Gergő Bocsárdi

Exploring some of the stores still open at the Lloyd Center #portland #mall TikTok video by Here is Oregon

This is unexpectedly delightful: a dying mall in Portland Oregon is filling up with a beautiful collection of weird and interesting businesses because the rent is now affordable for them www.tiktok.com/@hereisorego...

memes aside, with the lack of algorithm to me Bsky feels decidedly closer to my RSS feed than any sort of shortform brainrot

now it’s time for us to start betting on what you’ll get

@ssp.sh saw the convo on replacing raycast on arch, have you seen this one?😃

I forgot that I have a workshop proposal deadline tomorrow, so today I frantically wrote 2800 words😃

I guess that paper draft that I have been procrastinating on is now done💀 nothing gets me working like a deadline

old story in rob's head: ahhhhhhh i'm feeling super overwhelmed by my to do list, so i should prob run away and do a bunch of recreational numbing in the name of "self care"

new story: the more i do the better i feel

Any best practices for prompting the deep research?

I tried for the first time a few days ago, but it kept getting lost in “collecting additional data” endless web browse spirals, with 4-500 web pages opened, to complete the task of collecting submission guidlines of a list of academic journals😅

YouTube video by Gergő Bocsárdi llm brainrot is good, actually

Today I tried to reconcile some rather ambivalent thoughts on LLMs, thinking, and work😃

Insanely cool setup!

But can it run Doom?😃

I realized I mostly just needed a formatter, so the injected language formatting feature of conform.nvim took care of this!

I just replicated a 40 minute processing pipeline from a spark cluster in 33 minutes on my macbook😃
I have to say I now get the hype

Or in the middle the @simonwillison.net -ism of “give me 10 ways to solve this problem” and then pick the one you like the most to code down!

superwhisper AI powered voice to text for macOS

I was looking for a dictation tool on my iphone that’s better than the built-in, but still not a cloud, and lo and behold, there is one!

I just tried superwhisper’s local model, and it provided me with flawless transcripts in both english and hungarian, all made locally.

yap walks here we come!!

Hahaha not exactly, though the source is all the same😃

I’m using Quarto’s website framework! This was basically anything I write (papers, blogposts, presentations) come from a single unified source of QMD files, just rendered differently

Found that it reduces a lot of friction around formats!

Just simple stuff I think, my site is a good example, it’s not even a VPS, just hosted storage and I push the generated files over via FTP😃

Quarto’s revealjs is my go-to!

Only caveat with revealjs is that you need a Chrome to export nice PDFs (Safari and Firefox usually mangle the aspect ratios for me)

I certainly agree that taking steps to reduce environmental impact is necessary, but I think the main point is more that ChatGPT usage isn’t a Big New High-Impact Thing, rather than a drop in an already pretty large bucket, which itself deserves more attention.

Hey #duckdb friends!

I am endlesy fascinated by the tool, and I’m yet to try it, but the one thing that I’m worried about is the lack of autocomplete during my data wrangling in Python.

Is there some magic way to get SQL completion inside Python? Some Neovim magic perhaps?

Or is this a no-issue?

A cheat sheet for why using ChatGPT is not bad for the environment The numbers clearly show this is a pointless distraction for the climate movement

also a great piece by @andymasley.bsky.social, with lots of great visuals!

Congrats! Well deserved, amazing piece, I share it with vim-curious people often😃

Neural correlates of interspecies perspective taking in the post-mortem Atlantic Salmon: an argument for multiple comparisons correction

Mandatory dead salmon mention when talking about brain scans😃

perhaps even more so picking a market you enjoy being in!

i can’t imagine building something for a crowd you don’t feel a kinship with

Yes, vision LLMs are pretty good geo-guessers (I had wondered about this) easily beating humans

Bigger LLMs are more accurate and data leakage (having seen the same picture in training) does not seem to be a big problem. The models are better at urban & more developed areas arxiv.org/pdf/2502.11163

Firefox 136.2 (51977) on IOS 18.3.2 is still struggling (“NSURLErrorDomain”), but the Safari version (same as IOS) resolved after about 30 minutes yesterday!

The 100% CI Always Confident. Sometimes credible.

I’d love to read it, but the link (and the whole site btw) is throwing a “can’t connect to server” error, both on safari and firefox on both iphone and mac, not sure if it’s on my end or not!

Thanks brother, I actually got running again two weeks ago, but my feet and hips didn’t exactly appreciate the extra mass they have to carry since the last time I ran, so now I’ll go nice and conservative, gradual increases😃

I was looking for a free, simple Couch to 5K app to get back into running, and I was disappointed to see that all of them tried to shove a subscription fee down my throat

Then I found “Just Run”. Indie developer, free and ad-free, the only IAP is “support Gary who makes the app”. Never paid quicker

I am so tired of rewriting papers 😭😭

#PhDSky

starting the day with “what would done look like for today” is such an underrated move

it really gave my todolist a different tone, shoutout @austinkleon.bsky.social and @oliverburkeman.com for the inspiration

YouTube video by Gergő Bocsárdi Local LLMs - Why, What, How?

This week I gave a presentation on how to run LLMs locally!

We touched on models, tools, and technical stuff (like quantisation), I designed this in a way that it gives a good baseline to explore afterwards.

Enjoy🫶
#llm #ai

Surely adding AI to printers is somehow against the Geneva Conventions