2,296 karma · joined November 27, 2023
On my phone there are two swipe zones at the bottom. The one that's actually on the screen opens the Google Maps menu. But if I swipe from off screen up it works as I described. Maybe your phone doesn't have enough surface area below the screen to distinguish the two?
As a business, each additional human or company you add to your supply chain represents additional risk that you're taking on. You can go some ways towards mitigating those risks—one of the most common is to sign a contract with them rather than doing business ad hoc—but the risk doesn't go away entirely. Given that additional risk for each additional downstream supplier, it's generally safer to use code written by someone who's already on your payroll than it is to use code written by someone you've never met and have no way of vetting.
> like when I upgraded to 15 my home button no longer exists when the screen is locked, which effectively makes google maps navigation stuck on the screen unless you stop navigation by pressing back repeatedly
Smaller things like this, sure.
FYI, swiping up from the bottom will get into the lock screen and then once unlocked into the home screen.
This is really bad reasoning for a few reasons:
1) We've gotten much better at training LLMs since 2022. The negative impacts of AI slop in the training data certainly don't outweigh the benefits of orders of magnitude more parameters and better training techniques, but that doesn't mean they have no negative impact.
2) "Outperform" is a very loose term and we still have no real good answer for measuring it meaningfully. We can all tell that Gemini 2.5 outperforms GPT-4o. What's trickier is distinguishing between Gemini 2.5 and Claude 4. The expected effect size of slop at this stage would be on that smaller scale of differences between same-gen models.
Given that we're looking for a small enough effect size that we know we're going to have a hard time proving anything with data, I think it's reasonable to operate from first principles in this case. First principles say very clearly that avoiding training on AI-generated content is a good idea.
I looked through the highlights linked here and the full "What's New" page [0] and am pleasantly surprised to see a few new features but no major overhauls of existing ones.
Out of 5 comments:
* This one is saying that Europe's biggest AI company isn't keeping up with China.
* Two comments are talking about Europe needing to become more independent from American tech companies.
* A fourth is talking about ElevenLabs's Portuguese support and noting that one of the Portuguese voices has a Spanish accent.
* The fifth is unrelated to geography or language.
I'm not sure where in this you get that they have a US-centric perspective. The only comments they have that mention the US at all are taking the very distinctly 2025-European tack of advocating for increased independence from America. Coupled with this one contrasting with Chinese tech that presents a pretty unified picture of someone who has a lot of interest in Europe having its own strong tech independent from the rest of the world.
The ElevenLabs comment is the most telling for me for pinpointing the geography: in order to know what Portuguese spoken with a Spanish accent sounds like you really have to have spent a significant amount of time in Iberia and most likely speak Portuguese as a native language.
EDIT: The comment below got flagged to death, so I'm going to link my evidence up here for visibility: https://news.ycombinator.com/item?id=44238841
Not only does the top-level comment make no mention of the US, but their comment history is strongly suggestive that they themselves are European, not US-based.
Using salaries as a proxy for value of the asset encourages only the safest shovelware bets, discouraging risk taking lest your asset be taxed at substantially higher than it's worth.
Avoiding that risk-adverse dynamic is why Section 174 was written the way it was since the 50s to encourage R&D, and it's paid off in spades.
The same is true for all true R&D, which is why historically the government has tried to provide protections for R&D work to incentivize people to not just churn out the safe bet over and over again. Patents fall into this category, but software patents are (rightly) hard to come by. Through 2022, the risk of software development was offset by the ability to expense the costs and avoid a tax bill, and this was good policy if your aim is to encourage innovation.
The capital asset theory could still work if there were some way to appraise the value of the actual asset you created. But absent such a way, this thinking is deeply flawed for all but the most shovelware of jobs.
The Section 174 changes altered the accounting method that software development companies must use for calculating their profit for tax purposes. Starting in 2022 software dev must be treated not as an expense but as an investment in an asset, such that you're now required to amortize the expense over 5 years instead of deducting it from your revenue the year you spent the money.
The gigantic problem with this change is that without the ability to expense software development expenses as expenses, a new software startup can very easily be considered profitable in their first year because only 10% of their software development-related expenses get to be counted as expenses.
And note that, contrary to what you say, most white-collar work is not treated this way, and software is further singled out from other R&D-type work in that it is the only type of work that is explicitly called out in the section as being required to be marked as R&D. So we're not asking for software to be treated specially, we're asking why it suddenly changed in 2022 to be treated specially in an extremely negative way.
> 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.
The rest I don't know enough to comment on, so I'll assume you're correct.
Karpathy used it as an example in a recent interview: https://www.msn.com/en-in/health/other/ai-expert-asks-grok-3...
ATC is a safety-critical function that has what amounts to a 100% uptime requirement. Whatever system they're running currently either works or has known flaws that they know how to work around, and air traffic controllers have been trained on these systems for more than a generation now. Upgrading merely for the sake of being up to date would have been foolish no matter how much funding Congress would have given them.
