frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

LLMs Don't Quite Beat Classical Hyperparameter Optimization Algorithms

https://github.com/ferreirafabio/autoresearch-automl
3•achierius•1h ago

Comments

achierius•1h ago
One of those rare papers where the code speaks for itself. They do a bunch of comparisons but the most salient is comparing Karpathy's autoresearch (verbatim, best as I can tell) vs. some HPO algorithms, and as of yet the Tree-structured Parzen estimator still wins out -- but just barely!

More interesting though is that the best results come from 'centaur' approaches, where an LLM is hooked up with a standard HPO. Somewhere around 1:3 LLM:HPO control seems to work best, with more LLM control degrading performance. But either way this method far outperforms either the naive autoresearch loop or the bare HPO approach.

achierius•1h ago
> Centaur outperformed all methods including CMA-ES alone by using the LLM on only 30% of trials. The LLM receives CMA-ES's full internal state (mean vector, step-size, covariance matrix), the top-5 configurations, and the last 20 trials. A 0.8B LLM already suffices to outperform all classical and pure LLM methods. Scaling from 0.8B (0.9766) to 27B (0.9763) to Gemini Pro (0.9767) yields no improvement, suggesting a capability plateau [which Claude sliightly beats]

> We ablate the LLM ratio: higher ratios degrade performance, confirming that CMA-ES should retain majority control.

Sniffing EU Smart Meters with a Flipper Zero (WM-Bus / 868MHz)

https://github.com/i12bp8/wmbuster
1•i12bp8•3m ago•0 comments

GlowGoblin – a gift to mb pro users

https://github.com/jtc268/glowgoblin
1•husky8•8m ago•0 comments

Japan intervened in its Curreny price

https://www.fxstreet.com/news/usd-jpy-drops-over-2-as-intervention-warnings-lift-yen-after-move-a...
1•mark336•9m ago•1 comments

US telecom agency votes to expand tech crackdown on China

https://www.reuters.com/business/media-telecom/us-telecom-agency-votes-expand-tech-crackdown-chin...
2•l2dy•10m ago•0 comments

Anthropic Model inference runs fastest on AWS

https://twitter.com/theo/status/2050078772507124134
1•albert_e•10m ago•0 comments

How People ask Claude for personal guidance

https://www.anthropic.com/research/claude-personal-guidance
1•pseudolus•11m ago•0 comments

No action taken against PimEyes: noyb lawsuit against Hamburg DPA

https://noyb.eu/en/no-action-taken-against-pimeyes-noyb-lawsuit-against-hamburg-dpa
1•latexr•12m ago•0 comments

The Starlink hack that doomed Russian troops [video]

https://www.thetimes.com/video/originals/article/starlink-hack-doomed-russian-troops-jgr0w9jkp
1•petethomas•16m ago•0 comments

Insights into software job openings – may2026

https://corvi.careers/blog/global_software-engineering_jobs_may_2026/
2•sp1982•19m ago•0 comments

Shivon Zilis Operated as Elon Musk's OpenAI Insider

https://www.wired.com/story/model-behavior-why-everything-in-musk-v-altman-leads-back-to-shivon-z...
1•aanet•29m ago•0 comments

Japan Airlines begins humanoid robot trials at Tokyo's Haneda airport as labor s

https://www.cnbc.com/2026/05/01/japan-airlines-humanoid-robots-haneda-labor-shortage.html
2•bookmtn•33m ago•0 comments

Elon Musk Seemingly Admits xAI Has Used OpenAI's Models to Train Its Own

https://www.wired.com/story/elon-musk-distill-openai-models-partly-xai/
3•satai•40m ago•0 comments

LA-SF rail will cost about $126B, with service beginning around 2040

https://ktla.com/news/california/california-high-speed-rail-cost-increase/
3•swyx•43m ago•0 comments

Show HN: Git repositories hosted directly on Freenet

https://github.com/freenet/freenet-git
3•sanity•44m ago•0 comments

U.S. Aims to Penalize Disabled Adults Who Live with Their Families

https://www.propublica.org/article/trump-social-security-ssi-disability-benefits-cuts-parents-chi...
3•petethomas•44m ago•0 comments

AliothPress – self-hosted, cloud-native CMS with a wizard-based installation

https://aliothpress.com/
3•Strodt•51m ago•0 comments

One solar storm could trigger a catastrophic collision in orbit

https://spectrum.ieee.org/kessler-syndrome-crash-clock
2•thread_id•51m ago•0 comments

Show HN: A vdom-less framework that's local first

https://github.com/terajs/terajs
2•thecodergabe•53m ago•0 comments

AI Tips and Tricks

https://www.youtube.com/watch?v=w_m5RmVsmtE
2•frag•54m ago•0 comments

Show HN: ChatToMap – Scans your chats to find all your activity and trip ideas

https://chattomap.com
1•nathan_f77•57m ago•0 comments

Kolmogorov Complexity [video]

https://www.youtube.com/watch?v=UGN9D0n4AJA
1•sgschlesinger•58m ago•0 comments

10x Faster Real-Time High-Quality AI Video Generation

https://tenstorrent.com/solutions/real-time-video
1•montyanderson•1h ago•0 comments

LLM Summaries Are Ruining Your Learning

https://arpitbhayani.me/blogs/do-not-rely-on-summaries/
1•menonNN•1h ago•0 comments

AI and the Future of News 2026

https://reutersinstitute.politics.ox.ac.uk/news/ai-and-future-news-2026-what-we-learnt-about-its-...
2•jruohonen•1h ago•0 comments

What do you think of people buying Mac mini's to run AI?

1•namegulf•1h ago•3 comments

Task-Specific LLM Evals That Do and Don't Work

https://eugeneyan.com/writing/evals/
1•eigenBasis•1h ago•0 comments

Mark Zuckerberg Says AI Costs Contributed to Layoffs of 8k Staffers

https://www.forbes.com/sites/antoniopequenoiv/2026/04/30/mark-zuckerberg-says-ai-costs-contribute...
2•ZeidJ•1h ago•0 comments

Workforce Transparency Act [pdf]

https://www.warner.senate.gov/wp-content/uploads/2026/04/Warner.-Budd-Workforce-Transparency-Act.pdf
1•petethomas•1h ago•0 comments

Call centres dismantled and ten arrested in EUR 50 million online fraud case

https://www.europol.europa.eu/media-press/newsroom/news/call-centres-dismantled-and-ten-arrested-...
1•jruohonen•1h ago•0 comments

C8s: A Confidential Kubernetes Architecture

https://arxiv.org/abs/2604.26974
2•badcryptobitch•1h ago•0 comments