Gentle tip from a lifelong aviation enthusiast: wait one week before reading on causation.
Exposing yourself to first-week speculation isn’t just unproductive, it’s often counterproductive since the actual findings can rhyme with the false speculation closely enough that you wind up muddling the two in your mind.
I think it's neat that this summary is written by an author of the scientific manuscript. Oversimplification is a risk, but this approach eliminates the possibility that the writer did not understand the underlying science.
I'd like to offer a cautionary tale that involves my experience after seeing this post.
First, I tried enabling o3 via OpenRouter since I have credits with them already. I was met with the following:
"OpenAI requires bringing your own API key to use o3 over the API. Set up here: https://openrouter.ai/settings/integrations"
So I decided I would buy some API credits with my OpenAI account. I ponied up $20 and started Aider with my new API key set and o3 as the model. I get the following after sending a request:
"litellm.NotFoundError: OpenAIException - Your organization must be verified to use the model `o3`. Please go to: https://platform.openai.com/settings/organization/general and click on Verify Organization. If you just verified, it can take up to 15 minutes for access to propagate."
At that point, the frustration was beginning to creep in. I returned to OpenAI and clicked on "Verify Organization". It turns out, "Verify Organization" actually means "Verify Personal Identity With Third Party" because I was given the following:
"To verify this organization, you’ll need to complete an identity check using our partner Persona."
Sigh I click "Start ID Check" and it opens a new tab for their "partner" Persona. The initial fine print says:
"By filling the checkbox below, you consent to Persona, OpenAI’s vendor, collecting, using, and utilizing its service providers to process your biometric information to verify your identity, identify fraud, and conduct quality assurance for Persona’s platform in accordance with its Privacy Policy and OpenAI’s privacy policy. Your biometric information will be stored for no more than 1 year."
OK, so now, we've gone from "I guess I'll give OpenAI a few bucks for API access" to "I need to verify my organization" to "There's no way in hell I'm agreeing to provide biometric data to a 3rd party I've never heard of that's a 'partner' of the largest AI company and Worldcoin founder. How do I get my $20 back?"
I like their Material Expressive design a lot more now after Apple's big reveal. While it's still a bit too colorful and whimsical for my liking, it does stay much closer to what I think should be the top UX design ideal - be clear, legible, and get out of the way. On the new iOS every screen feels like some UX designer is shouting "look how amazing I am at this!!" at me.
I feel like I need a button on HN for, as another commenter put it, "folksy wisdom porn", where an article superficially touches all the right buttons to get it to the front page (hey, I always fail to reach my goals, I need a new framework!), but is just anecdotes and shows the results of the author's own Rorschach test.
The section on NASA made absolutely no sense to me:
> NASA had a fixed budget, fixed timeline, and a goal that bordered on the absurd: land a man on the moon before the decade was out. But what made it possible wasn’t the moonshot goal. It was the sheer range of constraints: weight, heat, vacuum, radio delay, computation. Each constraint forced creative workarounds. Slide rules and paper simulations gave us one of the most improbable technological feats in history.
Wut? The constraints are what made it a hard problem, but the only reason they were able to hit this goal in an impossibly short timeline is the huge amount of resources that they put toward a very clear goal (which was, honestly, less "let man explore the heavens" than "beat the Soviets").
Because this isn't about that. This is about having a perceived enemy that only you can fight. If it wasn't immigrants (legal or illegal), it would be a different group, within or outside of your borders.
It's fascism 101.
If illegal immigration is such a problem, why not fine businesses 5x salary for using the labor, for as long as it was used? There are a lot of systems in place to verify working status at this point. It eliminates any incentive to hire this cheaper labor willing to work for lower wages.
The people coming will be coming for a variety of reasons but it won't be to take the jobs of the uneducated Americans
There's a reason you separate military and the police. One fights the enemies of the state. The other serves and protects the people. When the military becomes both, then the enemies of the state tend to become the people.
I installed it. I really wanted to love it but it’s bad. It’s very busy and the proportions in the Settings app are awful. It’s on the “cozy” side of things (as opposed to “compact”). This means you see less options at one time on the screen and have to scroll more around the OS to get where you need to.
As for accessibility… It’s hell. Have a look: https://imgur.com/a/6ZTCStC
This is also likely a performance nightmare. Funny that they mention that "new hardware has enabled us to..." which means that this will perform poorly on old devices.
At a previous company, we were forbidden from using translucency (with a few exceptions) because of the performance cost of blending. There are debugging tools we'd use fairly often to confirm that all layers were opaque.
The whole thing is Windows Vista Aero Glass and iOS 7 all over again. Repeating all the SAME mistakes with 3D translucent design.
Right now I really want skeuomorphism back.
Much like iOS 7 they will have to spend another 2 - 3 years "tweaking" or basically walking back some of these design decisions.
I believe the problem is when Tim Cook decided to merge "Design" under one umbrella. So the Design team now takes over both Hardware and Software Design when they kicked Scott Forstall out. A lot of Apple's UX went down hill from there.
Oh no. It looks like every button and menu is now a translucent layer, so that any noise from the background shows through and muddles the text. This seems like an accessibility nightmare.
Translucent layers generally make software unusable for me. In the video, I saw several instances that would be really really bad for me, where I’d be straining to understand the text. Looks really cool and futuristic though. Just like a movie. Big whoop.
I’m autistic, but this won’t only affect autistic people. A lot of people are going to have problems with this. I hope there’s a very prominent way to turn it off.
I'm really showing off my age here, but it has been all down hill since skeuomorphic design; because the focus was primarily on usability and teachability as first-class concepts. Heck, companies were spending millions on usability research at the time, much of which was used.
I taught people to use computers in the 90s and early 2000s, and having those concepts matching to real world objects helped immensely. Recently I had to teach my kids to use a PC (they no longer teach that in "computers" at school, by the way, iPads only), and everything was arbitrarily designed without even internal rules/consistency let alone building on real-world metaphors.
You've also had this ongoing trend of content density getting consistency worse, and now Apple is accelerating a trend to make UI elements difficult to see/harm discoverability further. Liquid Glass is going to be a painful period, and all the clones that do it even worse are going to be pure hell.
Okay, the AI stuff is cool, but that "Containerization framework" mention is kinda huge, right? I mean, native Linux container support on Mac could be a game-changer for my whole workflow, maybe even making Docker less of a headache.
A lot of people don't know what this Section 174 is about, so here's a brief explainer.
Normally, when you have expenses, you deduct them off your revenue to find your taxable profit. If you have $1 million in sales, and $900k in costs, you have $100k in profit, and the government taxes you on that profit.
Section 174 says you can't do this for software engineers. If you pay a software engineer, that's not "really" an "expense", regardless of the fact that you paid them.
What you've actually done, Congress said, is bought a capital good, like a machine. And for calculating tax owed, you have to depreciate that over several years (5 in this case).
Depreciating means that if you pay an engineer $200k in a year, in tax-world you only had $40k of real expense that year, even though you paid them $200k.
So the effect is that it makes engineers much more expensive, because normally when a company hires an engineer, like they spend on any other expense, they can at least think "well, they will reduce our profit, which reduces our tax obligation," but in this case software engineers are special and aren't deductible in the same way.
In the case of the $200k engineer, you deduct the first $40k in the first year, then you can expense another $40k from that first year in the second year, the third $40k in the third year, and so on through the fifth year. So eventually you get to expense the entire first year of the engineer's pay, but only after five years.
The effect is that companies wind up using their scarce capital to loan the federal government money for five years, and so engineers become a heavy financial burden. If a company hires too many engineers, they will owe the federal government income tax even in years in which they were unprofitable.
These rules, by the way, don't apply to other personnel costs. If you hire an HR person or a corporate executive, you expense them in the year you paid them. It's a special rule for software engineers.
It was passed by Congress during the first Trump administration in order to offset the costs of other corporate tax rate cuts, due to budgeting rules.
