HNHacker News
TopNewBestAskShowJobs

adamzwasserman

261 karma · joined February 2, 2022

IT Strategist, Startup Positioner, Cargo Cult Programmer

https://github.com/adamzwasserman

submissionscomments
adamzwasserman··on How the Brain Parses Language
On effect size: my primary goal at this stage is falsification. If French and English models show no meaningful differences at matched compute, that's informative: it would support the scaling hypothesis. If they do differ, I'll need to be careful about causal claims, but it would at least challenge the "transformers are magic" framing that treats architecture as the main story.

The L2 regularization and information theory pointers are helpful, it will go on my reading list. If you have favorites, I'll start there.

On the "we know nothing" point: I'm sympathetic. The Stroop example is exactly why I'm skeptical of strong claims in either direction. 197k papers and no mechanism suggests language processing has properties we don't yet have frameworks to describe. That's not mysticism. It's just acknowledging the gap between phenomenon and explanation.

adamzwasserman··on How the Brain Parses Language
This is great, thanks for the link. IMHO it actually supports the broader claim: if ICL emerges in both language models and genomic models, it suggests the phenomenon actually is about structure in the data, not something special about neural networks or transformers per se.

Genomes have statistical regularities (motifs, codon patterns, regulatory grammar). Language has statistical regularities (morphology, syntax, collocations). Both are sequences with latent structure. Similar architectures trained on either will repeat those structures.

That's consistent with my "instrumentation" view: the transformer is revealing structure that exists in the domain, whether that domain is English, French, or DNA. The architecture is the microscope; the structure was already there.

adamzwasserman··on Show HN: TabHere – AI autocomplete for almost any editable field on the web
from the chrome store. I will just go to gihub
adamzwasserman··on Show HN: TabHere – AI autocomplete for almost any editable field on the web
I am getting a "this item not available"
adamzwasserman··on Show HN: TabHere – AI autocomplete for almost any editable field on the web
interesting idea. I will try it out and feedback as soon as I can squeeze out a few moments
adamzwasserman··on How the Brain Parses Language
Thank you for the Leiden references. I hadn't encountered this framework before. The "language symbiont" framing resonates with what I've been circling around: a system that operates with its own logic, sometimes orthogonal to conscious intention.

The mule analogy is going to stick with me. LLMs have inherited the statistical structure of the symbiont without the host: pattern without grounding. Whether that makes them useful instruments for studying the symbiont itself, or just misleading simulacra, is exactly what I'm trying to work out.

Going to dig into Kortlandt tonight.

adamzwasserman··on How the Brain Parses Language
The confound concern is fair: no cross-linguistic comparison is perfectly controlled. The bet is that the effect size (if any) will be large enough to be informative despite the noise. But you're right that it's not ceteris paribus in a strict sense.

Your proposal is interesting though. Synthetic manipulation of morphology within a single language. Have you seen this done? The challenge I'd anticipate is that "genderized English" wouldn't have natural text to train on, so you'd need to generate it somehow, which introduces its own artifacts. But comparing French vs artificially gender-neutralized French might be feasible with existing parallel corpora. Worth thinking about as a follow-up.

On the neural network → brain distance: agreed it's a leap. The claim isn't that transformers are brains, but that if both are extracting structure from language, they might reveal something about what structure is there to extract. Fedorenko's own comparison to "early LLMs" suggests she thinks the analogy has some merit.

adamzwasserman··on How the Brain Parses Language
These are good points that sharpen the hypothesis. The word order question is interesting — positional encoding vs morphological encoding might have different computational properties for a parser.

One difference I'm betting on: morphological agreement is redundant (same information marked multiple times), while word order encodes information once. Redundancy aids error correction and may lower pattern extraction thresholds. But I'm genuinely uncertain whether that outweighs the structural information carried by strict word order.

Do you have intuitions on which would be "easier" for a statistical learner? Or pointers to relevant literature? The vocabulary size / morpheme count tradeoff is also something I hadn't fully considered as a confound.

adamzwasserman··on How the Brain Parses Language
Incredibly, I had not thought to use that data set.

Now I will. Thanks.

adamzwasserman··on How the Brain Parses Language
I don't think we're far apart. My claim isn't that Level 1 is "mere reflex". It's that language can produce effects at a level that operates independently of (and sometimes in opposition to) conscious evaluation. The hypnosis example is just a clean demonstration of that separation.

Whether LLMs are useful models for studying that level is an empirical question. They're not conscious, but they do learn statistical regularities in language structure, which may be exactly what Level 1 is optimized for.

adamzwasserman··on How the Brain Parses Language
The hypnotism worked. There was an unconscious process at work whose relationship to the words was completely different from my conscious reaction.
adamzwasserman··on How the Brain Parses Language
The dyslexia point is interesting; yes, English orthography causes more reading disorders than languages with more regular spelling-to-sound mappings (Italian, Finnish, etc.). That's consistent with the parser having to work harder when the signal is noisier.

