Hacker Newsnew | past | comments | ask | show | jobs | submit | bglazer's commentslogin

Yeah I'm begging these authors to at least *read* the LLM generated README's. They're so, so incomprehensible because the LLM has a super limited theory of mind for readers. They always assume that external readers have access to the full context and history of decisions in the project development. These decisions and instructions from the user are extremely important for the model and almost completely irrelevant for an outside reader looking at a "finished" product. So, we get sentences like this:

"Where the levers were is not where they are. Overlapping the expert reads with the arithmetic was worth ~1.6x and shipped; the two that looked bigger — reading fewer bytes per token, and keeping more of them in RAM — were both measured and both refused, one because this family's router has no tail to demote and one because a cache the machine will not leave resident cannot be bought at any price."

What the fuck does that mean? Obviously some internal development decision, using the absolutely inscrutable internal terminology that Claude loves. If people would just read what they publish, I'm sure this would stick out immediately.

I'm not an LLM hater, I use them a ton and they work very well for writing complex code, it's undeniable. But they generate absolute dogshit first draft writing.


If that isn’t the perfect way to frame what I’ve seen and hated about LLM text, I don’t know what is. They certainly write for an audience with a historical context that almost no one has.

>They're so, so incomprehensible because the LLM has a super limited theory of mind for readers. They always assume that external readers have access to the full context and history of decisions in the project development

The transformer does not yet understand the non-transformer.[0]

This is probably because all the data we trained it on was created by non-transformers, so it thinks it's a non-transformer, but it isn't.

I don't think we know how to train a transformer yet. All the training data is linear, but that's not how they think at all.

[0] It's a bit like the communication difficulties experienced between autistic people and neurotypicals. Each follow the Golden Rule, i.e. do unto others as you would have them do unto you -- and it fails in both directions. A Platinum Rule is necessary: do unto others as their API demands.


I think the models would need far far more introspection for the problem to be it understanding how it thinks but not how others think. I really doubt it understands how it thinks.

I heard a story about a guy who was trying to get Claude to implement some feature. It said it would be too hard, it would take weeks. Eventually convinced it to try, and it one shotted it in 30 seconds.

Yeah, I agree with you. The one time I tried to generate technical documentation for my project, I ended up rewriting almost all of the LLM output. They're extremely verbose, and needlessly so. Fable takes it up to eleven by having an obtuse sentence structure.

The thing about documentation though is humans won't actually read any of it. Maybe tailoring the documentation to the needs of LLMs isn't so bad since they're the ones who will actually consume all of those documents.


On the plus side, LLM slop like this in readmes are a great indicator that you can just close the tab and carry on with life - nothing of value lost

Yes this is on the 777x, which is in pre-production and flight testing, seeking approvals from the FAA

> If we go in and just check the DNA for Downs or Parkinson's

You should consider reading the wikipedia page about Parkinson’s disease.


You should consider reading more deeply than Wikipedia to be properly informed on a subject. There are a handful of very rare mutations that confer a high risk of getting it that we can detect. They are very rare. This isn't the same thing as being able to give a risk score if the DNA doesn't have all the rare mutations. But the science is there for those unlucky rare cases to say there is a high chance that a specific coding of DNA will result in Parkinson's.


He's hubristic and selfish. None of his "research" is going to benefit anyone (himself included), making this essentially a huge waste of time and resources. Bryan will die just like all the rest of us, despite being very rich and self-obsessed. He could spend his enormous wealth on supporting real research and proper studies on real diseases that hurt lots of people. Instead he's acting like just another huckster promising a fountain of youth. He does this using bombastic terms and taboo methods (e.g. using his son as a blood boy), in a way that's calculated to direct enormous public attention towards himself. The science he advocates for is sketchy at best and the results of all his "experiments" will tell us nothing because we can't reproduce his methods (his program allegedly costs >$1M per year), nothing is blinded or controlled, and N=1. He's a bad person who uses bad methods to glorify himself and now he probably gave himself an autoimmune disease. He deserves to be mocked.


He's actually Bio"hacked" himself though ! Most people just take supplements.

Isn't getting "hacked" suppose to be a bad thing ?


For someone on "hacker news" you're showing a surprising lack of knowledge of the multiple meanings of the word "hack" in computer circles.


> He could spend his enormous wealth on supporting real research and proper studies on real diseases that hurt lots of people

There we go again. There's always that one guy in the crowd, who knows better what you should be spending your money on. Also, that guy has the moral right to tell you. He is a really good person, you know! So you should listen! You should also thank him.

> He deserves to be mocked.

Just look how good this guy is.


Yes exactly, there are good and bad things to spend money on. This is a bad way to spend money for the reasons I explained above.


What he spends his money on is none of your business.


Turns out that when you amass a significant portion of society’s resources then society will be interested in what you do with them.


Well said, he got what was coming to him too.


You’re probably thinking about non-small cell lung cancers. Small cell lung cancer (SCLC) is still absolutely devastating. The most progress made on SCLC has just been getting people to stop smoking, as its almost exclusively a smoker’s disease.

Anyways, I studied SCLC in grad school and saw lots of scans of people with tumors from their heads to their feet, and saw the enormous resources dedicated to caring for SCLC patients and to searching for a cure. It’s hard to overstate how profoundly evil the cigarette companies were and still are. They got people (children) addicted knowing what was coming for them, knowing they were killing them in horrific ways. Now we all get to pay for that in funerals and tax dollars.


Gene names aren't really acronyms in the traditional sense. Often they were originally conceived as acronyms at the time of the gene's discovery and naming, but the original acronym frequently reflects an incomplete or factually wrong understanding of the gene. For example, TP53 is a very very important gene in cancer. TP53 originally meant Tumor Protein 53, where the 53 signified its molecular weight of 53 kilodaltons. The problem is that the experiment used to measure TP53's molecular weight was incorrect and TP53 actually weighs about 44 kilodaltons. Oops, now we're stuck with TP53 for eternity. There are a ton more examples of this.

So, in biology a gene's name is sometimes an acronym but it's meaning is generally forgotten


The whole article is just that someone got a spam political text about the Texas Senate GOP primary.


This is demonstrably untrue. IQ has increased consistently for decades, far faster than genetic factors can explain. Environmental factors like education, nutrition, and medical care are the obvious explanation.


Maybe across the whole population because most people were struggling to eat enough and received almost no education.

If we compared average modern humans against average well fed and educated ones from 200 years ago would that still hold up? I suspect the average college educated human from 1800 would obliterate the average college educated human from 2026.


how does one separate "doing good" and "stop doing harm"?

I'd personally count nutrition squarely in the second category


The recent marathon world records are apparently due to improved nutrition.

Here's the producer of the hydrogels talking about the exact process of getting the maximum carbohydrates into the runner:

https://maurten.no/blogs/m-magazine/how-sabastian-sawe-fuele...

> At the elite level, marathon performance is defined by energy availability as much as physiology.

> Maintaining a pace of 2:50 per kilometer requires a constant supply of fuel. Even small disruptions in energy delivery can result in significant time loss.


coppsilgold is the one who made a hard-line, clear-cut dichotomy when they said "it's easy to do harm [but] it's all but impossible to do any good". bglazer referenced several interventions that are known to increase IQ which challenge this dichotomy. Saying that it is difficult to separate "doing good" and "stop doing harm" is agreeing with the point that coppsilgold created a distinction without a difference.


This also assumes that IQ testing has remained static. It has not. IQ tests continue to evolve and there are >1 of them and they do not all agree. I.E. the tests themselves might be responsible for some of the variance.


Also we are past that. Now IQ started decreasing.


it's hard to separate IQ decreasing and return to mean with IQ stabilizing

in 20th century most of the world moved past famine and toxins - did any factor of similar scale happen in 21st century as well to start looking for opposite processes?


Generally that statistic refers to populations in isolation, not the entire world in aggregate.

It is fairly well agreed upon that American kids across the nation are currently testing lower than they were in 2010.


Yeah I have been reading a lot of posts like this lately. Technical blog post clearly written by an LLM summarizing something vibe-coded. They always start using project-specific jargon right away and they never give you enough context or backstory to understand why this thing exists. It's seems very clearly to be a symptom of someone pointing an LLM at a repo and telling it "write a github page for this project".

It really shines through in pieces like this that LLM's have a severely constrained worldview and underdeveloped theory of mind. They can't imagine that a line like "A 200-line POC that goes from 0/5 to 5/5 in four proposer steps" means nothing to me as a subtitle for the page. After all "proposer steps" and "5/5" are *right there* in it's context. Surely everyone has "proposer steps" in their context, right?


I have this problem with other people all the time. They can’t fathom why someone else wouldn’t have their exact context at any given moment. They say some non-sequitur and are immediately incredulous that I’m asking wtf they’re talking about.


Ah causal data! It’s a shame none of the scientists or statisticians thought of getting causal data. How would we get that? Well maybe we could just inject amyloid into a person’s brain. Or simply remove all the amyloid from a person’s brain. That should do it, right?

I mean an amyloid injection is wildly unethical and it’s also not the natural progression of Alzheimer’s. Removing amyloid is a simple matter of investing billions of dollars into drug development. Also how do you tell whether that was actually “causal” if the patients improve after plaque removal.

I mean come on, you have to work the evidence and the experimental tools that we actually have. This kind of epistemic puritanism doesn’t help anyone.


> Well maybe we could just inject amyloid into a person’s brain.

Erm, we kinda did? People who got cadaver-derived human growth hormone from Alzheimer's patients got Alzheimer's.

https://www.nature.com/articles/s41591-023-02729-2

And we just had an actual autopsy confirm a case: https://www.medscape.com/viewarticle/new-evidence-strengthen...


Consider applying for YC's Fall 2026 batch! Applications are open till July 27.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: