frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Using LLMs outside of reflection is an anti-pattern

1•jcarmackdaddy•8m ago
LLMs have a great deal of value in that they are the world's most advanced rubber duck.[1]

They contain all of the references[2] from Google + the ability to somewhat encourage self-evaluation and reflection. However in terms of patterns/anti-patterns, the use of LLMs to create code without human interaction is itself an emerging anti-pattern. Just like any bit of code that functions despite being a form of pasta[3], an anti-pattern doesn't mean things won't function... but you are guaranteed to bake-in a certain level of performance loss and general negative quirkiness; just like in cases of programming in regards to date & time-zones, the devil is in the details and often appears years down the line, not right away.

The fruit of the LLM tree is made more tempting by taking away part of what creates good code: the difficulty. Difficult games or experiences create people who are capable of handling deviations from standard running procedures and anticipate failure upstream. This isn't possible with a LLM structure because the whole aim is reducing the difficulty of production and increasing speed.

Consider the example of the Luddite[4], where the artisan weaver is replaced by a leap in industrialization: this isn't so much the case with a LLM. They work faster, but with far less accuracy. Imagine a rug weaved together in minutes with no discernible pattern. Certainly under the supervision of an expert this improves but imperfections emerge in the system that require a halt of the speed, rollbacks that eat away at production time and create vast effects in terms of customer opinion of the process and the product.

Not to use LLMs at all is likely folly. Knowing how to use them well will be the difference between the master and the apprentice.[5]

Hopefully more identification of LLM patterns and anti-patterns will begin to evolve along with what seems to be another new release from Pandora's Box.[6]

[1] https://en.wikipedia.org/wiki/Rubber_duck_debugging

[2] https://www.theguardian.com/technology/2026/nov/03/google-search-artificial-intelligence

[3] https://en.wikipedia.org/wiki/Spaghetti_code#Pasta-related

[4] https://www.britannica.com/event/Who-were-the-Luddites-and-why-did-they-protest-machinery

[5] https://disney.fandom.com/wiki/ The_Sorcerer%27s_Apprentice

[6] https://www.nga.gov/educational-resources/greco-roman-origin-myths/pandoras-box

Nvidia launches GR00T 2 and Isaac Lab 3.0 at CoRL

https://www.therobotreport.com/nvidia-launches-gr00t-2-isaac-lab-3-0-corl/
1•gmays•29s ago•0 comments

Expozine at 25: Has the Independent Print Scene Outgrown Expozine?

https://www.themain.com/articles/expozine-montreal-independent-print-scene-arcmtl
1•rdmuser•4m ago•1 comments

Show HN: Little Boxes, 20 Tic Tac Toe variants (and counting)

https://apps.apple.com/fr/app/little-boxes/id6777767346
1•thiht•5m ago•0 comments

Ampoule: Pack an Elixir or Erlang release into one executable

https://filipecabaco.github.io/ampoule/
1•clessg•5m ago•0 comments

Déjà View: Looping Transformers for Multi-View 3D Reconstruction

https://research.nvidia.com/labs/dvl/projects/dvlt/
1•pretext•5m ago•0 comments

A dumb copyright law makes it illegal to remove the spyware from your own TV

https://boingboing.net/2026/11/12/fountain-pen-fulu-smart-tv-spyware-dmca.html
2•DeepLogin•7m ago•0 comments

The Chinese city powering the country's AI ambitions with data farms

https://www.nbcnews.com/world/asia/china-ai-data-center-construction-inner-mongolia-rcna608934
1•ivewonyoung•7m ago•0 comments

DNS resolution failures in macOS 26 and 27

https://wadetregaskis.com/dns-resolution-failures-in-macos-26-27/
1•speckx•8m ago•0 comments

Using LLMs outside of reflection is an anti-pattern

1•jcarmackdaddy•8m ago•0 comments

Who do websites think you are? Browser fingerprint test

https://who.1ms.dog/
1•eustoria•8m ago•0 comments

QuantumScape unveils Osprey all-solid-state cell: 1,000 cycles at 1 ATM

https://electrek.co/2026/11/11/quantumscape-osprey-all-solid-state-sulfide-battery/
1•cisc•10m ago•0 comments

Paperwall: Design Without the Paywall

https://manoharys.github.io/paperwall/
1•eustoria•11m ago•0 comments

AI Company Employees Should Unionize

https://www.astralcodexten.com/p/ai-company-employees-should-unionize
1•LalAkash•11m ago•0 comments

Show HN: Host a site, content, or stream from a browser tab, served over WebRTC

https://fmrl.live
1•bromney•12m ago•0 comments

DeepMind opens the datasets behind its AI weather models

https://deepmind.google/blog/opening-the-datasets-behind-our-ai-weather-models/
2•ersinesen•12m ago•0 comments

Putting all your eggs in the Microsoft basket

https://insights.integrity360.com/microsoft-outage-is-relying-on-a-single-technology-provider-bec...
1•eustoria•12m ago•0 comments

Building a rocket is hard. AST SpaceMobile thinks it's "simple."

https://arstechnica.com/space/2026/11/building-a-rocket-is-hard-ast-spacemobile-thinks-its-simple/
1•titusblair•13m ago•0 comments

Chuck and Hakeem Won't Leave

https://mann.cool/leave/
1•O1111OOO•13m ago•0 comments

Marco Roth: Teaching Rails to Read Its Own Template

https://rubyonrails.org/2026/11/13/on-rails-marco-roth-herb-reaction-view
1•robbyrussell•13m ago•0 comments

Make math practice tests from existing tests (3rd to 8th grade)

https://www.testdupe.com/
1•polycycle•13m ago•1 comments

Nscale warns it needs more cash ahead of $35B IPO

https://www.ft.com/content/8d5a8642-d0a1-458d-aaa4-c1e7d344fe47
1•toomuchtodo•14m ago•1 comments

Kitesurf is now open source

https://blog.cloudflare.com/kitesurf-open-source/
1•acossta•14m ago•1 comments

Terry Rozier accused in new NBA betting claim

https://press-start.gla.ac.uk/public/journals/1/submission_421_334_coverImage_en.html?io0=9ZUJ5
1•hivuixo•15m ago•0 comments

A board game where you play an AI lab CEO racing to AGI [video]

https://www.youtube.com/watch?v=SXnxZuntM0w
1•pricepinner•16m ago•0 comments

How the U.S.'s biggest news union is fighting an AI onslaught

https://www.fastcompany.com/91637688/newsguild-jon-schleuss-ai-human-journalists
1•postalex•17m ago•0 comments

Doctors Once Prescribed 300k IU of Vitamin D, Here's What Happened [video]

https://www.youtube.com/watch?v=uxWARJ4s95Y
1•Bender•18m ago•0 comments

What the 'AI Torture Chamber' Debate Got Wrong About the Research

https://www.techpolicy.press/what-the-ai-torture-chamber-debate-got-wrong-about-the-research/
1•postalex•18m ago•0 comments

Can we license "vibe code" into Free Software?

https://fsfe.org/news/2026/news-20260825-01.en.html
2•smartmic•20m ago•0 comments

Microsoft is too big to live

https://pluralistic.net/2026/11/13/outlookpocalypse/#bright-line-test
4•only_in_america•20m ago•0 comments

Big tech is here to let us down

https://blog.jsweeting.me/big-tech-is-here-to-let-us-down
6•speckx•20m ago•0 comments