It really is crazy. I have been contributing the Melee decompilation project for the past year-ish, and things have really accelerated in 2026. Just today I decided it would be nice to have a better "permuter" (program that randomly modifies C in the hopes of finding a better asm match) so I...just asked Claude to make one, custom-tailored to my needs. It almost feels pointless to publish it to GitHub when I can just tell the other contributors "hey fyi you can ask Claude to make you a better permuter"
You should not swear at LLMs, for the same reason you should not shout a slur even if no one is around to witness it: You witness it, and witnessing yourself being toxic updates you in the direction of "I am capable of toxicity" and eventually "I am toxic." In other words, it stains your soul.
I 100% agree with you, which is why I try to be polite to the LLMs. Also, on a practical note, I've never seen an agent session go right-side-up after I crash out; the model seems to perform worse than if I just gently guide it.
LLM writing tells are getting more subtle, but they still jump off the page for me, in particular the word "genuine:"
"This is the area where Go genuinely shines, and it’s worth being precise about why"
"the lack of GC pauses is a genuine selling point"
"Humans are genuinely bad at reasoning about memory"
"There are cases where the borrow checker is genuinely too strict"
tbc I don't think the article was fully AI-generated, just AI-assisted. If so, the author did a genuinely good job of it! No one else is commenting on it, so clearly it didn't detract much from the substance. It's just weird that this is becoming increasingly common, and increasingly hard to detect.
I don't know about the author's background, but there is now a generation of non-native programmers who learned to write English by using LLMs for corrections (yeah including this comment).
The irony is that studies show LLM detectors have a much higher false-positive rate for non-native speakers [1]. If most of what you read stems from LLMs, you end up writing like an LLM.
> but there is now a generation of non-native programmers who learned to write English by using LLMs for corrections (yeah including this comment)
LLM writing has not been overly abundant for more than a couple years. I don't know where you got the idea that an entire generation of people have already learned to write like an LLM.
I have to agree here, but I'm not sure why. I don't have any clue what makes something sound AI generated or not. I got to about here "Go is clearly working for a lot of people," -- before I became suspicious that it was AI-assisted (but also maybe I'm wrong and it's not AI-assisted, I am very bad at telling). It's more about vibes (ironically) than anything else in particular. If something "sounds" AI-assisted then I instantly lose interest even if the article itself is otherwise fine. I wish people were more ok with writing their own thoughts with how it comes to them.
Agreed. In fact, one of the things I now watch for is my mind starting to "slide off" the text, or finding myself re-reading a section multiple times. It's like the brain subconsciously recognizes a lack of substance even if we can't point to a specific tell.
I've noticed LLM writing over the past year has had an unusually high tendency to talk about surfaces and, in particular, substrates. I don't expect LLM generated text to be anything other than rich with clichés. I simply wish we would all demonstrate a better editorial hand so we weren't reading the same voice, over and over.
Perhaps more people are using AI as part of an editorial process that is largely driven by what they wish to convey but where they have stopped fighting the AI on its preferred style. It’s supremely annoying when AI updates your prose with its own formulation despite plenty of instructions otherwise. Too often AIs mangle meaning which can be especially worrisome as it’s not easy to catch subtle word/grammar changes that dramatically shift meaning. Overall though, defects aside, for me, and only very recently, it’s been more helpful than not. I think AIs will continue to improve in this regard and be better editorial partners. For competent writings, it won’t replace human authorship or expert review.
Specifically, I’ve recently used ChatGPT for legal/administrative writing where the AI seems to be trained on a large corpus and seems to know the conventions and vocabulary well; a lawyer who reviewed the work had important corrections. Before AI, I would have sought model filings and have had less success at emulating the genre. So it’s lowered time/cost somewhat but it takes lots of diligence. By default, current AI outputs seems intelligible but are still really far off the mark. I’ve found a structured interview is a good way to start rather than jumping into draft generation.
Author here. I use the term 'genuinely' too often, but that's just me. I do that when speaking here as well. Suffice to say that I'm not a native speaker, so that might have something to do with it. I will go over the text and replace some of those. thx.
And it’s a good contrast with ‘just fcking use Go’ article he linked.
Go article is much more human. I love that and would choose a human centered language and human centered culture over LLM-centered everything every time
the psychosis has gone off the charts! anything that sounds odd to someone can now be labelled as an LLM text "smell"?
why scoff over someone doing assisted writing? i might age myself but kids back in the day would try to sound better by using synonym feature in ms word (or through web thesaurus) for their assignment essays. this all looks familiar to the same practice, now only made more accessible.
There's nothing in the comment you're replying to that could be described as "scoffing." What are you on about?