Your intuition about "slower more intentional parsing" connects to something I'm exploring: we may parse language at two levels simultaneously; a fast, nearly autonomic level (think: how insults land before you consciously process them) and a slower deliberate level. Whether those levels interact differently across languages is an open question.

adamzwasserman··on How the Brain Parses Language
The body of knowledge on Ericksonian hypnotherapy is pretty clear that the effect of language on Level 1 is orthogonal to, and sometimes even opposed to, conscious processes.

I became interested after being medically hypnotized for kidney stone pain. As the hypnotist spoke, I was consciously thinking: "this is dumb, it will never work." And yet it did.

That's exactly your point — I was fully conscious and "at home" the whole time, yet something was processing and acting on the language independently. The question is whether that something shares any computational properties with LLMs, not whether the whole system does.

adamzwasserman··on How the Brain Parses Language
The disanalogy you quote might actually be the key insight. What if language operates at two levels, like Kahneman's System 1/2?

Level 1: Nearly autonomic — pattern-matched language that acts directly on the nervous system. Evidence: how insults land before you "process" them, how fluent speakers produce speech faster than conscious deliberation allows, and the entire body of work on hypnotic suggestion, which relies on language bypassing conscious evaluation entirely.

Level 2: The conscious formulation you describe — the translator between perception and meaning.

LLMs might be decent models of Level 1 but have nothing corresponding to Level 2. Fedorenko's "glorified parser" could be the Level 1 system.

adamzwasserman··on How the Brain Parses Language
There's an interesting falsifiable prediction lurking here. If the language network is essentially a parser/decoder that exploits statistical regularities in language structure, then languages with richer morphological marking (more redundant grammatical signals) should be "easier" to parse — the structure is more explicitly marked in the signal itself.

French has obligatory subject-verb agreement, gender marking on articles/adjectives, and rich verbal morphology. English has largely shed these. If you trained identical neural networks on French vs English corpora, holding everything else constant, you might expect French models to hit certain capability thresholds earlier — not because of anything about the network, but because the language itself carries more redundant structural information per token.

This would support Fedorenko's view that the language network is revealing structure already present in language, rather than constructing it. The "LLM in your head" isn't doing the thinking — it's a lookup/decode system optimized for whatever linguistic code you learned.

(Disclosure: I'm running this exact experiment. Preregistration: https://osf.io/sj48b)

adamzwasserman··on If you're going to vibe code, why not do it in C?
Fair point on UB — LLMs absolutely do not reason about it (or anything else). They just reproduce the lowest-common-denominator patterns that happened to survive in the wild.

I’m not claiming the generated C is “safe” or even close. I am sure that in practice it still has plenty of time-bombs, but empirically, for the narrow WASM tasks I tried, the raw C suggestions were dramatically less wrong than the equivalent JavaScript ones — fewer obvious foot-guns, better idioms, etc.

So my original “noticeably better” was really about “fewer glaring mistakes per 100 lines” rather than “actually correct.” I still end up rewriting or heavily massaging almost everything, but it’s a better starting point than the JS ever was.

adamzwasserman··on If you're going to vibe code, why not do it in C?
Practical argument for vibe coding in C: the training corpus is probably much higher quality.

C code that survives in the wild tends to be written by experienced devs who care about correctness. The JS corpus includes mountains of tutorial code, Stack Overflow copypasta, and npm packages with 3 downloads. In my experience, generated C is noticeably better—and this might be why.

adamzwasserman··on [dead]
I wonder what it was you were taught growing up.

My grandfather was a through-and-through Communist - a Red Russian who came to the US after escaping a Nazi firing squad and living in the forest for months. Once in the US, he organized labor, and got blackballed by McCarthy. And yet what he taught me, for as long as I can remember, was that life was not fair and I had no business expecting it to be.

He taught me that if I failed, it was no one's fault but my own. He taught me to live by my wits - how to survive outdoors in winter, how to shoot, how to fix machines, how to work wood. He raised me to survive a collapse of civilization because he had lived through one.

For him, self-reliance wasn't opposed to his politics. It was forged by them. I'm not sure what to make of your post, but I wanted to offer a different picture of what a Communist might teach.

adamzwasserman··on JSON Schema Demystified: Dialects, Vocabularies and Metaschemas
Yep!
adamzwasserman··on JSON Schema Demystified: Dialects, Vocabularies and Metaschemas
I am dark at the moment.

I am doing final testing, packaging, integrating strip, etc. My target was before end of month. If I work hard I just might make it.

I'm releasing a bunch of things at once: 1. multicardz.com 2. a paper on using DOM as single source of user state dataos.software 3. an OSS repo of hyper performant implementation of that for React (stateless.software) 4. an OSS repo that might best be described as Tailwind for frontend behavior. (genX.software)

they are all somewhat interconnected

adamzwasserman··on JSON Schema Demystified: Dialects, Vocabularies and Metaschemas
The deeper principle here: "stop lying about where truth lives."

I've been exploring how this generalizes beyond side effects. Every React state library creates a JavaScript copy of state that must sync with the DOM. This is the original sin. Two truths = lies.

