I were 17, I'd learn how to build LLMs from scratch
https://twitter.com/paulg/status/2091544343589060625
Comments: https://news.ycombinator.com/item?id=49412396
#HackerNews #learnLLMs #buildFromScratch #techEducation #futureSkills
#ChatGPT says they will analyze the talks of all the citizens and show ads "relevant for the current talk in the EU".
@EUCommission
Did you got this? Did clowns forgot your laws again?
32x32 pixel sprites of the classic Doom motley crew. Drawn last year following a little trend on social media.
(I made an exception for the Cyberdemon & Spider Mastermind, which are a little bigger)
@pita @cosmicvoid Turn them into stickers!
Finished CIRCE by Madeline Miller 📚
What a beautiful rumination on change and the ups and downs of life. I basically had tears for the last quarter of the book (and I read the last 45% just today! Couldn’t stop) https://micro.blog/books/9780316556323
Prior to this post I have 4 that use “delve”, 5 that use “interplay”, and 2 that use “testament”.
I frequently use en dashes – and many that use em dashes.
I use the Oxford comma when it clarifies, but not when it’s superfluous.
I’m sceptical of conclusions drawn from use of tools like Pangram, linked to from https://www.pewresearch.org/data-labs/2026/08/20/how-much-of-the-internet-is-written-with-ai/, which is linked to from https://m.slashdot.org/story/457268.
It’s not just that the conclusions are shaky, it’s also that they do nothing to alter the incentives to slop
So, been reading some of the hand wringing over on #Bluesky about their dismal MAU numbers, and what is interesting is the fear that they engender among users who are worried that the company will not continue to be able to operate if their venture capitalist owners decide it will never be the 'next big thing.'
Basically, like all VC backed companies, there is a 'runway' during which the investment has to either make money or prove worthy of more invested capital.
-more-
@mastodonmigration ultimately, the deal we make is to use good things while they are free, and then move on when the monetization ruins them.
This is inevitable.
I am apparently NOT allowed under ANY circumstances to buy a drinks globe despite always wanting one.
@TheBreadmonkey Your self-expression and joi de vivre is being repressed! Rebel! Break free of the chains of mundanity!
Alonzo-dev news: I fixed a few nil-dereference bugs so function bodies can be rendered. Once I've synthesized something rudimentary to render.
Its still not rendering sufficiently for me to interact with, but at least it is rendering!
Now I'm in my issue tracker planning out next steps...
https://codeberg.org/alcinnz/Alonzo/issues
RE: https://mastodon.social/@danluu/111037093079443322
With all the discussions of how much bot traffic is flooding websites, I was surprised to find that my CDN is "only" showing 3.5 TB used in the past month, up 3x-ish from 3 years ago.
Human traffic is probably down because I haven't been active for the last couple years, so maybe we can call that a 4x to 5x increase due to bots?
Maybe my site just isn't very appealing to bots, but from the way people talk, I was expecting a higher ratio.
GeofCox
boosted
Uber faces fine of nearly $1B over automated driver suspensions
https://techcrunch.com/2026/08/23/uber-faces-fine-of-nearly-1b-over-automated-driver-suspensions/
The Dutch Data Protection Authority is fining Uber €825 million (around $966 million) — the second largest penalty issued so far under Europe’s General Data Protection Regulation
Titus Stahl
boosted
#introduction I work with external research funding at the IT University of Copenhagen, Denmark.
I’m interested in #Bibliometrics #Scientometrics #PeerReview #ResearchAssessment #ResponsibleResearchAssessment and a lot of other research related things.
I love live music. Especially #MetalMusic and #CountryMusic
Moving my account to Helvede. My Mastodon home since 2022 is closing. I'm not tooting much but appreciate all the inspiration I get from other accounts and hashtags.
Unfortunately, theoretical work often seems to gain provable security only at the cost of efficiency.
how are you being anti-intellectual while being cryptographic academics
This is due in part to the following. Theorists view certain primitives (e.g., one-way functions) as “basic”
one-way functions are bullshit, that's what merkle said about fucking DES. it's NP-completeness or go home
and build more powerful primitives (e.g., pseudorandom functions) out of them in inefficient ways;
i don't even know what a pseudorandom function is but i bet it sucks. also "inefficient"????? maybe yeah if you mean the BLAKE3 reference implementation
but in practice, powerful primitives are readily available and the so-called basic ones seem to be no easier to implement.
literally indistinguishable from LLM boosters
In fact theorists deny themselves the capabilities of practical primitives which satisfy not only the strongest kinds of assumptions they like to make, but even have strengths which have not been defined or formalized.
this is a crazy fucking thing to say. everyone in this field is first against the wall
Leo
boosted
Mannaggia a sant'Audoeno, vescovo di Rouen
Get a book from the library (free) or bookstore about some topic you’ve been interested in, but have never taken the time to learn.
The problem with using LLMs to research information is that this technology generates text, it does not report information.
Generating a map is not the same as finding an accurate map that already exists. Same for any other pieces of information.
And even if it was able to find and report information without modifying it (what a search engine does basically), it cannot evaluate if this information is accurate.
Adjust your level of trust accordingly.
How would you categorize Claude Code? I’m using it to critique my Swift code and find it remarkably instructive. It just makes suggestions, sometimes involving language features I’ve never seen before. I try them out, implement them, and they invariably lead to cleaner, tighter code. Is Claude Code properly considered to be an LLM?
Che brutto il socialismo, che brutta la proprietà statale, che orrore la gestione pubblica, viva il libero mercato
https://tg24.sky.it/cronaca/2026/08/22/distributore-benzina-comune-prezzi-sconto
Elevated Errors for Multiple Models
https://status.claude.com/incidents/vgz5psbjmt1h
Comments: https://news.ycombinator.com/item?id=49415405
#HackerNews #ElevatedErrors #MultipleModels #TechNews #DataScience #IncidentReport
Skull & Suit: An Eldritch Executive Interview
idk why they let me see those papers from the 1980s explaining how to use DH to do everything
Unfortunately, theoretical work often seems to gain provable security only at the cost of efficiency.
how are you being anti-intellectual while being cryptographic academics
This is due in part to the following. Theorists view certain primitives (e.g., one-way functions) as “basic”
one-way functions are bullshit, that's what merkle said about fucking DES. it's NP-completeness or go home
and build more powerful primitives (e.g., pseudorandom functions) out of them in inefficient ways;
i don't even know what a pseudorandom function is but i bet it sucks. also "inefficient"????? maybe yeah if you mean the BLAKE3 reference implementation
but in practice, powerful primitives are readily available and the so-called basic ones seem to be no easier to implement.
literally indistinguishable from LLM boosters
