Acko.queue(function () { renderMathInElement(document.querySelector('article'), {delimiters: [ {left: "$$", right: "$$", display: true}, {left: "$", right: "$", display: false}, ]}); }); Improving on a timeless recipe The bezier curve is a staple of CAD and computer graphics. Like Bic pens, the design is decades old and they're everywhere. You'll often find them as the default or only choice in various illustration and animation tools. Conceived by Paul de Casteljau in 1959, and refined by Pierre Bézier in the 1960s at Renault, the enduring appeal of the bezier curve lies in its simplicity. While more sophisticated curves have been invented, and new ones continue to be proposed, these are limited to specific domains, like high-precision CAD or road design. In general use, the bezier stubbornly refuses to be dethroned, despite its shortcomings. Hence bezier curves are a piece of legacy tech we appear to be stuck with. As a software engineer, my question then is: can we make beziers better without invalidating all the tech built on and with them? The answer is yes. Source Lerp-a-derp Drawing a bezier curve is a surprisingly simple and linear process: Tip: All the diagrams in this post are fully interactive. Given a series of control points, we connect them with lines. We then run along those lines simultaneously, to produce new points, which can be connected again. This process is repeated until we are left with a single point, which lies on the curve. This construction makes beziers far more regular than they might first appear. The linear interpolations (aka lerps) can be summarized into a single compact formula, e.g. for 4 control points $ \left(A, B, C, D\right) $: $$ \gamma \left( t \right) = A \cdot \left(1 - t\right)^3 + B \cdot 3 \left(1 - t\right)^2 t + C \cdot 3 \left(1 - t\right) t^2 + D \cdot t^3 $$ The rule is simple: descending powers of $\left(1 - t\right)$, ascending powers of $t$, with coefficients taken from the n'th row of Pascal's triangle: $$ \begin{array}{ccccccccccccc} &&&&&& 1 \\ &&&&& 1 && 1 \\ &&&& 1 && 2 && 1 \\ &&& 1 && 3 && 3 && 1 \\ && 1 && 4 && 6 && 4 && 1 \\ &1 && 5 && 10 && 10 && 5 && 1 \\ ... &&&&&& ... &&&&&& ... \\ \end{array} $$ For curves in 2D and 3D, the formula is applied to the individual X, Y or Z coordinates. While beziers can be constructed for any number of control points, the common practice is to only use cubic beziers with 4 control points. This is because the curve is only guaranteed to cross through the first and last control point, which makes higher degree beziers more difficult to shape. Larger curves are instead constructed by joining together multiple cubic bezier segments, with the tangents lined up to create a segmented curve that appears smooth: This is the cubic bezier spline, as commonly understood. The precision "pen tool" in most drawing apps then consists of drawing and editing the control points, rather than drawing curves directly. A Lie Told Everywhere Pen tools typically have a few different modes for the control points: Intuitively these represent various degrees of smoothness. Symmetric tangents are offered as the smoothest option, with some qualities of smoothness being lost as you relax the constraints. In reality this is completely wrong, and this is easy to demonstrate. Bezier curves can be split exactly, by reading off the new control points from the interpolation diagram: The left and right segments are 100% identical to the original curve, and always join up perfectly at the seam. Yet the tangents in the middle will be asymmetric except for one split near the middle. This can be confirmed using a curvature comb which represents the (inverse) radius of curvature at every point: The curvature comb remains continuous, with no jumps. This means that whether or not adjacent tangents are of equal length, i.e. symmetric, is completely irrelevant. Attempting to draw smooth and intuitive bezier curves this way is a fool's errand. A simple way to improve this is to treat the tangents as relative rather than absolute. e.g. We can make them proportional to the distance between the start and end of each segment: This spline is easier to edit, because as you move each curve point around, the adjacent segments naturally flex to get out of the way. There are far fewer cusps created this way. Editing the tangents remains the same. However if