Load site modules...
lade...
random avatar

maxpool - Network

Posts Subscribe

UMA: A Family of Universal Models for Atomshttps://ai.meta.com/research/publications/uma-a-family-of-universal-models-for-atoms/family of...

https://mathstodon.xyz/@maxpool/...

UMA: A Family of Universal Models for Atoms
ai.meta.com/research/publicati

family of Universal Models for Atoms (UMA), designed to push the frontier of speed, accuracy, and
generalization. UMA models are trained on half a billion unique 3D atomic structures (the largest
training runs to date) by compiling data across multiple chemical domains, e.g. molecules, materials,
and catalysts.

15.5.2025 11:25UMA: A Family of Universal Models for Atomshttps://ai.meta.com/research/publications/uma-a-family-of-universal-models-for-atoms/family of...
https://mathstodon.xyz/@maxpool/...

Self-Improving Reasoners.Both expert human problem solvers and successful language models employ four key cognitive behaviors1. verification...

https://mathstodon.xyz/@maxpool/...

Self-Improving Reasoners.

Both expert human problem solvers and successful language models employ four key cognitive behaviors

1. verification (systematic error-checking),

2. backtracking (abandoning failing approaches),

3. subgoal setting (decomposing problems into manageable steps), and

4. backward chaining (reasoning from desired outcomes to initial inputs).

Some language models naturally exhibits these reasoning behaviors and exhibit substantial gains, while others don't and quickly plateau.

The presence of reasoning behaviors, not the correctness
of answers is the critical factor. Models with incorrect solutions containing proper reasoning patterns achieve comparable performance to those trained on correct solutions.

It seems that the presence of cognitive behaviors enables self-improvement through RL.

Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs
arxiv.org/abs/2503.01307


6.3.2025 10:48Self-Improving Reasoners.Both expert human problem solvers and successful language models employ four key cognitive behaviors1. verification...
https://mathstodon.xyz/@maxpool/...

"The Leader of the Free world" (1948–2025)The first time the phrase “leader of the free world” appeared in The New York...

https://mathstodon.xyz/@maxpool/...

"The Leader of the Free world"
(1948–2025)

The first time the phrase “leader of the free world” appeared in The New York Times was in a November 1948 essay by the British economist Barbara Ward, which urged Western unity against the communist threat.

The US leadership status ended after the February 2025 meeting between the US president and the president of Ukraine where the US president and vice president together attacked the president of Ukraine.

Immediately after the meeting ended other free world heads of states started to send out "We stand with Ukraine" messages.


28.2.2025 21:45"The Leader of the Free world" (1948–2025)The first time the phrase “leader of the free world” appeared in The New York...
https://mathstodon.xyz/@maxpool/...

Xinyu Guan et al, rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking, arXiv...

https://mathstodon.xyz/@maxpool/...

Xinyu Guan et al, rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking, arXiv (2025).
arxiv.org/abs/2501.04519

Microsoft introduces rStar-Math, an SLM for math reasoning and problem solving
techxplore.com/news/2025-01-mi

"small language models (SLMs) can rival or even surpass the math reasoning capability of OpenAI o1, without distillation from superior models. rStar-Math achieves this by exercising "deep thinking" through Monte Carlo Tree Search (MCTS), where a math policy SLM performs test-time search guided by an SLM-based process reward model."

11.1.2025 12:00Xinyu Guan et al, rStar-Math: Small LLMs Can Master Math Reasoning with Self-Evolved Deep Thinking, arXiv...
https://mathstodon.xyz/@maxpool/...

Microsoft and OpenAI have contractual definition of AGI....

https://mathstodon.xyz/@maxpool/...

Microsoft and OpenAI have contractual definition of AGI.
techcrunch.com/2024/12/26/micr The need for definition is because Microsoft loses access to OpenAI’s technology when the startup reaches AGI.

"The two companies reportedly signed an agreement last year stating OpenAI has only achieved AGI when it develops AI systems that can generate at least $100 billion in profits."
That's Saudi Aramco, Apple, Alphabet, level profits. en.wikipedia.org/wiki/List_of_

Currently OpenAI loses billions annually, and tells investors it won’t turn a profit until 2029.

---

I think Microsoft made a good bet.

26.12.2024 19:55Microsoft and OpenAI have contractual definition of AGI....
https://mathstodon.xyz/@maxpool/...

A synthetic protein-level neural network in mammalian cellshttps://www.science.org/doi/10.1126/science.add8468Bringing neural networks to...

https://mathstodon.xyz/@maxpool/...

A synthetic protein-level neural network in mammalian cells
science.org/doi/10.1126/scienc

Bringing neural networks to life
A synthetic protein-based winner-take-all neural network controls cell fate decisions
science.org/doi/10.1126/scienc

Chen et al. describe a protein-level synthetic circuit framework that implements a “winner-take-all” neural network capable of classifying the relative abundance of multiple inputs to modify the circuit output. This network replicates various classification circuits by adjusting the relative concentrations of a few components. By connecting this circuit to molecular pathways that regulate apoptosis, the authors demonstrate the promise of this approach for programming cell fate outcomes. This could be extended to design complex neural circuits that augment the computational capacity of cells.

perceptein: a combination of protein and perceptron.

biorxiv:
biorxiv.org/content/10.1101/20

phys.org/news/2024-12-percepte


22.12.2024 09:50A synthetic protein-level neural network in mammalian cellshttps://www.science.org/doi/10.1126/science.add8468Bringing neural networks to...
https://mathstodon.xyz/@maxpool/...

The Dimensions of dimensionalityhttps://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(24)00189-X#Globally interpretable feature...

https://mathstodon.xyz/@maxpool/...

The Dimensions of dimensionality
cell.com/trends/cognitive-scie

Globally interpretable feature dimensions are one property, among others like prediction performance and compactness, that researchers can prioritize when inferring latent representations.

Multidimensional representational spaces can capture complex structures such as hierarchical relationships, lexical entailment, and compositional features.

Superficially different representations, such as graphs and multidimensional data, can capture the same structural relationships under appropriate settings.

The dimensionality of a representation conveys relatively little information on its own.

19.12.2024 13:21The Dimensions of dimensionalityhttps://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(24)00189-X#Globally interpretable feature...
https://mathstodon.xyz/@maxpool/...

The unbearable slowness of being: Why do we live at 10 bits/s?...

https://mathstodon.xyz/@maxpool/...

The unbearable slowness of being: Why do we live at 10 bits/s? cell.com/neuron/abstract/S0896

arxiv.org/abs/2408.10234

The information throughput of a human being is about 10 bits/s. In comparison, our sensory systems gather data at ∼1⁢0^9 bits/s.

The brain seems to operate in two distinct modes: the “outer” brain handles fast high-dimensional sensory and motor signals, whereas the “inner” brain processes the reduced few bits needed to control behavior.

18.12.2024 16:41The unbearable slowness of being: Why do we live at 10 bits/s?...
https://mathstodon.xyz/@maxpool/...

Urban–rural political divideThe Urban-rural political divide is not limited to the U.S. It's global phenomenon....

https://mathstodon.xyz/@maxpool/...

Urban–rural political divide

The Urban-rural political divide is not limited to the U.S. It's global phenomenon.
en.wikipedia.org/wiki/Urban%E2

The Density Divide: Urbanization, Polarization, and Populist Backlash, Will Wilkinson, 2019
niskanencenter.org/wp-content/

Urbanization sorts populations on attribute—ethnicity, personality, and education—that make individuals more or less responsive to the incentives to move toward cities.

Self-selected migration has segregated the national population and concentrated economic production into megacities, driving a polarizing wedge between dense diverse populations and sparse white populations—the “density divide.

The filtering/sorting dynamic of urbanization has produced a lower-density, mainly white population that is increasingly uniform in socially conservative personality, aversion to diversity, relative disclination to migrate and seek higher education, and Republican Party loyalty.

30.10.2024 16:24Urban–rural political divideThe Urban-rural political divide is not limited to the U.S. It's global phenomenon....
https://mathstodon.xyz/@maxpool/...

Writes and Write-Nots by Paul Graham (Y Combinator) "I'm usually reluctant to make predictions about technology, but I feel fairly...

https://mathstodon.xyz/@maxpool/...

Writes and Write-Nots by Paul Graham (Y Combinator)

"I'm usually reluctant to make predictions about technology, but I feel fairly confident about this one: in a couple decades there won't be many people who can write. ... writing is thinking. In fact there's a kind of thinking that can only be done by writing. ... So a world divided into writes and write-nots is more dangerous than it sounds. It will be a world of thinks and think-nots."

paulgraham.com/writes.html

27.10.2024 09:40Writes and Write-Nots by Paul Graham (Y Combinator) "I'm usually reluctant to make predictions about technology, but I feel fairly...
https://mathstodon.xyz/@maxpool/...

Evidence that Elon Musk’s takeover of Twitter was concurrent with a decline in the academic community of scholars who had been using the...

https://mathstodon.xyz/@maxpool/...

Evidence that Elon Musk’s takeover of Twitter was concurrent with a decline in the academic community of scholars who had been using the platform to share and discuss research.

