@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 random
@kellogh@hachyderm.io avatar

my 3yo exhibits this thing i call stochastic walking, where her exact walking path is wholly unpredictable, except that (1) it’s always in my way and (2) if i step into her probabilistic path cloud, her actual path will definitely converge with mine with absolute certainty

kellogh, to random
@kellogh@hachyderm.io avatar

i’ve seen a lot in my career, but until now, i’ve never seen religious fervor like there is against AI. there’s a lot of religion around AI in most directions, but crypto had that too

what’s crazy about this phase is that if people took time to understand what accelationists see, they wouldn’t come to the same conclusions, and we’d have far more interesting and productive conversations about it

but instead we’re caught in tired arguments about “it’s just linear algebra”

kellogh, to LLMs
@kellogh@hachyderm.io avatar

one thing i love about #LLMs is asking it “how tf do i do X” and it responds with 5 ideas, four of which are terrible but one is far better than anything i’d thought of. or their all terrible but one makes me realize i’ve been thinking about the problem wrong

kellogh,
@kellogh@hachyderm.io avatar

also, #LLMs cause me to think a lot about the multifaceted nature of intelligence. we used to over-weight language skill, but now that LLMs have that in spades, it’s apparent that there’s more going on

for example, spontaneity. if it had an ounce of sponteity, it could suggest approaching the problem differently

annmlipton, to random
@annmlipton@esq.social avatar

Always be closing

kellogh,
@kellogh@hachyderm.io avatar

@annmlipton @andrew having talked to a few people about this, the charges actually were unclear — if you consume Fox News. speak it into existence, i guess

futurebird, to random
@futurebird@sauropods.win avatar

I've largely dismissed the strain of AI alarmism based on the notion the a computer will be so smart that the danger it poses to humanity is outsmarting us.

There are real dangers in AI, most of them relate to people using these technologies in improper ways due to having a poor understanding of what they really are... and most important the exploitation & degradation of the human body of knowledge creativity represented by publicly available digital information. 1/

kellogh,
@kellogh@hachyderm.io avatar

@futurebird i also tend not to believe it, but i’m skeptical of myself because 100% of the top AI experts believe in the scenario where AI becomes that intelligent

idk, we’ve been talking about anti-science regarding climate change, are we doing the same thing with AI?

kellogh,
@kellogh@hachyderm.io avatar

@futurebird every single one of the deep learning “pillars”, the experts that are working on AI — Hinton, LaCunn, Ng, Bengio, etc.

all of the disagreement with that viewpoint are from people who are experts in other fields, like linguistics — Bender, Chomsky, Marcus, etc.

kellogh,
@kellogh@hachyderm.io avatar

@bennett @futurebird what breakthrough discoveries has he made? i’m not aware of any. the question isn’t how they function now, but how they’ll function in the future. for that, i’d trust the people making breakthrough discoveries, the scientists not the engineers

kellogh,
@kellogh@hachyderm.io avatar

@bennett i agree, disparate perspectives are good. however, the thing i can’t square with is the unanimity of the experts in the thing we’re trying to make predictions about.

scientific consensus isn’t always right the first time, but it’s a pretty strong signal

further, similar to climate change, there are these scaling laws that predict capabilities of LLMs. they’ve been right so far, doesn’t mean they always will be, but it’s another strong testable signal

kellogh,
@kellogh@hachyderm.io avatar

@dnavinci @bennett i’m always suspicious of the “you’re stupid if you even try to think that” argument. it smells like religion.

on this topic, the smartest scientific minds believe it’s true. writing them off as “incredulous” seems telling

scientific consensus isn’t everything, it’s been wrong before. but they do have a testable hypothesis, and they’re testing it, and so far it’s held

kellogh,
@kellogh@hachyderm.io avatar

@dnavinci the hypothesis is that performance increases predictably with increases in data size, model size and compute. it’s held since 2020, and hallucinations have decreased at a predictable rate

if i’m not an expert (i never claimed to be, so you’re right), why are you demanding to see my credentials? that makes zero sense outside simply being an asshole

kellogh, to random
@kellogh@hachyderm.io avatar

talking to my conservative family, they’re completely baffled that there’s an election interference angle to the . my dad consumes a metric ton of news media, had no idea

carnage4life, to random
@carnage4life@mas.to avatar

I’d watch this as a reality show.

kellogh,
@kellogh@hachyderm.io avatar

@carnage4life do we really need MORE prison gangs?