we plot curvature again, we can tell this isn't a great solution: Scaling tangents proportionally doesn't guarantee that curvature is preserved, nor does curvature remain continuous from one segment to the next. This also shows that offering users a curvature comb visualization as a "helpful tool" is really quite mean: adjusting the curvature on one end will also affect the other side, requiring repeated adjustments back and forth until it's close enough. Handle Carefully A much more effective strategy is to work with curvature directly. While this is a difficult problem in general, it turns out there are some surprising relationships here, which we can observe directly: Consider the curvature at the start of segment $A-B-C-D$. This is affected only by the positions of $A$, $B$ and $C$. Point $D$ can be moved freely if it's detached from $C$. Furthermore, because the tangent $A-B$ is horizontal, only the vertical position of $C$ matters. This is a result of the underlying linear interpolations, which end up cancelling out a lot of terms in the formulas. The curvature at $A$ is therefor only affected by the length of $A-B$, and the perpendicular distance from $C$ to $A-B$. The same applies to $D$ on the other side with $C-D$ and $B$. This isn't very useful by itself though. When we move a curve point ($A$ or $D$), typically the adjacent control point ($B$ or $C$) is moved as well to preserve the tangent. And when we turn a symmetric or asymmetric tangent, the adjacent tangent in the next segment is turned by the same angle, altering the curvature on that side. Still, this shows that preserving curvature is not by itself a crazy idea and can be done simply by scaling the tangents appropriately. The rules would be simple: When we move a curve point, we have to preserve the start and end curvature in the adjacent segments When we move a tangent, we have to preserve the curvature on the opposite end of the current segment, as well as on both sides of the adjacent segment This ought to produce an editing experience where the curve actually respects your intent. Except not quite: Moving curve points works great, but when turning a tangent, the length of that tangent is itself a poor representation of the user's intent. Small changes can cause huge shifts in curvature, which can cause the curve to jump around and explode unexpectedly as it tries to find a matching solution. Hence it's better to work with curvature handles instead, where the length corresponds directly to curvature: Unlike the curvature comb, the length of the handles is the non-inverted radius of curvature, which is the more natural choice. These handles are very stable and can be converted just-in-time to classic bezier control points, without needing to round-trip back and forth between the two representations. Helpfully, this also avoids numerical drift. Goldilocks To actually pull this off, we need to solve for the lengths $l_0$ and $l_1$ of the tangents, given the desired curvatures $k_0$ and $k_1$, the start/end points $A$ and $D$, and the unit-length tangents at the start/end. Given a curve $\gamma\left(t\right)$, we can express the unit tangent vector $\mathbf{T}\left(t\right)$ as the normalized derivative: $$ \mathbf{T}\left(t\right) = \frac{\displaystyle \gamma'\left(t\right)}{\displaystyle |γ'\left(t\right)|} $$ This can be used to find the curvature vector $\mathbf{K}\left(t\right)$ via two vector cross products using the first and second derivative: $$ \mathbf{K}\left(t\right) = \mathbf{T}\left(t\right) \times \frac{\gamma''\left(t\right)}{|\gamma'\left(t\right)|^2} \times \mathbf{T}\left(t\right) $$ The cross products ensure that $\mathbf{K}\left(t\right)$ is perpendicular to $\mathbf{T}\left(t\right)$, i.e. they extract the normal vector component of the middle term. Now we can solve for $\mathbf{K}\left(0\right) = k_0$ and $\mathbf{K}\left(1\right) = k_1$. After working through the math, we end up with a quadratic system of equations in $l_0$ and $l_1$: $$ \begin{cases} a_0 \cdot l_0^2 + b \cdot l_1 + c_0 = 0 \\[0.3em] a_1 \cdot l_1^2 + b \cdot l_0 + c_1 = 0 \end{cases} $$ Where: $$ \begin{array}{lcl} a_0 & = & ±|k_0| \\[0.3em] a_1 & = & ±|k_1| \\[0.3em] b & = & 2 \cdot |\mathbf{T}(0) \times \mathbf{T}(1)| \\[0.3em] c_0 & = & 6 \cdot |\mathbf{T}(0) \times (D - A)| \\[0.3em] c_1 & = & 6 \cdot |\mathbf{T}(1) \times (A - D)| \\[0.3em] \end{array} $$ This captures a few things: The sign of the curvature $k_i$ defines whether the curve turns clockwise or counterclockwise. When moving curve points around, the curve may be forced to flip, hence the $±$ is necessary, and both signs can change independently. The coupling between $l_0$ and $l_1$ is influenced only by $b$. If $b = 0$, then the two equations are independent and the problem is trivial to solve. This corresponds to the situation where the two tangents are parallel. The constant term $c_i$ is influenced only by the perpendicular distance $D - A$ (or $A - D$) to the tangent $\mathbf{T_i}$. This matches the earlier finding that only perpendicular distance affects curvature. Hence, the problem is reduced to finding the intersection of two double-parabolas, one horizontal and one vertical: The two sides of each double-parabola represent bending clockwise or counterclockwise. We have to solve 4 times, one for each sign combination $\left(+,+\right)$, $\left(+,-\right)$, $\left(-,+\right)$, $\left(-,-\right)$. To solve this system, we can rewrite either of the equations to isolate $l_1$ or $l_0$: $$ l_1 = \frac{\displaystyle -a_0 \cdot l_0^2 - c_0}{\displaystyle b} \\[0.5em] l_0 = \frac{\displaystyle -a_1 \cdot l_1^2 - c_1}{\displaystyle b} $$ Pick one, and square it to plug it into the other original equation as the $l_1^2$ or $l_0^2$ term. This produces a quartic equation in the other variable (resp. $l_0$ or $l_1$), which can be solved directly using the quartic formula. We can then compute the other $l_i$, either: $$ l_1 = \sqrt{\frac{\displaystyle -b \cdot l_0 - c_1}{\displaystyle a_1}} \\[0.5em] l_0 = \sqrt{\frac{\displaystyle -b \cdot l_1 - c_0}{\displaystyle a_0}} $$ Not all solutions of the quartic will be actual solutions to the system however, because of the squaring. So we have to double check that the original equations hold before accepting a solution for $\left(l_0, l_1\right)$. We also want to reject solutions where $l_0 < 0$ or $l_1 < 0$, because these represent situations where the tangents have been flipped by 180º. In most cases, there will only be one valid solution, and this works pretty well: For corner-type points, one or both $k_i$'s are infinite. This is straightforward to solve as the matching $l_i$ is zero and the other can be computed directly. Grouper and Snapper Unfortunately if there are multiple solutions, these each represent a different curve: Having the curve jump around as you edit it would be a very bad editing experience, but there is no obvious way to pick the right one. One strategy would be to remember the previous $l_0$ and $l_1$ as you edit, and pick the solution that is closest to the previous curve. However this would introduce path-dependence into editing, where the order and direction that you moved points around in affects which curve you end up with. For a drawing tool, this is quite frustrating and undesirable. Different solutions to the equation can also appear and vanish suddenly, so this wouldn't eliminate popping either, merely reduce it. Ideally the resulting curvatures should always change smoothly themselves. This can be accomplished by designing an appropriate weighting heuristic for the solutions, and calculating their weighted average. 1) The forbidden regions: When a solution approaches either $l_0 These solutions are unstable and should be avoided. To detect this, I calculate the gradient vector of each equation. $$ \mathbf{g_0} = \begin{bmatrix}2 \cdot a_0 \cdot l_0 & b\end{bmatrix} \\[0.3em] \mathbf{g_1} = \begin{bmatrix}b & 2 \cdot a_1 \cdot l_1\end{bmatrix} $$ The angle between them can be found via their dot product, after dividing by their respective lengths: $$ \cos \theta = \frac{\displaystyle \mathbf{g_0} \cdot \mathbf{g_1}}{\displaystyle |\mathbf{g_0}||\mathbf{g_1}|} $$ A value close to $-1$ means two opposite-bending parabolas, while a value close to $1$ means two aligned parabolas. The latter is not as bad, as this implies many almost-equivalent solutions. 3) When the radius of curvature is very big compared to the bezier itself, one of the possible solutions is for the curve to cross through itself, forming a loop. This should be discouraged: We can divide by the length of $|\mathbf{l}|$ i.e. $\sqrt{l_0^2 + l_1^2}$, preferring solutions that have shorter tangents. Putting it all together, the weighting heuristic I chose is: $$ w_{heuristic}\left(\mathbf{l}\right) = \left(\frac{\displaystyle l_0 \cdot l_1 \cdot \arccos\left(-\cos \theta\right)}{\displaystyle |\mathbf{l}|^2}\right)^2 $$ By applying $\arccos \left(-\cos \theta\right)$, we remap the dot-product range $(-1..1)$ to $(0..