Create accountLogin

Recent searches

No recent searches

Search options

Only available when logged in.
med-mastodon.com is one of the many independent Mastodon servers you can use to participate in the fediverse.
Medical community on Mastodon

Administered by:

Server stats:

341
active users

med-mastodon.com: AboutProfiles directoryPrivacy policy

Mastodon: AboutGet the appKeyboard shortcutsView source codev4.3.8

#scaling

1 post1 participant0 posts today
Public
Ecologia Digital

"The vast investments in scaling... always seemed to me to be misplaced."

"The premise that #AI could be indefinitely improved by #scaling was always on shaky ground… In November last year, reports indicated that #OpenAI researchers discovered that the upcoming version of its GPT large language model displayed significantly less improvement, and in some cases, no improvements at all than previous versions did over their predecessors."

futurism.com/ai-researchers-te

Futurism · Majority of AI Researchers Say Tech Industry Is Pouring Billions Into a Dead EndBy Frank Landymore
Public
Hacker News

Can reinforcement learning for LLMs scale beyond math and coding tasks? Probably

arxiv.org/abs/2503.23829

arXiv.orgCrossing the Reward Bridge: Expanding RL with Verifiable Rewards Across Diverse DomainsReinforcement learning with verifiable rewards (RLVR) has demonstrated significant success in enhancing mathematical reasoning and coding performance of large language models (LLMs), especially when structured reference answers are accessible for verification. However, its extension to broader, less structured domains remains unexplored. In this work, we investigate the effectiveness and scalability of RLVR across diverse real-world domains including medicine, chemistry, psychology, economics, and education, where structured reference answers are typically unavailable. We reveal that binary verification judgments on broad-domain tasks exhibit high consistency across various LLMs provided expert-written reference answers exist. Motivated by this finding, we utilize a generative scoring technique that yields soft, model-based reward signals to overcome limitations posed by binary verifications, especially in free-form, unstructured answer scenarios. We further demonstrate the feasibility of training cross-domain generative reward models using relatively small (7B) LLMs without the need for extensive domain-specific annotation. Through comprehensive experiments, our RLVR framework establishes clear performance gains, significantly outperforming state-of-the-art open-source aligned models such as Qwen2.5-72B and DeepSeek-R1-Distill-Qwen-32B across domains in free-form settings. Our approach notably enhances the robustness, flexibility, and scalability of RLVR, representing a substantial step towards practical reinforcement learning applications in complex, noisy-label scenarios.
#HackerNews#reinforcementlearning#LLMs
Public
💧🌏 Greg Cocks

Scaling In Branch Thickness And The Fractal Aesthetic Of Trees [In Art From Da Vinci To Mondrian]
--
academic.oup.com/pnasnexus/art <-- shared paper
--
[not my usual fare, but love the spatial mathematical components, although the maths is way over my head!]
#spatial #maths #mathematics #art #painting #representation #daVinci #LeonardodaVinci #MurraysLaw #tree #vegetation #fractal #fractals #α #biology #artwork #radiusscalingexponent #branches #branching #nature #naturalpatterns #proportions #physiology #Mondrian #aesthetic #aesthetics #scale #scaling

a) da Vinci’s sketch of a tree illustrates the principle that combined thickness is preserved at different stages of ramification. b) Transects A and B have the same combined thickness and so limbs a and b together must have the same thickness as c. This has been taken to imply that c) combined cross-sectional area (⁠πR²⁠) is preserved across branching. Likewise, d) such transects can count the number of branches n that have radius approximately r to derive a fractal scaling relationship. Source: (a) from Institut de France Manuscript M, p. 78v.
sketch from one of Leonardo da Vinci's notebooks illustrating his original rules of trees
The scaling exponent α (a) determines the amount of detail at a given resolution and (b) proportions among adjoining boughs.
Trees in artwork across centuries, cultures, and styles exhibit fractal scaling. a) carved stone window screen of Sidi Saiyyed Mosque, Ahmedabad, Gujarat, India (1573 CE). b) Cherry Blossoms, Matsumura Goshun (1752–1811), ink on paper. c) L’Arbre de Vie (Tree of life), Gustav Klimt (1909), oil on canvas. d) data and inference of α±95% CI in two replicate measurements for each work along with simulated and living trees. Dotted vertical lines indicate the cutoff for excluding small-scale effects. Diameters from simulated and living trees are divided by this minimum value and therefore the horizontal axis is unitless.
Replied in thread
Public
Kent Pitman

@climatebrad

It feels to me like "build more housing" can't be the answer. You almost might as well say "make more land". It's not a durable solution. And it doesn't address the many other aspects of society that need to be addressed. Jobs food commerce in general, schools, the nature and flow of community itself.

A favorite quote comes to mind.

"Better implies different."
--Amar Bose, at an MIT Enterprise Forum event

(He was trying to explain to sales people at stores that would sell Bose speakers why they had to make changes in how they set them up. "Couldn't they just do what they'd always done?" The people would ask. They were used to that and did not want to change. He was trying to explain succinctly why you can't just radically improve something and leave it the same at the same time. So he, explained, that slogan had emerged.)

Surely higher population density at some point means using existing resources differently. I'm not pushing an agenda here, but I am observing that higher density feels less compatible within every person for themselves and traditional-ownership / rent-taking-for-profit model. Surely that brings a 2-tiered citizenship and breeds discontent/danger as inequality simmers.

In computer science, we talk about building systems that scale, planning for higher traffic. This could really be done in a system that did not plan for scale without the architecting the system entirely, and I've even seen some of pine that every factor of 10 in scale requires a redesign.

Sometimes the architectural plan is indeed to just add servers, but that has to be planned in, and there has to be a source of servers, and the system architecture has to be structured such that in the new model, all the necessary flows will happen correctly and resources won't be cut off from each other or too hard to access or too expensive.

"Build more housing." does not sound like the kind of answer I could give in a job interview and expect to be hired, with the hiring manager saying "this person has clearly demonstrated their understanding of operating at scale". The answer is not of a shape that seems right to me, nor does it offer sufficient detail.

A lot of capitalism seems to operate on a theory that you just twist some knobs and everything will just happen right without coordination. I think this is less and less true as either populations grow larger or resources grow smaller or resources become more stressed.

I did not write the accompanying article specifically to address this issue, and yet I feel like it says some important additional things I might say here if I were to ramble on. It is not a complete discussion of scale, but more discussion of why I don't think the traditional ways of thinking about just turning a few knobs is likely to keep working.

Losing Ground in the Environment
netsettlement.blogspot.com/201

It also just not addressed the issue of urgency, and the way in which urgency materially changes the set of usable solutions. I did try to address that issue here:

The Politics of Delay
netsettlement.blogspot.com/202

netsettlement.blogspot.comLosing Ground in the EnvironmentEssay on how we can't still see the world as an infinite resource. Things are interconnected and finite, so we need a fresh mindset when planning.
#climate#ClimateMigration#SystemsThinking
ExploreLive feeds
About