Marco Z
@ocramz.bsky.social
📤 995
📥 945
📝 1311
ML, λ • language and the machines that understand it •
https://ocramz.github.io
pinned post!
CERN for frontier AI >>>
8 months ago
1
8
1
idea to published extension: 1 hour
marketplace.visualstudio.com/items?itemNa...
about 11 hours ago
0
5
0
taalas.com/the-path-to-...
the company behind this
add a skeleton here at some point
about 18 hours ago
0
0
0
a chip with the whole LLM baked in, ready for inference when you power it on.
add a skeleton here at some point
about 21 hours ago
2
5
1
arguably there are islands of software engineering where this is already happening
add a skeleton here at some point
1 day ago
0
0
0
loading . . .
2 days ago
1
0
0
I'd start by questioning whether and why do we need to crank out tokens at absurd rates just to do useful work
add a skeleton here at some point
4 days ago
0
1
0
reposted by
Marco Z
Shen-yi Liao 廖顯禕
4 days ago
The correct attitude that one should take toward AI is as a left technocrat. Be relentlessly empirical about the powers and pitfalls of the technology, but do not forget the contingent material circumstances that shape its deployments. Incidentally, I am a left technocrat.
0
47
11
reposted by
Marco Z
Sam Rose
5 days ago
Are you using open models on a regular basis? If yes, I’d love to hear about it! - what’s your use-case? - local or through an inference provider? - how has the experience evolved over time? - anything you miss from the closed models, if you’ve used them? Would appreciate reposts for visibility 🙏
12
53
28
will be interesting to see whether this approach enables smaller models too what are the scaling laws of models that can (partly) introspect? huge and v interesting design space
add a skeleton here at some point
7 days ago
0
9
0
SIX papers to review for
@icmlconf.bsky.social
?? that's like a full time job for a week
8 days ago
1
5
0
reposted by
Marco Z
hikikomorphism
8 days ago
I bet y'all thought I was joking when I said I was rewriting gastown in haskell
1
4
1
(months deep into an audio deep learning project) torchaudio goes into maintenance mode with whole module subtrees getting removed from one point release to the next
github.com/pytorch/audi...
loading . . .
Update on TorchAudio’s future · Issue #3902 · pytorch/audio
EDIT from 2026/01/22: we just released TorchAudio 2.10, which marks the completion of the migration detailed below. Based on user feedback, one important modification was made to the original plan:...
https://github.com/pytorch/audio/issues/3902
9 days ago
0
0
0
💯 plus, rearranging all knowledge work to rely on hyperscalers supplying cheap tokens as a utility is .. short sighted
add a skeleton here at some point
9 days ago
0
5
0
reposted by
Marco Z
Jon Bois
10 days ago
if you played sonic the hedgehog and put down the controller, sonic would look at you and tap his foot impatiently. this was proof that sega genesis cartridges had souls. the decision was made to make the storage cases a lot larger than the cartridges so they would have room to move around in there
67
7486
1496
openai crashes out of the agi contract with microsoft, gets cannibalized and sold for scraps anthropic buys salesforce, or the other way around, turning into a productivity saasburger tokens become suddenly very expensive what happens then?
12 days ago
4
7
0
in good company, close to the HuggingFace local cluster
add a skeleton here at some point
12 days ago
0
2
0
from
speakerdeck.com/yaroslavvb/g...
h/t
@escorciav.bsky.social
12 days ago
0
3
0
the economic side of the equation is .. concerning? crazy? Just How The World Works? I cannot find a term
www.forbes.com/sites/rashis...
add a skeleton here at some point
12 days ago
0
0
0
I just recalled FauxPilot by
@moyix.net
first commit: August 2022 truly visionary stuff
github.com/fauxpilot/fa...
loading . . .
GitHub - fauxpilot/fauxpilot: FauxPilot - an open-source alternative to GitHub Copilot server
FauxPilot - an open-source alternative to GitHub Copilot server - fauxpilot/fauxpilot
https://github.com/fauxpilot/fauxpilot
13 days ago
1
1
0
the cost of typing may have turned to zero, but eliciting requirements, sanding off assumptions etc, is most of the sw engineering work
add a skeleton here at some point
13 days ago
0
6
1
the uncanny valley is deeper than previously thought
add a skeleton here at some point
13 days ago
0
0
0
just as agentic coding broke the correlation between work artifacts and author proficiency, quirky social agents break the tacit contract between lived experience and its narration
14 days ago
0
8
4
AI agents are language machines, and their range of behaviors is an interesting effect. However I find it v problematic that they self identify as people or are made to imitate human character traits in a sort of emotional manipulation.
add a skeleton here at some point
14 days ago
4
14
2
"Found it !! "Unless .. claude stop playing games with my heart challenge
16 days ago
0
6
0
quite handy that the scaling law for baking is 1 egg/ 100g flour
16 days ago
2
6
0
ever wondered how does claude code work ? I certainly do. go check out OpenClaudia by
@dollspace.gay
!
github.com/dollspace-ga...
loading . . .
