my_library

16,958 bookmarks · page 28 of 340
2026-01-12 · cavendishlabs.org

Sampling at negative temperature

Note: Looking at the equations above, if T<0T<0 then the sign of the exponent flips. That means that the states that were previously the least likely are now the most likely, and vice versa. As temperature approaches zero from the negative side, the model output will again be deterministic — but this time, the least likely tokens will be output.

Looking at the equations above, if T<0T < 0T<0 then the sign of the exponent flips. That means that the states that were previously the least likely are now the most likely, and vice versa. As temperature approaches zero from the negative side, the model output will again be deterministic — but…
2026-01-07 · thebeliever.net

The Worst Shot Ever Taken - Believer Magazine

I was sixteen when my mother gifted me a painting by Ernie Barnes. I’d seen one Barnes painting at that point on the cover of Marvin Gaye’s 1976 album, I Want You, on the brown carpeted floor of my Auntie Sue’s apartment. Then I saw it five days a week for at least a decade…

2026-01-07 · arstechnica.com

New battery idea gets lots of power out of unusual sulfur chemistry

Rather than being used as a storage material, the sulfur gives up electrons.

sulfur is a great storage material for lithium
Chinese researchers has managed to turn sulfur’s complex chemistry into a strength, making it the primary electron donor in a sodium-sulfur battery that also relies on chlorine for its chemistry.
2026-01-07 · marktechpost.com

Implementing Softmax From Scratch: Avoiding the Numerical Stability Trap - MarkTechPost

Learn about implementing Softmax from scratch and discover how to avoid the numerical stability trap in deep learning projects.

This function implements the Softmax activation in its most straightforward form.
numerically unstable—large positive logits can cause overflow, and large negative logits can underflow to zero. As a result, this version should be avoided in real training pipelines
2026-01-04 · worksinprogress.co

The triumph of logical English - Works in Progress Magazine

English prose has become much easier to read. But shorter sentences had little to do with it.

We think we are looking at a language that got simpler
Sentences can be simple (one main clause), compound (two or more main clauses linked with a conjunction), complex (a main clause with dependent clauses, i.e. clauses that cannot stand alone), or complex-compound (one or more dependent clauses and two or more main clauses).
2026-01-04 · semiengineering.com

Impact Of On-Chip SRAM Size And Frequency On Energy Efficiency And Performance of LLM Inference (Uppsala Univ.)

A new technical paper titled “Prefill vs. Decode Bottlenecks: SRAM-Frequency Tradeoffs and the Memory-Bandwidth Ceiling” was published by researchers at Uppsala University. Abstract “Energy consumption dictates the cost and environmental impact of deploying Large Language…

Our simulation methodology combines OpenRAM for energy modeling, LLMCompass for latency simulation, and ScaleSIM for systolic array operational intensity.
Counter-intuitively, high compute frequency can reduce total energy by reducing execution time and consequently decreasing static energy consumption more than the resulting dynamic power increase
2026-01-03 · clearerthinking.org

What makes something a cult? Here is what our data say

Short of time? Read the key takeaways. 💡 People rely on relational dynamics to judge cultishness. When people rate groups as cult-like, they focus most on signs of control, isolation, and suppression of outside information rather than on unusual predictions or strange practices…

2026-01-03 · bigthink.com

Why your brain needs everyday rituals

Rituals serve psychological functions that go far beyond mere habit or tradition.

Shared routines make cooperation easier in times of stress
Lastly, rituals transmit culture across generations