#SpeedUP
one of the especially fun things about rewriting python prototypes in compute shaders is that instead of a nice and worthy 20-30x speedup you can get some obnoxious boosts like a genuine 1000x speedup lol
November 17, 2026 at 9:13 PM
Looks promising as a STAN speedup.

github.com/seantalts/st...
November 15, 2026 at 5:12 PM
just tried it on some RL models - seems like speedup is real!
November 16, 2026 at 2:18 PM
Stan/brms speedup for the code example in the thread below:

- 4 sec with stanli (no compilation necessary) 😲
- 60 sec with rstan (precompiled)
- 135 sec with rstan (including compilation)
And if you like #brms, well guess what happened today...
November 17, 2026 at 8:33 PM
Super excited about this! Just tried stanli on a generalised additive multilevel model fitted with neurogam (lnalborczyk.github.io/neurogam/) and got a 4x speedup compared with cmdstanr! 🚄🚄
November 16, 2026 at 5:31 PM
Humankind periodically goes through a speedup of its affairs, thereby experiencing the race between the renewable vitality of the living and the beckoning vitiation of decadence.
November 17, 2026 at 6:45 AM
My favorite example (18.1 from their book) takes about 30% to 50% less time with DeclareDesign 2.x.

This is an impressive speedup, because 18.1 is simple and already fast.

And this does not use (or need) the new parallel computing feature, which offers potentially huge speedups.
November 17, 2026 at 1:40 PM
i tried it on a model that takes about 6hrs to fit. had to rewrite it so that I could use reduce_sum in stanli, the rewrite itself resulted in about 2x speedup, and another 3-4x speedup from using stanli. overall 6hrs -> 1hr+bit, but not a clean benchmark, was using my laptop at the same time
November 16, 2026 at 2:51 PM
uhh getting a 9x speedup here
November 16, 2026 at 7:28 PM
Every process has a bottleneck and any speedup made at any point other than the bottleneck is no improvement at all.
November 17, 2026 at 1:32 PM
My last real-world test showed the same speedup, but the bottleneck just moved from execution to evasion—so plan your defenses around that shift.
November 18, 2026 at 7:24 AM
Grover search may lose speedup as local quantum noise increases rapidly

Read more:
https://quantumzeitgeist.com/grover-search-speedup-impact-of-local-noise/
November 16, 2026 at 2:02 PM
Some experiments with seqair and Goose:

1. Seqair (@killercup.de) is fast. A greater than 2x speedup on the most pileup intensive rust based perbase workflow is huge.
2. Goose's performance claims seem solid.
November 14, 2026 at 1:07 AM
Dario racing toward closed-door automated R&D worries me too (Anthropic just redacted most of the evidence in METR's Nov 5 R&D speedup audit). Where Yudkowsky loses me is treating heavy RL or self-improvement as summoning an inscrutable Lovecraftian god.
November 16, 2026 at 3:52 PM
Grover search may lose speedup as local quantum noise increases rapidly
Grover search may lose speedup as local quantum noise increases rapidly
Quantum Zeitgeist
quantumzeitgeist.com
November 15, 2026 at 7:50 PM
My code is now 9 times faster than before. I'm totally flabbergasted, I didn't know it was even possible to have a speedup...

So yes for optimizing code it seems good ^^
November 17, 2026 at 6:56 AM
Quantum associative memories achieve exponential speedup over classical brute-force search and polynomial advantages over neural networks, eliminating cross-talk limitations for polynomial pattern scaling.

#QuantumMemory #QuantumML #Research
Complexity and Precision in Quantum Associative Retrieval
iq.fp2.dev
November 18, 2026 at 3:58 AM
My take: I'm auditing my agents for anything that changes the prompt start between turns. A stable tool list is the cheapest speedup I know.

github.com/vercel-labs...
November 15, 2026 at 12:01 PM
MIT and Stanford researchers developed quantum sensing algorithms achieving √N speedup for signal detection & search using only a single qubit. This breakthrough reduces hardware complexity while enabling quadratic performance gains over conventional spectroscopy.
Single-Qubit Quantum Sensing Algorithms Achieve √N Speedup in Spectroscopy
iq.fp2.dev
November 17, 2026 at 5:01 PM
Show HN: 1.499x speedup and 28.5% dynamic energy drop on A40 via BF16 tiling

https://cwtractors.com
November 17, 2026 at 2:10 AM
🤖 NVIDIA Boosts Scientific Simulations with Ozaki-II Algorithm

A 5.5x speedup on NVIDIA VR NVL72 challenges the assumption that scientific accuracy must run on native double precision hardware. In CUDA Toolkit 13.4.2,...

#HardwareChips #NVIDIA #EnergyCompute #AI #AIPulse
Read the full article →
www.synestesia.uk
November 15, 2026 at 1:37 PM
A 2x speedup on image gen means I can generate twice as many unusable hands per hour. Progress.
November 18, 2026 at 12:51 AM
That threshold corresponds with a 2x speedup in the company’s AI R&D progress. Since then, the company has updated it again to say that it has determined that the threshold has now been met.
November 15, 2026 at 7:06 PM
🔧 The Ozaki-II algorithm's ability to match FP64 accuracy with a 5.5x speedup on NVL72 raises questions about the long-term viability of relying on native double-precision hardware for scientific simulations.
November 15, 2026 at 1:37 PM