π)$ which effectively neutralizes the unstable solutions Large $|\mathbf{l}|$'s represent undesirable solutions, so should be strongly penalized and only chosen if they are the only option When the entire heuristic approaches $0$, the influence should vanish smoothly, hence the overall squaring Using this as the weight, the resulting bezier curves always change smoothly, even when the start and end are moved around and through each other. But when multiple solutions are averaged together, the resulting $(l_0, l_1)$ is not necessarily a solution to the original system. This means the resulting curvatures are not always exact, even when there is an exact solution nearby. To mitigate this, we can add an additional resampling step, where we re-weigh the solutions according to the inverse distance^4 to the weighted average: $$ w_{sample}\left(\mathbf{l}\right) = \frac{1}{|\mathbf{l} - \mathbf{l}_{average}|^4} $$ This causes the solution to snap smoothly to exact solutions when available. Migratory Curves What's especially nice about this approach is that it can largely be slotted into existing systems with little modification required. The UI of adjusting and editing bezier handles can be kept, including the notion of the 4 point types. Just the meaning of the lengths of the tangents changes. Converting an asymmetric point to a symmetric point will actually equalize the curvature of the left and right side, which is what the user actually wants. The internal representation can mostly remain the same type, i.e. a series of points, just those points represent curvature tangents, not bezier tangents. Though it's better to explicitly store the unit length tangents and curvatures directly, which is more numerically stable. For the demos here I made pairwise conversion functions between curvature handles ↔︎ points/tangents/curvatures ↔︎ bezier handles. Splitting a curvature bezier is a lot simpler than splitting a classic bezier, because the start and end handles don't change at all, and the newly created handles are symmetric: There is one remaining flaw though. When splitting a curve with an S-bend, the splits near the bend are not exact: The heuristic picks the wrong curve on one or both sides. This can be fixed by remembering the specific orientation of the curvature, and only flipping the curvature sign when necessary. Though this does introduce some minimal path-dependence into editing. So really you would need to expose this in the UI as an extra orientation toggle for ambiguous control points. It would also mean that the internal representation has to be in tangent + signed curvature form to be fully stable, not just curvature handles. Near the S-bend, the radius of curvature—and thus the handle size—effectively becomes infinite, because $|k|$ becomes $0$. The bezier control points lie on a straight line, as perpendicular distance to the tangent is zero. So even when handled exactly, this can still create impractically large curvature handles. Perhaps what's needed is a 5th "Inflection" point type, for points with exactly zero $|k|$, though such points tend to be very unstable, and sometimes no solution exists: To migrate a classic bezier to a curvature bezier, you can measure the unit tangents and curvature at the start and end. In the vast majority of cases this will result in the exact same curve. If it doesn't, then you must split the original bezier to avoid the ambiguity. For complex operations on bezier curves, e.g. boolean union or intersection, these still have to be done in bezier form. This implies some amount of round-tripping between the two representations, which would introduce numerical drift. However, only the newly created points need to be converted back, because the curvature and tangents elsewhere remain the same. * * * This post started from the observation that "symmetric" bezier tangents aren't actually symmetric as you'd expect. Hopefully you found the resulting exploration enlightening. The practice of showing curvature combs to diagnose problems with splines is really backwards. If we can show a curvature comb, we can also compute an appropriate spline to match the desired curvature. So beziers as they commonly exist are quite inadequate. We can retrofit a better system to edit them without breaking the expectations of a smooth editing experience, or introducing a new class of curves that behaves very differently. The code for the diagrams and implementation can be found on GitLab.

