“Bump_hq” — Fediverse search results on home.social

Barcelona were flying. Real Madrid and PSG brought them back to earth with a bump https://www.nytimes.com/athletic/7726038/2026/10/26/barcelona-real-madrid-clasico-psg-defeats/ Hansi Flick's side are still two points ahead of the pack in La Liga, but defeat by Jose Mourinho's side raises questions #Soccer #Football

#soccer #football

Kyle Memoir 🍉🐧 @[email protected]

2026-10-25

Today, I realized I'm so old I remember when elders used to fret if the fridge door was open more than five seconds because it would bump the hydro bill.

Well over 50% of my current electricity cost is 'delivery' and various bullshit charges included therein.

#energy #climatechange #environment

Rachel J Graves @[email protected]

2026-10-26

Apple pushing creator studio everywhere inside pages, numbers and (whatever the slides one is called) is so annoying I’ve uninstalled the suite.

If there’s a filter to hide all the premium features and just let me use the software I’d love to know.

Most annoying is I’ve actually paid for these apps once upon a time.

It’s a strange direction imo since so many companies offer apps and document storage for free. At least make it added value to my Apple One subscription? Then when they bump its price it seems fairer.

#Apple #CreatorStudio #Pages #Numbers

#apple #creatorstudio #pages #numbers

Fish in the Percolator @imrehg

2026-10-26

I pulled the trigger on a Linux kernel update a bit too eagerly, and the ZFS version I was running wasn't ready yet, there goes the dynamic kernel module update compilation...

Roll back? 🤔
💡 Not quite, the compatibility change just shipped while I was building, I mean failing to build the last version. Here's to bump-and-rerun! 🚢

#bleedingedge #linux

Fossery Tech :debian: :gnome: @[email protected]

2026-10-25

DeepSeek-V4.1-Flash-1017 brings faster project building:
https://alternativeto.net/news/2026/10/deepseek-v4-1-flash-1017-brings-faster-project-building/

Meta launches Muse Spark Open, an 816B open-weight AI model for Macs requiring 1TB of RAM:
https://alternativeto.net/news/2026/10/meta-launches-muse-spark-open-an-816b-open-weight-ai-model-for-macs-requiring-1tb-of-ram/

FLUX 3 Image Klein adds 4K image editing on 8GB GPUs:
https://alternativeto.net/news/2026/10/flux-3-image-klein-adds-4k-image-editing-on-8gb-gpus/

Strata 0.2.1 Released: Vision, Benchmarks, Fixes, and Docker Images:
https://www.linuxcompatible.org/story/strata-021-released-vision-benchmarks-fixes-and-docker-images/

Strata 0.2.2 Released With Memory-Light Gyro Variant:
https://www.linuxcompatible.org/story/strata-022-released-with-memorylight-gyro-variant/

Strata 0.2.3 Adds Intel Arc Support and Strix Halo Speed Bump to 125B-Parameter PC Model:
https://www.linuxcompatible.org/story/strata-023-adds-intel-arc-support-and-strix-halo-speed-bump-to-125bparameter-pc-model/

Strata 0.2.4 Released: Run a 763-Billion-Parameter AI Model Locally on Your Gaming PC:
https://www.linuxcompatible.org/story/strata-024-released-run-a-763billionparameter-ai-model-locally-on-your-gaming-pc/

Microsoft’s AI Coding Agent Testing Tool Has Been Open-Sourced:
https://itsfoss.com/news/microsoft-scope-self-hosted-agent-testing/

You Can Reserve This Small-Factor, De-Googled Phone for $30:
https://itsfoss.com/news/brax-wink-pre-order/
(Pretty good option for Blackberry fans I guess)

VirtualBox 7.2.22 Releases with Security Fixes:
https://www.linuxcompatible.org/story/virtualbox-7222-releases-with-security-fixes/

OpenBSD 8.0 brings Clang/LLVM 22 base compiler, 64 VCPUs VM & Intel Arc A-Series support:
https://alternativeto.net/news/2026/10/openbsd-8-0-brings-clang-llvm-22-base-compiler-64-vcpus-vm-and-intel-arc-a-series-support/

Red Hat Funded a Report That Found a Quarter of Incoming Pull Requests May Be AI Slop:
https://itsfoss.com/news/red-hat-lf-research-ai-slop-report/

FFmpeg 9.1 Preps For Release With APV Vulkan Encoder, NVIDIA Optical Flow FRUC:
https://www.phoronix.com/news/FFmpeg-9.1-Releasing

Vulkan 1.4.366 Released With SMPTE ST 2094-50 HDR Support:
https://www.phoronix.com/news/Vulkan-1.4.366

(Post 7/7)

#WeeklyNews #OpenSource #FOSSNews #FOSS #OpenSourceNews #News #DeepSeek #Muse #MuseSpark #FLUX #Strata #AI #AIAgent #AgenticAI #BraxWinkModularSmartphone #BraxWink #VirtualBox #OpenBSD #OpenBSD8 #FFmpeg #Vulkan #FosseryTech

#weeklynews #opensource #fossnews #foss #opensourcenews #news

Dev Community @[email protected]

2026-10-25

I have watched the same release go wrong three times. Someone tunes a prompt or swaps a model, the eval average moves from 0.81 to 0.83, the PR gets approved, and two weeks later a specific intent is failing in production. The regression was real. The improvement was not. Nobody did anything careless — the eval reported a number that looked like a measurement and was actually a coin flip.

The unglamorous part of LLMOps is not prompt management or tracing. It is that most eval suites cannot detect the changes teams ship on. Fixing that is a half-day of statistics and one CI job.

Three kinds of noise, only one of which is your change

Before you compare two configurations, know what is moving:

Example sampling noise. Your golden set has 200 examples. Each one is worth 0.5 points. A "3-point gain" can be six examples flipping, several of which are borderline judgments.
Judge noise. If an LLM grades the outputs and you left temperature above 0, the same output can pass on Tuesday and fail on Wednesday. Order matters too: ask a judge "which is better, A or B" and then "which is better, B or A" and you will find disagreement.
Task noise. Sampling temperature, tool timeouts, and retrieval index drift all change outputs without any code change on your side.

You cannot remove all three. You can remove the first one almost entirely by comparing paired results: run baseline and candidate on the same examples and analyze the per-example difference. Example difficulty cancels out. A prompt that fixes one hard case and breaks another now shows up as one +1 and one -1, not as an aggregate you have to reason about.

Do the arithmetic before you build the gate

For a paired comparison of pass/fail outcomes, the standard error of the mean difference depends on the disagreement rate — how often the two configs disagree on the same example — not on the accuracy.

If 15% of examples flip between your two configurations and you have 200 paired examples:

SE = sqrt(0.15 × 0.85 / 200) ≈ 2.5 percentage points
Minimum detectable effect at 80% power, α=0.05 ≈ 2.8 × SE ≈ 7 points

So on that setup, a 2-point bump is invisible. To see a 5-point change under the same assumptions you need roughly 400 paired examples; to see 3 points, about 1,100. These are arithmetic, not benchmarks — plug in your own disagreement rate, which is usually higher than you expect for prompt changes.

Two consequences follow. First, stop gating on raw averages. Second, if your golden set is 200 examples and you cannot grow it, your gate should be "no regression beyond X" rather than "improvement required."

The gate itself

The whole thing is about thirty lines of stdlib Python. Bootstrap the paired differences, look at the interval, and ship only when the interval clears zero.

python
"""paired_gate.py — decide whether an LLM change is real or noise.
Run: python paired_gate.py Deps: none (stdlib only)."""
import random

def paired_bootstrap(baseline, candidate, iters=5000, seed=7):
"""Scores aligned on the SAME example ids, 0/1 or 0..1.
Returns (mean delta, 2.5th pct, 97.5th pct)."""
diffs = [c - b for b, c in zip(baseline, candidate)]
n = len(diffs)
rng = random.Random(seed) # fixed seed: the gate must be reproducible
means = sorted(
sum(diffs[rng.randrange(n)] for _ in range(n)) / n
for _ in range(iters)
)
return sum(diffs) / n, means[int(0.025 * iters)], means[int(0.975 * iters)]

def required_n(disagreement_rate, mde, z_alpha=1.96, z_power=0.84):
"""Paired binary comparison: examples needed per side."""
se = mde / (z_alpha + z_power)
p = disagreement_rate
return int((p * (1 - p)) / se**2) + 1

if __name__ == "__main__":
# 200 paired examples; candidate looks "+3 points" better.
baseline = [1] * 160 + [0] * 40
candidate = [1] * 166 + [0] * 34
delta, lo, hi = paired_bootstrap(baseline, candidate)
print(f"delta#ai #llm #devops #python #software #coding #development #engineering #inclusive #community
Your Eval Suite Has a Flake Rate: Gating LLM Changes on Paired, Powered Tests

#devops #ai #community #software #coding #python

D. Moonfire @[email protected]

2026-10-24

#WritersCoffeeClub October 23:

How do you make sure you deliver on your writerly promises?

I try not to make any. I have a think about authors who announce "this will be X books" and then have to write a few hundred pages of filler to make it happen.

But I do write a lot of commissioned works and I strive to finish those. I may not always make a deadline, so I don't usually give one to the commissioner, but I will get it done unless I find I'm unable to for some reason; since I don't get paid until I finish, the only thing I waste is time.

Many years ago, I decided to take some advice from Jim C. Hines and Scott Lynch that I heard at ICON (IA, which is today) which was "write more words, don't be a dick". What followed was that I wrote a chapter or story a week for three hundred consecutive weeks.

https://d.moonfire.us/blog/2021/03/17/three-hundred-weeks/

I put a lot of effort into making sure I had good buffers to avoid missing a week. I wrote programs to ensure it was published. I made sure I was going to post on Thursday every day for basically six years.

And then I burned out, which is where we are. Five years and I only published one book, write two stories, and have been working on my fifth novel ever since.

The only real promise I make is that I will not have sexual assault in my books. No matter how grim, or dangerous, it isn't there. It isn't an option, it isn't even a concern. It does not exist.

Occasionally, though, I make a mistake (fucking reality). When I do, I go back and edit the chapter/story, bump up the version, and then move on. I'll acknowledge it because I live in a world were SA does exist, but I have loved ones who never want to worry about that. So, no SA by anything written by D. Moonfire.

#writerscoffeeclub

Search