rickintoplace

And finally, we need those clip-on Game Boy lights again.

On why we shouldn't:

- bacterial populations are being regulated by bacteriophages (viruses) or predators. They have spent millions of years evolving to hunt standard chiral life and wouldn't be physically able to digest or hunt mirror bacteria.

- the same is true of our immune system. Our pattern-recognition receptors only work for non-mirror molecules and wouldn't recognize a reversed molecular fingerprint. You may be able to engineer antibiotics and phages, but a vaccine would require a creative approach.

- the claim "yes but mirror bacteria would starve since they wouldn't find mirror food to eat in nature or in our bodies" doesn't hold up, because there are enough achiral (non-handed) nutrients in the soil, our blood and our gut that mirror bacteria could eat.

Here is an example where you can see it in action:

https://navignine.com/c/g4PDSWbLgy2HPLROeDfz#key=1f84bcb51d7...

Or just start a new chat and try it out for yourself.

This is a link to the mentioned research tool, but it's not used commercially anymore. As long as the rate limits allow it, you can play around with it a bit as a demo without a login.

I recently measured something similar on whether a quote supports the claim it's cited for. I tested on 240 examples from ALCE (labeled by human annotators).

Jev 1.13 flagged 87.5% of the citations that the annotators marked as not supported. The best regular LLM judge I tried was GLM-5.3-flash with reasoning which flagged 83.9% (without reasoning it was 75.6%).

The same also works with open models though. Give Qwen3.5-397B the three possible answers, let it output a single token, and read off the probability of each answer. No reasoning, one token per check. That flagged 89.3%.

There were only 56 unsupported cases, so these numbers are within each other's error bars. But my conclusion was that the speed and the confidence score come from reading out a single token, not from Jev itself.

Thanks :) I'm also building a tool for spoonerisms that might be useful for rhyming: https://wordlab.rickinto.place/spoonerize

It uses the same data as the misheard tool and produces cross-pairs like these: Be known Knee bone

No guts Go nuts

And there is a rude word filter for SFW that can also be inverted for extra "edginess"

I initially built this as a toy and it turned into an interesting small phonetics project.

Enter a sentence and it finds other ways to break down the same sounds into English words: “kiss the sky” gives you “kiss this guy”. “Four candles” turns to “fork handles”. “Ice bank mice elf” ... I think you get the idea

Each word is looked up in the CMU Pronouncing Dictionary. The sentence is converted into a sequence of phonemes, and a beam search breaks it back down into words. Exact re-parsing is rare, so sounds may be distorted under certain conditions. For example, keep in mind that a double consonant at a word boundary sounds like a single one (mistake/miss steak). A /ɡ/ before a nasal almost completely disappears (recognize -> “reco'nize”). The sound alone is not enough to distinguish “wreck a nice beach” from “reckon eyes beach,” so variations that actually appear in 40 million lines of movie subtitles are also taken into account.

The system was calibrated using 42 known oronyms (38 in the top 12, 25 in first place), and I disabled each rule individually to see how much it contributed.

What doesn’t work: long sentences, proper nouns (“Euthanasia” -> “Youth in Asia” is impossible, since names are filtered out), and it only works with American English, since CMUdict only contains that.

Static site, no server, no LLM. Code: https://github.com/rickintoplace/wordlab

That's actually fun to watch. Did you experiment with nicknaming before you turned it off? I'd be a little curious to see how it behaves.

Looks great :) It would be nice to be able to manually tweak the plotting algorithm or some underlying metrics.