On Evitability in Use of AI If the hype is to be believed, software development as we know it is over. Strangely though, despite now years of LLM-powered tooling, the results look, feel and function mostly the same as they ever did: barely. It's undeniable there's a metric gigaton of hype surrounding the technology. It drives the enormous amounts of money and infrastructure being poured into it, which in turn demands more hype to justify the investment. The history of hyperbole is already evident, as new models continue to be trained to reach promises which now-retired models were already supposed to deliver. So allow me to drop a line that would shock a weathered San Franciscan more than open defecation on Market Street: it's perfectly okay not to use AI. It doesn't make you a troglodyte. It won't leave you choking behind in the dust as self-fashioned techno-wizards bring their agents to bear. In fact, it seems far less stressful and far more satisfying than the alternative. Source M.C. Escher, "Reptiles" – 1943 Craft vs Kraft In all the talk of what it is that LLMs do and don't, there are a lot of ways to frame what is happening. The positive spin includes helpfulness, cleverness, creativity and productivity. The negative spin points at lazyness, disposability, theft, and decay of knowledge. But there's one word that's remarkably absent in the discourse. That word is forgery. If someone produces a painting in the style of Van Gogh, and passes it off as being made by Van Gogh, by putting his signature on it, that painting is a forgery. If someone produces a legal document by mimicking the format, impersonating the parties, and faking their agreement, that document is a forgery. If someone produces a study by inventing or altering data, making up citations, and cherry-picking results to fit a particular conclusion, that study is a forgery. Whether something is a forgery is innate in the object and the methods used to produce it. It doesn't matter if nobody else ever sees the forged painting, or if it only hangs in a private home. It's a forgery because it's not authentic. P. Picasso, "Buste De Femme" – 1942 In this perspective, LLMs do something very specific: they allow individuals to make forgeries of their own potential output, or that of someone else, faster than they could make it themselves. The act of forgery is the act of imitation. This by itself is strictly-speaking legal, as a form of fiction or self-expression. It's only when one attempts to use a forged item as a substitute for the authentic thing that it creates problems. How this plays out in practice depends on the situation, and mainly depends on what authenticity would signify. That is, nobody will be arrested for "forging" a letter from Santa Claus, but also, no jurisdiction would allow you to have extremely convincing "imitation money" purely as a collector's item. This sort of protectionism is also seen in e.g. controlled-appelation foods like artisanal cheese or cured ham. These require not just traditional manufacturing methods and high-quality ingredients from farm to table, but also a specific geographic origin. There's a good reason for this. © Producing French "Brie de Meaux" abroad isn't allowed, because it would open the floodgates to inevitable cheaper imitations. This would degrade the brand of the authentic product, and threaten the rare local expertise necessary to produce it, passed down from generation to generation. The judgement of an individual end-consumer simply isn't sufficient here to ensure proper market function. The range of products that you can get in the store, between which you can choose, has already been pre-decided by factors out of your control. The quality of the artisanal cheese is a stand-in for an entire supply chain, often run on modern methods, which cannot simply be transplanted elsewhere without enormous investments in human capital, infrastructure and agriculture. This isn't mere romanticism. