kellogh

kellogh, 1 month ago to Deathcore

ngl i'm deeply disappointed with #StableAudio. i can't get it to generate decent #deathcore or any kind of #metal. none of it is even remotely passable

kellogh, 1 month ago to ChatGPT

the best use case for #chatgpt is to paste in errors and have it rephrase it in a way that doesn’t look like an error, thereby bypassing the dumbuglobus (the area of your brain responsible for dumbness) and allowing you to see what is wrong

kellogh, 1 month ago to random

i don’t understand how people see the xz incident and conclude that open source is insecure. That level of social engineering could easily have worked on a company as well, but it was detected because it was open source. All other mechanisms failed, and it was just some random guy poking around that discovered it. That kind of scrutiny doesn’t happen on closed source systems

kellogh, 1 month ago to random

@ivory can we have a “go to instance” feature?

kellogh, 1 month ago to random

i had an old grumpy coworker who oft repeated, “dependencies are like children. You have high hopes for them, but in the end they just disappoint you”

kellogh, 1 month ago to random

this month is going by fast, i can’t believe it’s already Feb 62nd

kellogh, 1 month ago to random

you know shits real when someone breaks out the while loop

kellogh, 1 month ago to random

yesterday @hgrsd announced his tool, drivel, that describes the schema of JSON and I. JUST. CAN'T. GET. OVER. IT.

such a simple thing, but it's incredibly useful for hacking on scripts alongside curl

It goes beyond data types and also tells you if a field is optional, a datetime string, the numeric range in the sample, etc.

https://github.com/hgrsd/drivel

kellogh, 1 month ago to random

Favorite part of church today: a little kid crying in the back made everyone acutely aware that there was a child in the audience while the pastor was describing the death of Jesus in graphic detail. The pastor, upon seeing discomfort on faces, interprets this as he's striking a nerve, so he doubles down with extra graphic descriptions. The flesh ripping out of Jesus' back as the whip pulls back, etc.. The kid cries more, we all cringe even more, the pastor triples down, quadruples down, …

kellogh, 1 month ago to random

so, who was intended to come through said xz backdoor? who commissioned it?

kellogh, 1 month ago to random

i’m hoping @simon is one of the ten developers that get access to this, so we can hear all about it https://www.theverge.com/2024/3/29/24115701/openai-voice-generation-ai-model

kellogh, 1 month ago to random

how often do you search your house for dead space that might have a secret room hidden within?

kellogh, 1 month ago to llm

Automatic refutation of misinformation.

A new paper offers a system to correct misinformation using an #LLM. The approach seems solid, and the results seem strong. I haven’t dug in deep yet, but I’m hopeful about this one

https://arxiv.org/abs/2403.11169 #LLMs #AI

kellogh, 1 month ago to random

are rainbows real?

kellogh, 1 month ago to random

i remember when smart devices were all the rage and it seemed like the best application of IoT was reordering groceries, e.g. measuring exactly when you used the last egg, and then Amazon offered “subscriptions” on a regular time interval and all that started to look pretty dumb

kellogh, 1 month ago (edited 1 month ago) to rust

ngl postfix match statements are probably the sexiest #rustlang feature i’ve seen in a while https://doc.rust-lang.org/nightly/unstable-book/language-features/postfix-match.html #rust

kellogh, 1 month ago to random

my daughter is learning to read & write. she seems to do better with rewarding activities. comics are the clear winner for reading, every completed frame is a pile of joy for her. for writing, writing prompts for Dall-E has that quick exciting fix. she also likes writing picture books, although i prefer Dall-E prompts because there’s a baseline level of spelling she has to cross in order to get results

kellogh, 1 month ago to LLMs

i’m skeptical of this paper. It’s hard enough to decide on a good evaluation metric, or to decide if the right one was chosen. This paper rides on the idea that you can just switch to a new metric and get different results, which yeah, that’s a well known phenomenon called bullshit https://arxiv.org/abs/2304.15004 #LLMs

kellogh, 2 months ago to random

there’s a set of technologies that i don’t want to see innovation. incremental improvement is fine, i just don’t think my head can take a new disruptive approach

• networking
• VMs
• version control
• programming languages
• data formats

probably some others. stuff that works fine and i don’t care enough to invest more time into learning something new

kellogh, 2 months ago to LLMs

I keep seeing this link posted with “gotcha!” comments, like “see, #LLMs can be trained without copyrighted data”. Honestly, I’d love to believe that’s true, but it’s still detached from reality. This dataset is only 500B words and claims to be the largest, whereas, e.g. Falcon used 2T and it hasn’t been competitive for 6-12 months https://huggingface.co/blog/Pclanglais/common-corpus

kellogh, 2 months ago to 3DPrinting

weaponize your #3dprinter!

is there someone you truly hate? do they have small children?

if you answered ‘yes’ to both of those questions, use your 3d printer to create several recorder flutes and give them to their children as gifts! the kids will love you and your message will be received

#3dprint #parenting

kellogh, 2 months ago to random

they got all the positive stuff right about me, but oddly they’re far off on all the negatives 😂
https://noc.social/@todayilearned/112128279468396023

kellogh, 2 months ago to ai

a critical skill in the age of #AI is knowing if you’re in a critical situation that demands accuracy or the other kind, where you just need a guess to point you in the right direction. You’ll always have huge negative impacts (e.g. quality) from AI if you don’t know when their usage is appropriate.

i like this phrasing, “a drunk guy at a bar”. He could be right, but maybe not. How much are you risking by trusting him?

https://attractive.space/@Setok/112123708506055669 via @Setok

kellogh, 2 months ago to ai

Fantastic paper! Detecting #AI generated text is hard. We’ve had disappointing results so far. So, the obvious (well, should be obvious) thing to do is to tackle the problem at a higher level, e.g. at the journal level

• 10M times more efficient
• 3.4-4.6 times more accurate

https://arxiv.org/abs/2403.07183
#LLMs #Science #GenAI

kellogh, 2 months ago to random

RethinkDB