The solution isn't better syncing, it's refusing to duplicate. The DOM is already a perfectly good state container. All you have to do is read it.

Releasing a paper (DATAOS) and React implementation (stateless, <1KB) soon. It's the architecture behind multicardz (hyper-performing kanban on steroids, rows AND columns, 1M+ cards, sub second searches, perfect lighthouse scores, zero state sync bugs). Because there's no state to sync.

adamzwasserman··on We're losing our voice to LLMs
I did, but it did not have much to add
adamzwasserman··on We're losing our voice to LLMs
See above
adamzwasserman··on We're losing our voice to LLMs
see above
adamzwasserman··on We're losing our voice to LLMs
It did some light spell checking and suggest I add one word - see above
adamzwasserman··on We're losing our voice to LLMs
here is the entire exchange. you decide:

Me: I want to reply to this HN article:

We're Losing Our Voice to LLMs Social media has become a reminder of something precious we are losing in the age of LLMs: unique voices. Over time, it has become obvious just how many posts are being generated by an LLM. The tell is the voice. Every post sounds like it was posted by the same social media manager. If you rely on an LLM to write all your posts, you are making a mistake. Your voice is an asset. Not just what you want to say, but how you say it. Your voice is unique. It is formed from your lifetime of lived experiences. No one's voice will be exactly like yours. Your voice becomes recognizable. Over many posts it becomes something people subconsciously connect with, recognize, trust, and look forward to. Your voice provides the framework for the impression you leave in a job interview, while networking at a meet-up, or with a co-worker. Years ago I got a job thanks to my blog posts. A manager wanted my voice influencing their organization. Your voice is an asset. Your voice matures and becomes even more unique with time and practice. LLMs can rob you of that voice, and the rest of us lose something precious in the process. Having an LLM write "in your voice" is not the same. Your voice is not static. It changes with the tides of your life and state of mind. Your most impactful message may come because it was the right moment and you were in the right frame of mind. Let your voice grow with use. Let it be unique. Do not let one of your greatest assets fade into atrophy, wilted by cognitive laziness. Write in your own voice. I do not care what the linguistic remix machine juggles into being. I care what you have to say. Join my newsletter. Get my posts, best coupons, and most interesting links direct to your inbox. Email address Join © 2025 Tony Alicea. All rights reserved. tonyalicea.dev

Here is my proposed reply:

It is not a zero sum game.

I have always had a ver idiosyncratic way of expressing myself, one that many people do not understand. Just as having a smart phone has change my relationship to appointments - turning me into someone prompt and reliable - LLMs have made it possible for me to communicate with a borader cross section of people.

I write what I have to say, I ask LLMs for editing and suggestions for improvement, and then I send that. SO here is the challenge for you: did I follow that process this time?

I promise to tell the truth.

Claude: Identified typos and validated strong argumentative content.

Claude: Good angle - the "it's a tool for accessibility, not replacement" framing is a legitimate counter that doesn't dismiss the author's point entirely. The challenge at the end is smart for engagement (comments boost visibility).

Typos to fix:

"ver idiosyncratic" → "very idiosyncratic" "has change my" → "has changed my" "borader" → "broader" "SO here" → "So here" Claude: One question: the smartphone/punctuality analogy is a bit of a stretch - do you want to tighten it or cut it? It's not essential to your point.

Me: it illustrates how - as a cyborgh - I am more sucessful personally with the aid of technology

Claude: Validated cyborg framework strengthening personal capability through technology.

Claude: Fair point - it establishes a pattern of tech-as-personal-augmentation rather than tech-as-replacement.

Claude: Do you want to make the "cyborg" framing explicit in the post? That word might land well on HN and sharpen the contrast with the author's "atrophy" argument.

Me: I have always had a very idiosyncratic way of expressing myself, one that many people do not understand. Just as having a smart phone has changed my relationship to appointments - turning me into a prompt and reliable "cyborg" - LLMs have made it possible for me to communicate with a broader cross section of people.

Claude: Validated cohesive metaphor and polished prose for clarity.

Claude: That works. "Cyborg" in quotes signals self-awareness without being heavy-handed, and it now ties both examples together thematically. Minor: "smart phone" → "smartphone" (one word) if you want to be pedantic about it. Otherwise the post is clean and ready.

adamzwasserman··on We're losing our voice to LLMs
hahahahahaha
adamzwasserman··on We're losing our voice to LLMs
I hear you. LLMs have been a major improvement in my life in that regard.
adamzwasserman··on We're losing our voice to LLMs
True, I do not do that for friends
adamzwasserman··on We're losing our voice to LLMs
I have been writing for professionally since I published my book in 2017 (with human editors): https://a.co/d/6ZFFb2z

If you want to read my unedited writing most of the articles on Hacker Noon prior to this year are "raw" https://hackernoon.com/u/azw

Most (all except the last two) of these are "raw": https://emusings.substack.com/

I never garnered a large readership. When I was the CTO at a large NGO, they literally called me "big yoda" because they considered my pronouncements so inscrutable. <shrug>

← PreviousPage 4 of 7Next →