AlexJimenez, to ai
@AlexJimenez@mas.to avatar

Inside Anthropic, the Company Betting That Safety Can Be a Winning

https://time.com/6980000/anthropic/

kellogh,
@kellogh@hachyderm.io avatar

@AlexJimenez there’s some good quotes in there. i never heard the phrase that ML models are “grown” not designed. i like that analogy

jimfl, to random
@jimfl@hachyderm.io avatar

TIL that Binance does not rhyme with Beyoncé

kellogh,
@kellogh@hachyderm.io avatar

@jimfl lol, it should though

deech, to random
@deech@mastodon.social avatar

Feel like streaming services should lean into how most people only sign up for one or two shows and make it super easy to suspend and restart their subscription.

kellogh,
@kellogh@hachyderm.io avatar

@deech daily subscription billing period

kellogh, to random
@kellogh@hachyderm.io avatar

new rule, in order to have “lab” or “labs” in your company name, you must register as a public benefit corporation

kellogh,
@kellogh@hachyderm.io avatar

@LChoshen yeah! that too

mekkaokereke, (edited ) to random
@mekkaokereke@hachyderm.io avatar

Some of my friends that are convicted felons are 100% innocent of what they were accused and convicted of. Some of my friends that are convicted felons are guilty, and absolutely did what they were accused of. I don't make fun of people for being convicted felons, and I don't ostracize them. Instead, I don't let bad people around me. My definition of bad focuses on the choices that a person made in the past, and the choices that they make now. Some of the worst people, will never be felons.

1/N

kellogh,
@kellogh@hachyderm.io avatar

@mekkaokereke yes, but when confronted with a person who has strong negative preconceived notions of felons, i will definitely make a strong association between Trump and felony, and use those biases against Trump.

OTOH, when they try to downplay felonies in order to excuse Trump, i will also definitely use that as an opportunity to educate them on the racial divide regarding felony convictions by agreeing heartily with them

kellogh, to random
@kellogh@hachyderm.io avatar

whoah, trump is convicted? i guess we can have nice things

sue, to random
@sue@glasgow.social avatar

I'd like back all the hours I've spent over the years trying figure out the correct incantations of quotation marks and escape characters to get computers to do what I want with some text.

kellogh,
@kellogh@hachyderm.io avatar

@sue yeah, that too. LLMs are very good at that. i use the llm CLI tool for better tool integration (e.g. use it from vim). they’re great at copying formatting. if you give it one or two examples it’ll usually get perfect accuracy, and often pickup on edge cases you didn’t think of

kellogh,
@kellogh@hachyderm.io avatar

@sue in general my flow is

Thought: “damnit this sucks, i shouldn’t have to do this”

Action: “let’s see if an LLM works here”

kellogh,
@kellogh@hachyderm.io avatar

@nicholas_saunders @sue yo vim is great. it’s a program made for editing text. it’s so close to the CLI that you can pipe arbitrary command output to it. you can shell out to arbitrary programs to process highlited text. i tend to use the macros to automate some one-off repetitive process

kellogh,
@kellogh@hachyderm.io avatar

@nicholas_saunders @sue pretty good. LLMs are designed around sequences of tokens (i.e. language tasks), so despite the weight, they’re unparalleled as a general purpose text manipulation tool

kellogh,
@kellogh@hachyderm.io avatar

@nicholas_saunders @sue full disclosure, i have no idea what IDE even means at this point. it used to be clear, back when it was IntelliJ, VS, eclipse, netbeans, etc., and when text editors just edited text and didn’t have evolved plugin ecosystems. today, it’s hard to draw a line between them

i typically use VSCode in vim mode, as well as neovim in the terminal if in need to edit something there. nowadays, vim is just for arbitrary text, maybe bash, most programming is in vscode

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