're both perfect)

https://timkellogg.me/blog/2022/04/11/three-plates

kellogh,
@kellogh@hachyderm.io avatar

idk, it's a very abstract way at looking at the issue, so i could be very far off, but i'm pretty sure model collapse isn't a certainty

kellogh,
@kellogh@hachyderm.io avatar

@michaelrowe01 orca uses outputs from one LLM and a totally different LLM (GPT4) to validate. Not really a GAN because you’re only evolving one side at a time. But honestly, GANs have stability issues for somewhat obvious reasons — you’re evolving the entire system at once, it’s like riding a unicycle across a tight rope

kellogh,
@kellogh@hachyderm.io avatar

@michaelrowe01 I think this is your GAN

  1. Fine-tune an LLM to extract knowledge graph tuples from text
  2. Fine-tune another to generate text from a KG
  3. Use #2 to generate a pre-training corpus with #1 to filter low quality samples
  4. Use #1 in the RL phase to score factual accuracy

https://arxiv.org/abs/2306.08302

Lazarou, to siliconvalleyhbo
@Lazarou@mastodon.social avatar

Just finished watching and it was spooky how contemporary the story lines were.
Everything from causing havoc to to Tech Bros in general, hilarious as well!

kellogh,
@kellogh@hachyderm.io avatar

@Lazarou is there a new season?

kellogh, to azure
@kellogh@hachyderm.io avatar

Oof, this is not a good look 🤦‍♂️ https://news.ycombinator.com/item?id=36347915

zlatiah, to reddit

https://kbin.social/m/RedditMigration/t/30318/Reddit-threatens-replacing-moderators-from-privated-subreddits-citing-vandalism-and

See the pic, sorry there's no alt text. Basic premise is that 's mod code of conduct is suggesting that it is feasible to remove "squatter" mods from their duty.

Implication is that Reddit might want to install their own mods to the indefinitely black-out subreddits

Make sure to grab some popcorn since this might get wild...?

kellogh,
@kellogh@hachyderm.io avatar

@zlatiah Reddit always was good for drama

kellogh, to ChatGPT
@kellogh@hachyderm.io avatar

imo lawyers who use to write court submissions aren’t “stupid”, they’re the victims of incomplete products with sharp edges and footguns

kellogh,
@kellogh@hachyderm.io avatar

OpenAI should do a better job of labeling their products as “bleeding edge”

kellogh, to random
@kellogh@hachyderm.io avatar

Does the Apache 2 license require that code is posted publicly? Can you privately share code under Apache 2?

kellogh,
@kellogh@hachyderm.io avatar

I’m trying to share code between 2 companies, neither cares enough to exchange money, but the providing side wants protections that Apache 2 provides. Unsure if that’s in bounds

kellogh,
@kellogh@hachyderm.io avatar

Seems like it is a thing you can do, but nobody talks about it

lzg, to random
@lzg@mastodon.social avatar

deleted_by_author

    •
    kellogh,
    @kellogh@hachyderm.io avatar

    @lzg yeah, we say “cut up”, which illustrates how uneasy we are with pineapples

    kellogh, to random
    @kellogh@hachyderm.io avatar

    trump: “i have nothing to hide”

    us: no please hide them, they’re national secrets and the whole reason you’re being indicted is bc you’re not hiding them

    my god this guy is a clown

    kellogh, to random
    @kellogh@hachyderm.io avatar

    imo this decade in tech is about use cases that are in the real world, not on a computer. let you interface with a computer without touching or looking at it. let’s you overlay compute on top of the real world. let’s you stand in someone else’s shoes and experience what they’re seeing. all of this needs a lot of work, but that’s where it’s going

    michaelrowe01, to apple
    @michaelrowe01@mstdn.social avatar

    how long before is used for support calls? Will really help in troubleshooting issues with devices. - one for @gamesatwork_biz

    kellogh,
    @kellogh@hachyderm.io avatar

    @michaelrowe01 i could see some killer applications in manufacturing and chemical engineering. a lot of what they do is incredibly visual. pics help a lot, but having the full 3D would be absolutely huge for helping someone diagnose issues that aren’t on a computer

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