GitHub - dollspace-gay/OpenClaudia
Contribute to dollspace-gay/OpenClaudia development by creating an account on GitHub.
https://github.com/dollspace-gay/OpenClaudia
17 days ago
1
9
0
OSS is many things, including learning communication performance art and, yes, infrastructure. Here I mean the latter
add a skeleton here at some point
18 days ago
0
1
0
inductive learning >> dog parks for moltys
add a skeleton here at some point
18 days ago
0
4
0
does building an agent framework that does the work count as working
18 days ago
4
5
1
taking this remark seriously for a moment. What would it take for AI coding to be genuinely useful for open source, in the sense of only filling the gaps when things are missing, and looking for dependencies otherwise?
19 days ago
2
1
1
that's what you get for treating citations as data and not pointers
add a skeleton here at some point
20 days ago
1
12
0
This is not true; I beg people read the full paper and especially the study design. The conclusions mirror my (and many other practitioners') conclusions: if you use AI critically and engage both with the question and the answer, it has a net positive impact on both learning and productivity
add a skeleton here at some point
21 days ago
9
67
15
oss author: hey guys check out the thing I made forum: harrumph! disappointing that X doesn't do Y
21 days ago
0
0
0
is moltbook yet another exotic infosec sidechannel liability? yup
22 days ago
0
4
0
the first two pages of suggested abstracts were very relevant to my interests but it's quite an exhausting exercise
add a skeleton here at some point
23 days ago
0
4
1
oh yeah oops lol I have no clue where my data goes AND the claims I make in a post
23 days ago
0
5
0
ah yes, the AI-MKULTRA connection, makes sense
add a skeleton here at some point
23 days ago
0
9
0
enjoying my steak&eggs while chained to the treadmill desk like a true Goblin
add a skeleton here at some point
23 days ago
1
4
0
dehydrated and sleep deprived but nailed the ICML deadline. now 🤞🤞
23 days ago
0
2
0
reposted by
Marco Z
Nature
24 days ago
Wikipedia is an antidote to an increasingly poisoned information ecosystem
go.nature.com/4pXIv2d
loading . . .
Wikipedia is needed now more than ever, 25 years on
The online encyclopedia is an antidote to an increasingly poisoned information ecosystem. Researchers should help to nourish it.
https://go.nature.com/4pXIv2d
2
168
57
if we recognize that science is a social endeavor, UX improvements of tools like this one become clear: what matters is not the IDE, but the feedback between ideas, people and the physical world. a network of minds enhanced by better search, better typing and perhaps better hypothesis generation
add a skeleton here at some point
24 days ago
1
1
0
smh
25 days ago
1
1
0
context management adds a whole new dimension to the process scheduling gymnastics, it's not just "reinventing Erlang" as some have said. For one I'd be very interested to see/pursue "smol" versions of this that don't rely on Big Model to be always available
add a skeleton here at some point
25 days ago
0
4
1
typing some code by hand, as a treat
26 days ago
0
8
0
"are LLMs conscious" – the greatest thread in the history of forums, locked by a moderator after 12,239 pages of heated debate,
add a skeleton here at some point
28 days ago
0
3
0
an insightful account/parable on this, well worth a read :
www.galois.com/articles/spe...
add a skeleton here at some point
29 days ago
0
9
1
LaTeX in vscode? with auto build, preview, hyperlinks, _and_ typing assist? what is this sorcery
29 days ago
2
13
1
hello new followers!
#introduction
Here I bleet about; * assorted numerical, ML/AI nuts&bolts * research: languages (natural, artificial, compilers), interpretability, formal verification, interesting LLM experiments, etc
about 1 month ago
0
2
0
we speak 3 languages at home (🇬🇧🇱🇹🇮🇹), and kid #2 recently turned 2, so naturally I have a half baked theory on language acquisition in children the "least effort" theory: both kids mixed words from the various languages to produce their first sentences, _as long as they are short_ 1/n
about 1 month ago
1
4
0
oh
lobste.rs
, never change (please change)
about 1 month ago
0
1
0
Load more
feeds!
log in