I feel the opposite, where AI hype is so extreme that merely someone pointing out an article may have had LLM involvement prompts a response like this. Someone incredulously painting people as ivory tower nose thumbers. If anything, it pushes me away from LLM writing more.
I also don't see how you can compare finding a synonym for a word to having your entire writing voice determined for you.
While reading the article, I remember feeling that I'm reading an LLM generated sentence a few times, but in general, this specific article look like an example of acceptable LLM usage to me. I wouldn't call it "AI slop".
It is, if I may say that, _genuinely_ hard to use LLM assist and not make the text look like LLM generated. Even when I write an email in gmail and it gives its suggestions to make the text better, each one individually makes perfect sense, but when I click a few of them, the whole email now looks like AI slop, so I would normally undo the changes, going back to my imperfect hand-written non-optimized version.
I also wonder if it's possible that this is just "blog-speak"
The author of this article has what seems like it could be a relatively thriving consulting business, so he probably writes more to advertise his services than anything else. That kind of writing surely lends itself to a particular writing style, which is a non-insignificant chunk of the kind of writing that LLMs were trained on.
I think the whole post is AI generated. The author could have given a draft as input and perhaps edited the output in a few places.
Take this paragraph as example:
> Go got generics in 1.18, and they’re useful, but the implementation has constraints (no methods with type parameters, GC shape stenciling, occasional surprising performance characteristics). Rust generics monomorphize, each instantiation produces specialized code with zero runtime cost. Combined with traits, this gives you real zero-cost abstractions.
Every sentence says something. Every sentence is important and holds its weight. I would expect that kind of writing from very specialized books or papers, not from a blog post. Also, it makes the post harder (and more boring) to read.
> Every sentence says something. Every sentence is important and holds its weight. [...] Also, it makes the post harder (and more boring) to read.
I actually prefer that style of writing! (When it's not AI-generated ofc.) And I also try to use it in my technical blog posts. I usually re-read my drafts asking myself: "Does the reader actually care about this? Is this sentence adding something or is it just fluff?"
And actually I feel like AI text usually produces more fluff, or anyway I notice it more, but I see how it can make the result "robotic and boring".
His stuff about generics in Go is also wrong. He says that Go's standard library "avoids" them. He forgot that it has `slices.SortFunc()`. He forgot about Go `Seq`. Maybe because he has stopped using Go and is no longer that familiar.
> After a disorienting visit from the FBI in May of 1990, I wrote a rant called Crime and Puzzlement, which led to my establishing with Mitch Kapor (who had previously founded Lotus Development Company) an organization called the Electronic Frontier Foundation.
It's interesting to revisit Brooks' "surgical team" in light of AI. For example, I frequently have Claude act as a "toolsmith", creating bespoke project-specific tools on the fly, which are then documented in Skills that Claude can use going forward. What has changed is that a) One person (or rather, one person-AI hybrid) plays all the roles within the surgical team, and b) Internal frictions such as cost, development time, and communication overhead have all been dramatically slashed.
How well does that work for you ? It's annoyingly inconsistent for me - I give it instructions on how to fetch JIRA ticket with a script that renders everything relevant to a .md and half of the time it will still default to reading it via ACLI. I have instructions on how to do a full build with warnaserror before commit but I still get pipeline errors regularly because it will skip the noincremental part, etc.
I have a harness for Claude Code "hooks" (https://code.claude.com/docs/en/hooks) which in my case execute a Go tool in a separate project which runs changes made by claude through a validator with various rules that can be defined (regex, semgrep, etc.). They can warn claude or they can block changes outright.
When I find claude is using tools or approaches that I have replaced with more specific ones, I ask claude to add a hook to prevent doing this in the future and point it to the instructions of what to do instead.
And of course I wrapped all that up in a Skill so it knows what approaches to take to add things to hooks.
It becomes fairly trivial to incrementally stop it making repeated mistakes like this.
You may want to try out pi-agent and create custom extensions instead.
Then codify this behavior into a process which automatically gets run through.
I.e. $repo/origin as bare repo, then prompt to create a shell script which creates the worktree and cds into it, running the script you mentioned, instantiating pi in it.
Potentially define explicit phases for your workflow and show the phase in the UI - and quality gates for transitions. Eg force the implment to finalize phase to only happen if all tests succeeded. Potentially add multiple review phases here too, with different prompts. This progressively gets rid of more and more inconsistencies.
Still not a perfect solution, but on average I've had less and less to manually address with that workflow. Albeit at cost of tokens (multiple reviews phases obviously ingest all changes multiple time)
Pi-agents extensibility is just a lot better then the other harnesses, but you could obviously also just introduce a different orchestrator to do the same. For me, pi-agent was just the least amount of effort necessary to get it going.
So why not fetch the ticket deterministically yourself as part of "staging" all the information that Claude will need? I.e. if Claude can't be relied upon to do something the way you want it done, take that task away from it, perform it some other way, and feed the results back to Claude.
I don't use AI much so I don't know if this is reasonable.
I rebuilt an app I found in rust and extended it in a bunch of ways that I use everyday for this use case and it works flawlessly if this is any help: https://github.com/dchuk/jarkdown-rs
On a local model, with open code, when I wrote a specific javascript way to run sql queries because bash and psql were error prone, what I did was when I saw it make a mistake, I told it in passive agressive tones something like: "please edit AGENTS.md to detail how to use the query.js tool to run a query and to never use psql"; I did this two times until it stopped wanting to use psql.
It seems like if you write the docs yourself that's not leveraging the probability that the model itself knows the anti-context guard rail that best prevents it from grabbing its average tool use.
> frequently have Claude act as a "toolsmith", creating bespoke project-specific tools on the fly, which are then documented in Skills that Claude can use going forward.
I also do this.
e.g. after watching Claude burn tokens building and then deploying a docker image multiple times (and it taking extra time), I asked it to just create a build.and.deploy.sh script. I also then have a test.deploy.sh script that Claude can use to confirm everything worked.
Saves a ton of time/tokens AND has the added benefit of being usable by me or other humans when doing manual tests or debugging outages etc.
I do something similar, but tell the agent to write a recipe into a justfile. Then it can run `just` and get a self-documenting list of all the tooling for the project (just build, just test, etc.)
I had a similar experience recently. I had a huge Swagger JSON file that would waste too many tokens if added directly to the context, so I told the agent in memory to use jq to retrieve what it needs when it wants to check this document. This saved a lot of cost.
LLMs are not inherently non-deterministic. This is a common misconception. You used to be able to set temp=0 and a fixed seed and get the same output every time. This broke when labs started implementing batching, and no one bothered fixing it because the benefits of batching vastly outweighed the demand for deterministic output.
I am hopeful deterministic output will return, though; DeepSeek v4 claims to have implemented "bitwise batch-invariant and deterministic kernels," though I haven't tested it myself.
Reproducible does not mean deterministic. You cannot determine in advance what a prompt will give as output, even with a temperature of 0 and a fixed seed, therefore they are not deterministic.
Huh? I'm not aware of anyone else who defines "deterministic" that way. "Deterministic" comes from "determinism," as in "the effects are fully determined by the causes" -- not "determine" as in "deduce."
Sufficiently-developed concentration gives you access to the jhanas, which are extremely blissful states of consciousness. Having reliable access to high valence reduces your need to seek pleasure in less wholesome things (drugs, food, twitter, etc.)
Sufficiently-developed attention gives you insight into how your brain is constructing what you perceive as reality, leading to a reduction in ego, permanent reduction in baseline suffering, and a pervading sense of unity with the rest of the universe.
I'm old enough to remember when companies worth $1 billion were called "unicorns." Now we have a company raising 122 times that? Valued at nearly 1000 times that...?
At least they're throwing consumers a bone via the ARK deal. It's crazy how little AI exposure is available to anyone who isn't already wealthy and/or connected.
I think this is reality-distortion field rivaling that of Jobs', and a crisis of faith. Nobody apparently believes that capital is worth investing into anything but AI.
> Nobody apparently believes that capital is worth investing into anything but AI.
This is the main reason we see this insane investment into AI imo. If you imagine having lots of money, where should you invest that currently?
Housing market: Seems very overvalued (at least in germany). Also with the current uncertainty and inflation its hard to make an investment that pays back over 20-30 years. So building is also difficult.
Stocks are very volatile currently. Not only since Iran. To me it seems since the financial crisis 2008 investors don't enjoy stocks as before.
Gold: Only if you are paranoid about collapse of society. It doesn't make sense to invest into s.th. without interest rates.
Crypto: Same as gold, but better if you like gamling. I would assume most people who are very rich don't gamble with most of their fortune.
Looking around, and especially forward, it would be military tech, e.g. [1], and its supply chain, e.g. [2] :-\ Valuations are not as crazy, but I bet there'll going to be a lot of demand in the coming decade, unfortunately.
Chip production, too, of course, but it's overflowing with money already, apparently. It's growing though, because there are real actual shortages of stuff like RAM and SSDs, there's money to be made immediately if you can. Chinese RAM manufacturers are building out like crazy.
Anduril is the only company in this sector in the US that has any promise and they aren't even public. Most of us are not going to get our hands on this.