If they're saying that they need the upgrade now, I'll trust them on that, but it was the right call to make it last.
I love this example! It perfectly illustrates a case where the government intentionally subsidizes a liability that no sane company would take on without government funding. Well said.
So we're agreed that the government should incentivize R&D with a favorable tax code that makes it not completely insane to take on the risk of doing something new.
Further, software is the only type of R&D explicitly called out as required to count as R&D. Which means it should be taken as a given that most other industries are finding ways to count their R&D as anything else, while we've been intentionally given the short straw for some reason by having our specific field be the only one identified by name so as to leave no wiggle room. I'd say that definitely counts as being a special case. The section is even labeled "Special Rules".
Tell me you're not an experienced software engineer without telling me you're not an experienced software engineer.
Code is a liability, not an asset.
Now, after this change, software is a special group singled out, a deviation in the tax code specially carved out for us to make our field specifically the exceptional one with no wiggle room. No other type of development is named in Section 174 to explicitly require companies to amortize.
So who is that subsidizing now?
This blocks most existing tracking methods. The only thing you're not protected from is first-party tracking by the site you're actually visiting, which is impossible to fully protect against.
As has been said repeatedly in this thread, this change is purely a boon for existing big tech companies that now have even less to worry about from startups. It takes a startup 5 years before they'll be playing on an even field with big tech.
> if you’re operating a profitable company
You keep saying this across this thread, and keep ignoring that Section 174 has now redefined "profitable" for tax purposes to include companies who:
* Are in year 1 with no history of expenses to draw on.
* Have spent <900% of their year 1 revenue on software development expenses.
i.e., a startup that earns $1mil and spent $8mil in software dev expenses is only able to deduct 10% * 8mil = $800k of expenses, which means that as far as the government is concerned they made a profit of $200k and owe taxes on that on top of their already-net-loss of $7mil.
You can keep ignoring this fact, but ignoring it doesn't help your case. If you want to argue that this is fine and dandy you need to explain why the above math doesn't prevent new companies from competing on fair terms with big tech.
Software firms across US facing tax bills that threaten survival (924 points, 981 comments) April 18, 2023 https://news.ycombinator.com/item?id=35614313
Ask HN: How are you handling Section 174 changes for bootstrapped companies? (298 points, 187 comments) Feb 2, 2023 https://news.ycombinator.com/item?id=34627712
Why the big tech firms that suddenly laid off a bunch of people the instant they started looking at their 2022 tax bill didn't tell everyone explicitly that that's what was happening I can't say, but it's not like this has been happening in secret.
Obviously interest rates also play a role, and probably a larger one. But this is objectively a very very bad contributing factor, far worse than the impact of coding LLMs.
OP is saying that the old system is clearly better because when they tried the new system the old system fought back and killed people, so they shouldn't have tried to replace the old system in the first place. It's democracy's fault that Somali warlords had to be brutal to keep democracy from working. Everyone would have been better off if they had just continued to put up with the warlords' old way of working.
This logic isn't comparable to the logic of Western democracies, it's comparable to the logic of criminal mobs everywhere. Play along and no one has to get hurt.
> Which depending on one's perspective on Somalia's current state could make it deeply problematic
It's pretty hard to look at Somalia's current state and say that their current social structures are "known to work" by essentially any metric.
Nope, clans are definitionally not states for several reasons. A state has definite territory, whereas clan-type structures have tended to overlap with each other geographically because they're usually embedded in some larger society. A state by definition has centralized authority whereas clans may not.
> or whatever label you want to put on the organization that starts with small family tribes and scales to multi-continent empires
There is no such label because these organizations are not just shades on the same theme at different sizes, they're fundamentally different in character.
> that kind of society organization was more or less everywhere in the history of every human society
True, but there are wide variances in how long in the past that form of organization was dominant.
You won't see Egypt reverting to decentralized non-state clan-oriented governance any time soon because they've been ruled by one state or another for 5000 years.
I'm not saying it's the actual story, but the timeline does track.
[0] Page 60, Sec 1306(e) sets the date: https://www.congress.gov/115/bills/hr1/BILLS-115hr1enr.pdf
Each and every startup will have a year 0 where they're spending more than they earn, and under the new Section 174 they will only get to deduct 10% of their employee's salaries that year. In year two they get to deduct 20% of year 1's salaries and 10% of year 2's salaries, which is still 30% of what the established players will be able to deduct. By year 4, if they make it that far (which most startups don't) they'll finally be at 90% of a full deduction.
Add to that the fact that startups also by definition have a much higher rate of growth than established companies and you'll find that a startup almost definitionally will be paying substantially more in taxes as long as it remains a startup, because they only get to deduct an average of the last 5 years of expenses from this year's revenue in order to calculate this year's profit. That's fine when your last five years are more or less similar to this one, but it's terrible when you've been growing.
The net effect of this change can only be to disincentive startups and cement big, slow established players.