Note that despite the small number, Kagi is profitable (or at least they were as of a year ago):
> We are also thrilled to report that we have achieved profitability. This significant milestone is a testament to our sustainable growth and fiscal responsibility. It demonstrates that our approach of offering a premium, ad-free search experience resonates with users who support a service aligning with their values. Becoming profitable allows us to reinvest in the business, further enhancing our offerings and ensuring that we can continue to provide a top-notch search experience.
As long as they're profitable I don't mind at all if they stay small. They're extremely useful to me as it is, and their small size means they aren't targeted for SEO nonsense, so their methods to cut through all that still actually work in my experience.
Not every business needs to become a unicorn. Some businesses are better at small scales serving a specific niche, and by their report Kagi seems to have found their niche.
https://blog.kagi.com/what-is-next-for-kagi
I wrote this after a bad week at a previous job trying to get an Android device to work with a CDC Ethernet adapter.
Since writing this, a couple people have let me know that there is a particular bit in the MAC address, that if flipped, will cause the kernel to assign an `ethX` name instead of `usbX` name. I haven't tried it myself or updated the post with that information because I moved on to a different job, and Android devices are no longer a large part of my life.
Of course, that only helps if you have a CDC device where you're in control of the MAC address (i.e. maybe another Linux device pretending to be a CDC adapter)
LLMs are divinatory instruments, our era's oracle, minus the incense and theatrics. If we were honest, we'd admit that "artificial intelligence" is just a modern gloss on a very old instinct: to consult a higher-order text generator and search for wisdom in the obscure.
They tick all the boxes: oblique meaning, a semiotic field, the illusion of hidden knowledge, and a ritual interface. The only reason we don't call it divination is that it's skinned in dark mode UX instead of stars and moons.
Barthes reminds us that all meaning is in the eye of the reader; words have no essence, only interpretation. When we forget that, we get nonsense like "the chatbot told him he was the messiah," as though language could be blamed for the projection.
What we're seeing isn't new, just unfamiliar. We used to read bones and cards. Now we read tokens. They look like language, so we treat them like arguments. But they're just as oracular, complex, probabilistic signals we transmute into insight.
We've unleashed a new form of divination on a culture that doesn't know it's practicing one. That's why everything feels uncanny. And it's only going to get stranger, until we learn to name the thing we're actually doing. Which is a shame, because once we name it, once we see it for what it is, it won't be half as fun.
I was lucky enough to be seen by MD Anderson doctors for my rare type of cancer. Within hours of the first painless infusion, the large tumors I could feel and see from the outside were completely gone. Hours, truly.
Now, the rest of the chemo wasn't as easy as the first. But the miracles of modern medicine, thanks to dedicated researchers and medical teams, truly blew my mind and that was a decade ago.
I was told I was one of 42 people ever that MD Anderson had seen with this cancer, yet they were able to tailor a treatment. I think Vox is right about the progress being amazing, and I hope it continues.
My father was on chemotherapy with fludarabine, a dna base analog. The way it functions is that it is used in DNA replication, but then doesn’t work, and the daughter cells die.
Typically, patients who get this drug experience a lot of adverse effects, including a highly suppressed immune system and risk of serious infections.
I researched whether there was a circadian rhythm in replication of either the cancer cells or the immune cells: lymphocyte and other progenitors, and found papers indicating that the cancer cells replicated continuously, but the progenitor cells replicated primarily during the day.
Based on this, we arranged for him to get the chemotherapy infusion in the evening, which took some doing, and the result was that his immune system was not suppressed in the subsequent rounds of chemo given using that schedule.
His doctor was very impressed, but said that since there was no clinical study, and it was inconvenient to do this, they would not be changing their protocol for other patients.
This was around 1995.
It might not be 100% clear from the writing but this benchmark is mainly intended as a joke - I built a talk around it because it's a great way to make the last six months of model releases a lot more entertaining.
I've been considering an expanded version of this where each model outputs ten images, then a vision model helps pick the "best" of those to represent that model in a further competition with other models.
(Then I would also expand the judging panel to three vision LLMs from different model families which vote on each round... partly because it will be interesting to track cases where the judges disagree.)
I'm not sure if it's worth me doing that though since the whole "benchmark" is pretty silly. I'm on the fence.
