Gergő Bocsárdi reposted
Gergő Bocsárdi
Gergő Bocsárdi
@agbocsardi.com
where i write: agbocsardi.com where i ramble: https://www.youtube.com/@agbocsardi Big on: 📊PhD: networks, strategy, ML ☕️Coffee: yirgacheffe, v60, co-ferment 📸Photos: fuji, edc, analog Third year PhD, 🇭🇺in🇳🇱
199 followers430 following332 posts
Gergő Bocsárdi reposted
Gergő Bocsárdi reposted
A glowing stellar nursery 👶⭐ Webb’s infrared sensitivity has revealed the protostars & stars of NGC 7129, a stellar nursery that lies 3,300 light-years from Earth. In the centre, lies the massive star LkH(alpha) 234, which has carved into its cloudy surroundings 👉 www.esa.int/ESA_Multimed... 🔭 🧪
Gergő Bocsárdi reposted
hey @pckt.blog! I've been loving the cozy editor, but I noticed that the standard.site "description" field get populated with the whole post's body, leading to rather lengthy descriptions :D Is this something on the roadmap to address, maybe like a separate description entry box? Thanks a lot <3
Gergő Bocsárdi reposted
If the tickets can be scalped, they aren’t being sold for their market value. The scalped price *is* the market price, theaters/concerts/etc are just generally unable to reliably price their products there (for a variety of reasons)
This is a perfect example of how Scalper-Friendly markets actl make for a more aristocratic society. Scalpers buy up the market-value tickets quickly, then price-gate them so that only the wealthy can afford to buy.
Gergő Bocsárdi reposted
this is half of management science theory building lmao
There's a certain satisfaction that comes from taking three things that are similar, determining the dimensions that define their differences, putting them into a 2x2 grid, and discovering a secret, likely daft, fourth thing that is distinct from the other three.
Gergő Bocsárdi reposted
Gergő Bocsárdi reposted
Gergő Bocsárdi reposted
Gergő Bocsárdi reposted
Gergő Bocsárdi reposted
Gergő Bocsárdi reposted
We already have superintelligences that have taken over the planet: we call them organizations. Example from: www.noahpinion.blog/p/what-will-...
"slop is something that took less effort to create than it takes to consume" unfortunately i don't remember the source but it rings true nonetheless
I am a firm believer that if text is meant to be read by human, it should be written by human (or at least adapted by human) Had an experience working in a startup with a culture of fully AI-generated PR descriptions/reviews and I might say it slightly depressing to work like that
this weekend I set up a file server and a photo library (immich) on my homelab (janky old laptop), and then had the whole thing set up to back up to a Hetzner storage box automatically, handling encryptcion/compression/etc would've taken me 2 weeks by hand, few hours with the robot
Linux is so much easier to engage with seriously for the first time when there’s a robot I can talk to that can fix all my problems.
This is absolutely insane, this might give a usable agentic KV-size context on my 16GB m1 macbook pro😃
holy smokes, PrismML released Bonsai-27B, a model based on Qwen3.6-27B that comes in two flavors: binary (1.125bpw) and ternary (1.71bpw) weights, while preserving benchmark scores better than standard IQ2 quants! the binary version has a footprint of only 3.9GB! this is wizardry.
after spending a few days figuring out how the different video codecs and compression mechanisms interface to find a smallest optimal storage format for my archive, i concur
Data compression is such an underrated topic and I think you should all learn about it. I cannot understate this. It is connected to more than you could possibly imagine, and it is now my favorite tech subject.
Gergő Bocsárdi reposted
the humble @kagi.com in the corner: 👀👀👀👀
Modern search: “You came looking for X, but wouldn’t you rather spend some time on the more addictive Y?”
Gergő Bocsárdi reposted
This article only earned my click because I had to confirm that the "Erdős method" they refer to isn't just "abusing amphetamines". www.quantamagazine.org/after-80-yea...
Gergő Bocsárdi reposted
@hankgreen.bsky.social casually dropping in a video that "AI already runs the world, it's just not LLMs but recommendation algorithms" very much rhymes with this, and is a take I regularly bring up to people
I think we are locked in an existential struggle with algorithmic solution media and that we are losing
The amount of times people told me that they can't use LLMs to do lit reviews because they hallucinate, and then they DON'T HAVE A SEARCH ENGINE HOOKED UP TO IT???? ps use openalex.org their free API tier is 10/10 and agents love it
Begging academics to include methodology with their twenty post "I tested an LLM and this is how it failed" threads, I just saw a huge thread on how they fail at summarizing literature and it turns out the author wasn't actually providing the text, just asking it for summaries based on training data
Seeing the Black Paintings by Goya in person in Madrid was an amazing experience! So much emotion and "color" expressed essentially only using yellow and black, mesmerizing
Art question 🎨 What painting, photograph, sculpture, or piece of visual art has stayed with you long after you first saw it? Not necessarily your favorite. The one that lingered.
Gergő Bocsárdi reposted
42, now it's time to find out what the question is to match the answer
the future of AI is you pay a million dollars to ask one question and then it says some shit like "If a hole's error text and your paper standstill disagree, trust the error and find the peel-step you fumbled — the "has type" line is your grader now. Score or red."
It seems my investment to go all-in on SQL via duckdb will pay off, now it's time to leave pytorch in the dust alongside polars, pandas, etc Absolutely mindblowing, I love it!
Sneak peek: I implemented a 3 layer MLP that predicts the fashion MNIST dataset using SQL (with my own autograd impl inside the database). Apparently einsum notation is better expressed as a JOIN. github.com/xqlsystems/x...
The Lord of the Rings is a good story, but “the tools of great power are inherently corrupting and you should destroy them before they destroy you” as a moral has some knock-on effects
Feels unreal that my janky little blogpost about using Quarto with Overleaf made the cut to he included in this wonderful resource! And like Nicola I also cannot wait for Quarto 2 so we can leave the workarounds behind for good😃
🚨 New blog post! 🚨 Collaborating on Quarto documents can be tricky, especially if you're working with people who don't need to write code 🫂 So I've pulled together some options for collaborating on Quarto documents into a blog post! Link: nrennie.rbind.io/blog/collabo... #QuartoPub #RStats
Gergő Bocsárdi reposted
Gergő Bocsárdi reposted
I think of this similarly as pre-ai times "compute power": universities/corpos with supercomputers and clusters that you access when needed, and normal users ticking along just fine on their modest compute For me GLM-5.2 is the equivalent of perfectly good daily driver CPU (like my M1 pro)
I agree with this — I think people underestimate the value of additional capability and are reading too much into open models reaching Opus 4.5-ish level. it does mean that capability will soon be free. it doesn’t mean people won’t also pay for more xcancel.com/deanwball/st...
Gergő Bocsárdi reposted
lowkey the best part about my umans max sub is not having to do this with multiple plans (opencode go, glm, codex)
After watching my token consumption like a hawk for a month I decided to go ahead and downgrade my Anthropic subscription to the next tier down from where I was. Should be no problem, should be easy. Except now I have terrible token anxiety. I'm still watching my usage like a hawk! I need to chill.