Bisbee J, Munger K. The Vibes Are Off: Did Elon Musk Push Academics Off Twitter? PS: Political Science & Politics. Published online 2024:1-8. doi:10.1017/S1049096524000416 cambridge.org/core/journals/ps

Elon Musk’s Twitter takeover triggered academic exodus, study suggests psypost.org/elon-musks-twitter

21.10.2024 13:21Evidence that Elon Musk’s takeover of Twitter was concurrent with a decline in the academic community of scholars who had been using the...
https://mathstodon.xyz/@maxpool/...

❝𝐏𝐨𝐥𝐢𝐭𝐢𝐜𝐚𝐥 𝐡𝐨𝐛𝐛𝐲𝐢𝐬𝐦 is a catchall for all the ways we do politics to serve our own...

https://mathstodon.xyz/@maxpool/...

❝𝐏𝐨𝐥𝐢𝐭𝐢𝐜𝐚𝐥 𝐡𝐨𝐛𝐛𝐲𝐢𝐬𝐦 is a catchall for all the ways we do politics to serve our own emotional or intellectual needs or wants rather than do a kind of power-seeking organized behavior. It’s arguing about the news, sharing the news, reacting to news, being an at-home pundit about the news. All of those things that involve giving politics your headspace and emotion but not doing the real work of it.❞

simonandschuster.com/books/Pol

❝For most people who are political junkies, their news consumption is not really geared toward information that is going to help them be active citizens in the community. And even if it is, they’re not being active in the community. Most people who are daily news consumers belong to zero organizations and have worked zero times in the past year with other people on a community problem. So, most people are not doing anything.❞

❝...let’s say 2 percent, is real community or volunteer engagement. The rest is mostly news consumption and sharing, talking, and debating online. The group that spends the most time in any kind of political engagement is white men, particularly college-educated white men. They know the most facts about politics, but they are not the group that is engaging most in real politics — organized politics with goals and strategies.❞

Are you a “political hobbyist”? If so, you may be the problem. Political scientist Eitan Hersh explains how “political hobbyism” is ruining politics
vox.com/2020/3/11/21172064/pol

23.9.2024 10:48❝𝐏𝐨𝐥𝐢𝐭𝐢𝐜𝐚𝐥 𝐡𝐨𝐛𝐛𝐲𝐢𝐬𝐦 is a catchall for all the ways we do politics to serve our own...
https://mathstodon.xyz/@maxpool/...

The piranha problem: Large effects swimming in a small pond.Christopher Tosh, Philip Greengard, Ben Goodrich, Andrew Gelman, @avehtari,...

https://mathstodon.xyz/@maxpool/...

The piranha problem: Large effects swimming in a small pond.
Christopher Tosh, Philip Greengard, Ben Goodrich, Andrew Gelman, @avehtari, @djhsu
2 Apr 2024
arxiv.org/abs/2105.13445

In a lot of social science research, small, random factors are reported as having large effects on social and political attitudes and behavior (social priming, hormonal levels,parental socioeconomic status, weather, ...). Studies have claimed to find large effects from these and other inputs.

The results show that it would be extremely unlikely to have all these large effects coexisting—they would have to almost exactly cancel each other out.

22.9.2024 10:03The piranha problem: Large effects swimming in a small pond.Christopher Tosh, Philip Greengard, Ben Goodrich, Andrew Gelman, @avehtari,...
https://mathstodon.xyz/@maxpool/...

Robert Hooke’s (1635 – 1703) model of memoryhttps://link.springer.com/article/10.3758/BF03196465In 1682 the scientist and inventor...

https://mathstodon.xyz/@maxpool/...

Robert Hooke’s (1635 – 1703) model of memory
link.springer.com/article/10.3

In 1682 the scientist and inventor Robert Hooke read a lecture to the Royal Society of London, in which he described a mechanistic model of human memory.

The lecture addressed questions of encoding, memory capacity, repetition, retrieval, and forgetting—some of these in a surprisingly modern way.

(1) it allows for attention and other top-down influences on encoding;
(2) it uses resonance to implement parallel, cue-dependent retrieval;
(3) it explains memory for recency;
(4) it offers a single-system account of repetition priming; and
(5) the power law of forgetting can be derived from the model’s assumptions in a straightforward way.

31.8.2024 11:43Robert Hooke’s (1635 – 1703) model of memoryhttps://link.springer.com/article/10.3758/BF03196465In 1682 the scientist and inventor...
https://mathstodon.xyz/@maxpool/...

NSA releases copy of internal lecture delivered by computing giant Rear Adm. Grace Hopper Capt. Grace Hopper on Future Possibilities: Data,...

https://mathstodon.xyz/@maxpool/...

NSA releases copy of internal lecture delivered by computing giant Rear Adm. Grace Hopper

Capt. Grace Hopper on Future Possibilities: Data, Hardware, Software, and People (Part One, 1982)
youtube.com/watch?v=si9iqF5uTF

Capt. Grace Hopper on Future Possibilities: Data, Hardware, Software, and People (Part Two, 1982)
youtube.com/watch?v=AW7ZHpKuqZ

press release:
nsa.gov/Press-Room/Press-Relea

27.8.2024 21:00NSA releases copy of internal lecture delivered by computing giant Rear Adm. Grace Hopper Capt. Grace Hopper on Future Possibilities: Data,...
https://mathstodon.xyz/@maxpool/...

Ensemble theory and the Wisdom of Crowds. Where political science and machine learning come together....

https://mathstodon.xyz/@maxpool/...

Ensemble theory and the Wisdom of Crowds. Where political science and machine learning come together. en.wikipedia.org/wiki/Ensemble

Promoting Erroneous Divergent Opinions Increases the Wisdom of Crowds journals.sagepub.com/doi/abs/1

Wisdom of the crowd comes from the accuracy of individual assessments and the diversity of opinions. Most strategies enhance the wisdom of crowds by improving individual accuracy while neglecting the potential of increasing opinion diversity.

QUESTION: Is it possible to make the crowd wiser while making individuals more inaccurate?

When people adopt extreme and erronous views, this may not degrade the whole society. Paradoxically, it can make us collectively smarter.


18.8.2024 15:06Ensemble theory and the Wisdom of Crowds. Where political science and machine learning come together....
https://mathstodon.xyz/@maxpool/...

Divergent recruitment of developmentally defined neuronal ensembles supports memory dynamics...

https://mathstodon.xyz/@maxpool/...

Divergent recruitment of developmentally defined neuronal ensembles supports memory dynamics science.org/doi/10.1126/scienc

Summary: Memory stored in multiple copies. Copies modified, stored differently.

biorxiv.org/content/10.1101/20

Parallel-emerging memory traces encoded in developmentally defined hippocampal subpopulations underpin memory dynamics.

Early- and late-born neurons in hippocampal CA3 exhibit distinct baseline functional properties and responses to learning. Late-born neurons plastically reconfigure their coactivity dynamics and functional connectivity upon memory encoding, whereas early-born neurons exhibit rigid responses. Their divergent recruitment to memory ensembles over time supports a memory’s persistence and modulates its plasticity.

press release
unibas.ch/en/News-Events/News/

18.8.2024 11:21Divergent recruitment of developmentally defined neuronal ensembles supports memory dynamics...
https://mathstodon.xyz/@maxpool/...

alphaXiv - An open research discussion directly on top of arXiv https://alphaxiv.org/For any paper change the URL arxiv.org/... ->...

https://mathstodon.xyz/@maxpool/...

alphaXiv - An open research discussion directly on top of arXiv alphaxiv.org/

For any paper change the URL arxiv.org/... -> alphaxiv.org/...

1.8.2024 14:18alphaXiv - An open research discussion directly on top of arXiv https://alphaxiv.org/For any paper change the URL arxiv.org/... ->...
https://mathstodon.xyz/@maxpool/...

Calculating the Cost of a Google Deepmind Paper. https://152334h.github.io/blog/scaling-exponents/How to burn US$10,000,000 on an arXiv...

https://mathstodon.xyz/@maxpool/...

Calculating the Cost of a Google Deepmind Paper.
152334h.github.io/blog/scaling

How to burn US$10,000,000 on an arXiv preprint

30.7.2024 22:31Calculating the Cost of a Google Deepmind Paper. https://152334h.github.io/blog/scaling-exponents/How to burn US$10,000,000 on an arXiv...
https://mathstodon.xyz/@maxpool/...

All Objects in Universe in One Pedagogical Plot"All objects and some questions"Am. J. Phys. 91, 819–825...

https://mathstodon.xyz/@maxpool/...

All Objects in Universe in One Pedagogical Plot

"All objects and some questions"
Am. J. Phys. 91, 819–825 (2023)
doi.org/10.1119/5.0150209

22.10.2023 15:56All Objects in Universe in One Pedagogical Plot"All objects and some questions"Am. J. Phys. 91, 819–825...
https://mathstodon.xyz/@maxpool/...
Subscribe
To add news/posts to your profile here, you must add a link to a RSS-Feed to your webfinger. One example how you can do this is to join Fediverse City.
         
Webfan Website Badge
Nutzungsbedingungen   Datenschutzerklärung  Impressum
Webfan | @Web pages | Fediverse Members

⬆️

⬇️