I was there, 3,000 years ago.
I remember fights over whether or not navigation in frames was bad practice. Not iframes, frames. Who here remembers frames?
I remember using HTTP 204 before AJAX to send messages to the server without reloading the page.
I remember building... image maps[1]... professionally in the early 2000. I remember spending multiple days drawing the borders of States on a map of the country in Dreamweaver so we could have a clickable map.
I remember Dreamweaver templates and people updating things wrong and losing their changes on a template update and no way to get it back because no one used version control.
I remember and handling where you clicked on an image in the backend.
I remember streaming updates to pages via motion jpeg. Still works in Chrome, less reliably in Firefox.
I remember the multiple steps we took towards a proper IE PNG fix just to get alpha blending... before we got the ActiveX one that worked somewhat reliably... Just for tastes to change and everything to become flat and us to not really need it anymore.
I remember building site navigations in Java, Flash, and Silverlight.
I remember spacer gifs and conditional comments and what a godsend Firebug was.
I don't know when I got old, it just happened one day.
1. https://developer.mozilla.org/en-US/docs/Web/HTML/Reference/...
It turns out that when elections are fought on the basis of identity (race, religion) etc corruption is actually considered a benefit! This is because the loyalists interpret this as "we" are winning and "they" are losing.
I witnessed this up close in India where parties openly exist to benefit certain constituencies based on caste, language, religion and so on.
It is horrifying to see this attitude take root in my adopted land.
From Walter Isaacson's _Steve Jobs_:
> One of Bill Atkinson’s amazing feats (which we are so accustomed to nowadays that we rarely marvel at it) was to allow the windows on a screen to overlap so that the “top” one clipped into the ones “below” it. Atkinson made it possible to move these windows around, just like shuffling papers on a desk, with those below becoming visible or hidden as you moved the top ones. Of course, on a computer screen there are no layers of pixels underneath the pixels that you see, so there are no windows actually lurking underneath the ones that appear to be on top. To create the illusion of overlapping windows requires complex coding that involves what are called “regions.” Atkinson pushed himself to make this trick work because he thought he had seen this capability during his visit to Xerox PARC. In fact the folks at PARC had never accomplished it, and they later told him they were amazed that he had done so. “I got a feeling for the empowering aspect of naïveté”, Atkinson said. “Because I didn’t know it couldn’t be done, I was enabled to do it.” He was working so hard that one morning, in a daze, he drove his Corvette into a parked truck and nearly killed himself. Jobs immediately drove to the hospital to see him. “We were pretty worried about you”, he said when Atkinson regained consciousness. Atkinson gave him a pained smile and replied, “Don’t worry, I still remember regions.”
It's wild that a president can say, "I don't like Elon anymore, so out of retaliation, I'm canceling all his government contracts," and ~40% of the country doesn't see that as corruption in any way, shape, or form.
Government contracts should not be based on whether or not the president likes the CEO, and the CEO says enough good things about the president.
If you can cancel contacts not based on merit, then it should extend you're likely willing to grant contracts not based on merit and based on nepotism instead.
This is literally the path that led the USSR to ruin. If anyone says anything you don't like, their funding is gone, even if it shoots the country in the foot. If people kiss your ass enough, they get contracts, even if it's clear they're just spending the money on hookers and coke and yachts and not delivering on promises, and it shoots the country in the head.
It's pretty bad.
It had a huge impact on my personally, I'm a small R&D shop and basically I have had to end all risky long-term research projects.
In addition to the research costs, I'd also have to pay taxes on the research costs mostly up-front. Significantly, if the project doesn't work out, I'm still out of pocket for the tax money. It's a penalty for taking a risk, and it kneecaps American innovators in a globally competitive technology race.
The rules are even worse than the article notes because it double-dings open source developers. See Section 6.4 of https://www.irs.gov/pub/irs-drop/n-23-63.pdf. The relevant bit is here:
> "However, even if the research provider does not bear financial risk under the terms of the contract with the research recipient, if the research provider has a right to use any resulting SRE product ... costs paid or incurred by the research provider that are incident to the SRE activities performed by the research provider under the contract are SRE expenditures of the research provider for which no deduction is allowed ..."
The rule as written means contractors who write Windows drivers could deduct their expenses (as they would have no residual rights to a closed-source work product), but contractors who write Linux drivers may not (as they would have some rights to open-source Linux drivers).
> Reading through these commits sparked an idea: what if we treated prompts as the actual source code? Imagine version control systems where you commit the prompts used to generate features rather than the resulting implementation.
Please god, no, never do this. For one thing, why would you not commit the generated source code when storage is essentially free? That seems insane for multiple reasons.
> When models inevitably improve, you could connect the latest version and regenerate the entire codebase with enhanced capability.
How would you know if the code was better or worse if it was never committed? How do you audit for security vulnerabilities or debug with no source code?
Transparency isn't the reason we use so much plastic. We like plastic because it is lightweight and not biodegradable. We like it because it lasts thousands of years. Because if it lasts thousands of years it will do a good job of storing your food products. Or it will stick around in various components without needing to worry about rain and such.
What we need to develop is something that doesn't degrade at all under most human living conditions, but does degrade rapidly if we expose it to some sort of not-common trigger, whether that is another chemical or temperature or pressure or whatever.
There are some misunderstandings in the comments that seem to stem from not having read the section, so I thought it was worth referencing the actual text [0]. It's quite short and easy to read.
The most important bits:
* Subsection (a) requires amortizing "Specified research or experimental expenditures" over 5 years (paragraph (2)) instead of deducting them (paragraph (1))
* Paragraph (c)(3) is a Special Rule that requires that all software development expenses be counted as a "research or experimental expenditure".
That's it. All software expenses must be treated as research and experimental expenses, and no research and experimental expense can be deducted instead of amortized. Ergo, all software expenses must be amortized over 5 years.
I strongly recommend reading the section before forming an opinion. It really is quite unambiguous and is unambiguously bad for anyone who builds software and especially for companies that aren't yet thoroughly established in their space (i.e. startups).
Also note that this makes Software a special case of R&D. It's the only form of R&D that Section 174 requires you to categorize as such and therefore amortize.
[0] https://www.law.cornell.edu/uscode/text/26/174
Bruce Dawson says: I like to call this Dawson’s first law of computing: O(n^2) is the sweet spot of badly scaling algorithms: fast enough to make it into production, but slow enough to make things fall down once it gets there.
https://bsky.app/profile/randomascii.bsky.social/post/3lk4c6...
This is the problem I had with all the content removal around Covid. It never ends with that one topic we may not be unhappy to see removed.
From another comment: "Looks like some L-whateverthefuck just got the task to go through YT's backlog and cut down on the mention/promotion of alternative video platforms/self-hosted video serving software."
This is exactly what YT did with Covid related content.
Here in the UK, Ofcom held their second day-long livestreamed seminar on their implementation of the Online Safety Act on Wednesday this week. This time it was about keeping children "safe", including with "effective age assurance".
Ofcom refused to give any specific guidance on how platforms should implement the regime they want to see. They said this is on the basis that if they give specific advice, it may restrict their ability to take enforcement action later.
So it's up to the platforms to interpret the extremely complex and vaguely defined requirements and impose a regime which Ofcom will find acceptable. It was clear from the Q&A that some pretty big platforms are really struggling with it.
The inevitable outcome is that platforms will err on the side of caution, bearing in mind the potential penalties.
Many will say, this is good, children should be protected. The second part of that is true. But the way this is being done won't protect children in my opinion. It will result in many more topic areas falling below the censorship threshold.
I like the way Jeff signed off the article, pointing out that whilst the video has been pulled for (allegedly) promoting copyright infringement, Youtube, via Gemini, is (allegedly) slurping the content of Jeff's videos for the purposes of training their AI models.
Seems ironic that their AI models are getting their detection of "Dangerous or Harmful Content" wrong. Maybe they just need to infringe more copyright in order to better detect copyright infringement?
Right, stealing training data from others is OK, having it stolen from you is not. What else is new?