Traditional defense sector looks more like Jeep, or Kodak...
Anduril has yet to deliver anything of consequence. I hope they shake up the industry but to say they are the next hot thing and write off the primes at this stage is premature.
Would you be fine with the ethical implications of funding the industry to fight WWII? Would you consider funding Ukrainian military unethical? Or Taiwanese?
This is, sadly, not theoretical, and I'm afraid we'll soon see more of such choices, not fewer.
"The Roaring Twenties roared loudest and longest on the New York Stock Exchange. Share prices rose to unprecedented heights. The Dow Jones Industrial Average increased six-fold from sixty-three in August 1921 to 381 in September 1929. After prices peaked, economist Irving Fisher proclaimed, "stock prices have reached 'what looks like a permanently high plateau.'"
You can argue that current market multiples are higher than 1929 [1] - and they're certainly high - but this also ignores the mechanism that drove that crash, focusing only on the symptoms. We simply aren't doing the kind of consumer margin buying that drove the '29 crash. It isn't even close. Average schlubs were leveraged to the stratosphere to buy shares of boring industrial stocks.
> The US stock market has nearly tripled since then. Literally the best period of stock growth in history.
The only thing I meant to point out was that a very high stock price by itself is no guarantee that there isn't a crisis around the corner. We plugged a lot of holes after 2008 and then reversed a lot of those fixes, I hear retail investors talking about their stocks at birthday parties again. Deja vu... of course this time it will be different. Or not. Let's just say that with the proverbial bull in the earthenware goods store on the loose if we only end up with another financial crisis that might actually not be so bad.
I actually calculated wrong. It went up 7.5x, not 3x.
In the roaring twenties stockbrokers allowed clients 10:1 margin. Investors were not as well-informed as they are today. There was no deposit insurance.
The SEC wasn't nearly as powerful as it was in 2024 and there was way more shady shit going on. In that respect, and the repeal of Glass-Steagall we're reverting to the pre-depression era.
Do you know the actual lessons of that crash? Because we don't allow retail investors to go 10:1 on leverage anymore. There are a lot more lessons and none of them apply to this situation (even Glass-Steagall). This is much closer to the dot com crash in 2001 in how it looks, just a lot more concentrated and probably a bit bigger. If all you got is "number go up too much" then you probably shouldn't be investing your own money.
The good news is that its almost all rich folks money on the line here and a small amount of dumb money. That's very different than, 2008 where it was mostly the indexes that got hit and that's more middle class/upper middle class concentrated.
That's the tao of hyper-financialization. It must keep growing irrational exuberance big and up forever like stonks or it bursts like DotCom and tulip mania. It's funny money that cannot be liquidated for real value for more than a tiny fraction of the imaginary trillions being thrown around. Similarly, Nvidia $4T mkt cap makes absolutely no sense when it has but a few incestuous customers-parters-investors throwing around tens of billions each per year devoid of fundamentals like essential service offerings that turn a profit. Those handful of whale customers will make their own chips or cease buying large qtys at any time.
And not even actual capital either, as much of the investment amounts into AI have been through cloud and GPU credits so that AWS or Microsoft Azure don't actually have to hand over billions in straight cash.
It's the result of too much echo chambered bullshit floating around daily about how capable LLMs really are. It's literally crypto/blockchain all over again. It's one big lie that a lot of people have bought into which causes it to self-perpetuate, like religion.
ARK funds has cult like following but then again they are a typical high beta player who outperforms in hot markets and heavily underperforms in cold ones.
Fees are high. The CEO (CIO) is a women who looks for investment advice in the Bible and asks God for his thoughts (I am not joking).
If anything being associated with ARK in any form is a big negative signal.
Also, the valuation for such a debt laden company should be viewed with great skepticism. I'm afraid a lot of mutual funds will end up holding the bags.
> At least they're throwing consumers a bone via the ARK deal. It's crazy how little AI exposure is available to anyone who isn't already wealthy and/or connected.
It is deliberate. Period.
It's always been known that you make money in the private markets and pre-IPO companies and retail is the final exit for insiders and early investors.
Retail is not allowed to be early into these companies (Because that would ruin the point of being an insider) and this "exposure" has to be at the near top.
Even a billion dollars is crazy money. If you have a company with a subscription service that costs $100 yearly, you have ~2m customers, with a 50% profit margin. Your company makes ~100m every year in profit. Imo that's what is actually worth a billion dollars, maybe even a bit less.
yep, I do a simple version of this in Google Sheets. Very useful to be able to "Ctrl-F" your life, especially when combined with Google Maps location history.
reply