HN user

docmechanic

1,042 karma
Posts361
Comments389
View on HN
www.spiegel.de 9mo ago

Polar Researchers Fearan Ecological Collapse in the Arctic

docmechanic
11pts1
www.semafor.com 10mo ago

A new lawsuit against OpenAI could challenge rule protecting online content

docmechanic
4pts0
www.theregister.com 10mo ago

Silver State goes dark as cyberattack knocks Nevada websites offline

docmechanic
3pts0
www.sciencedaily.com 10mo ago

Leprosy existed in America long before arrival of Europeans

docmechanic
5pts1
www.sciencedaily.com 11mo ago

Tiny microbes may rewire the brain before birth

docmechanic
3pts0
www.sciencedaily.com 11mo ago

Mysterious Denisovan interbreeding shaped the humans we are today

docmechanic
3pts0
www.semafor.com 11mo ago

Van Hollen accuses Trump of pay-to-play on chips

docmechanic
8pts0
www.theregister.com 11mo ago

Claude Code's copious coddling confounds cross customers

docmechanic
7pts1
www.theregister.com 11mo ago

GitHub CEO: Future devs will not code, they will manage AI

docmechanic
3pts2
www.independent.co.uk 11mo ago

Want a glass of wine with dinner? Blame our ape ancestors

docmechanic
1pts0
www.independent.co.uk 12mo ago

The unique human body part that evolution cannot explain

docmechanic
3pts1
www.independent.co.uk 12mo ago

Spiders may have originated in the ocean before adapting to live on land

docmechanic
13pts2
www.semafor.com 1y ago

OpenAI warns against tokenized stock offered by Robinhood

docmechanic
2pts0
english.elpais.com 1y ago

Testing free will: New experiments on the quantum property that baffled Einstein

docmechanic
3pts1
www.semafor.com 1y ago

Cognizant's CEO on staying ahead of the tech jobs 'tsunami' in an AI world

docmechanic
1pts0
www.semafor.com 1y ago

AI coding startup Replit CEO says companies soon won't need software developers

docmechanic
3pts3
www.sciencedaily.com 1y ago

Chimpanzee groups drum with distinct rhythms

docmechanic
3pts1
www.sciencedaily.com 1y ago

Study links fossilized flying reptile tracks to animals that made them

docmechanic
2pts0
www.sciencedaily.com 1y ago

Rare earth element extraction bolstered by new research

docmechanic
4pts1
www.sciencedaily.com 1y ago

Geobiology: Iron, sulfur, heat – and first life

docmechanic
4pts1
www.sciencedaily.com 1y ago

Landmark experiment sheds new light on the origins of consciousness

docmechanic
5pts2
www.sciencedaily.com 1y ago

Anatomy of a 'zombie' volcano: Investigating the cause of unrest inside Uturuncu

docmechanic
6pts2
www.sciencedaily.com 1y ago

Billion-year-old impact in Scotland sparks questions about life on land

docmechanic
3pts1
phys.org 1y ago

A new map of arthropod evolution, from fossils to embryos

docmechanic
3pts1
www.sciencedaily.com 1y ago

Female bonobos keep males in check – not with strength, but with solidarity

docmechanic
6pts1
www.sciencedaily.com 1y ago

The oldest ant ever discovered found fossilized in Brazil

docmechanic
6pts4
www.sciencedaily.com 1y ago

Skeletal Evidence of Roman Gladiator Bitten by Lion in Combat

docmechanic
3pts1
www.sciencedaily.com 1y ago

What rattlesnake venom can teach us about evolution: New USF study

docmechanic
2pts1
www.sciencedaily.com 1y ago

Parts of our DNA may evolve much faster than previously thought

docmechanic
4pts1
www.independent.co.uk 1y ago

Scientists stunned to find chimpanzees sharing boozy treats together

docmechanic
2pts1
[dead] 1 year ago

I’m surprised that the stability of genomic structure at the linear level across species may not be as essential as thought.

[dead] 1 year ago

Amazing that these 'living fossils' survived in these small pockets of forest that are the only remnant of the ancient supercontinent Gondwana. How many other endemic species are clinging on in tiny spots around the globe?

[dead] 1 year ago

I was unaware of the existence of purely self-pollinating plants.

[dead] 1 year ago

I love how chimps separated by different geography developed different drumming patterns.

[dead] 1 year ago

I find it fascinating that a genetic mutation from 7,000 years ago protects from a modern disease less than 100 years old.

Summary: 'A more efficient and environmentally friendly approach to extracting rare earth elements that power everything from electric vehicle batteries to smartphones could increase domestic supply and decrease reliance on costly imports.'

'The findings de-emphasize the importance of the prefrontal cortex in consciousness, suggesting that while it's important for reasoning and planning, consciousness itself may be linked with sensory processing and perception. In other words, intelligence is about doing while consciousness is about being.'

Summary: 'An experiment seven years in the making has uncovered new insights into the nature of consciousness and challenges two prominent, competing scientific theories: Integrated Information Theory (IIT) and Global Neuronal Workspace Theory (GNWT). The findings mark a pivotal moment in the goal to understand the elusive origins consciousness.'

'This model not only challenges traditional distinctions between short-germ and long-germ development—terms that refer to how many segments are defined early in an embryo's life—but also reframes the role of well-known genetic regulators such as Hox genes in defining segment and tagma identity.'

'Female bonobos team up to suppress male aggression against them -- the first evidence of animals deploying this strategy. In 85% of observed coalitions, females collectively targeted males, forcing them into submission and shaping the group's dominance hierarchy. This is the first study to test drivers of female dominance in wild bonobos. The study examined 30 years of demographic and behavioral data across six wild bonobo communities. The study suggests that power isn't solely determined by physical strength. It can be driven by social intelligence and coalition-building by females.'

'A 113-million-year-old hell ant that once lived in northeastern Brazil is now the oldest ant specimen known to science, finds a new report. The hell ant, which was preserved in limestone, is a member of Haidomyrmecinae -- an extinct subfamily that only lived during the Cretaceous period. These ants had highly specialized, scythe-like jaws that they likely used to pin or impale prey.'

'A new study found that some rattlesnakes are producing simpler venoms containing fewer and more focused toxin families than complex venoms -- a surprising discovery that challenges long-held ideas about how living alongside a variety of other species influences evolution in a world increasingly shaped by human activity.'

Source: Generative Deep Learning by David Foster, 2nd edition, published in 2023. From “Tokenization” on page 134.

“If you use word tokens: …. willnever be able to predict words outside of the training vocabulary.”

"If you use character tokens: The model may generate sequences of characters that form words outside the training vocabulary."

Pretty sure that we’re talking apples and oranges. Yes to the arbitrary byte sequences used by tokenizers, but that is not the topic of discussion. The question is will the tokenizer come up with words not in the training vocabulary. Word tokenizers don’t, but character tokenizers do.

Source: Generative Deep Learning by David Foster, 2nd edition, published in 2023. From “Tokenization” on page 134.

“If you use word tokens: …. willnever be able to predict words outside of the training vocabulary.”

"If you use character tokens: The model may generate sequences of characters that form words outside the training vocabulary."

That’s only true if you tokenize words rather than characters. Character tokenization generates new content outside the training vocabulary.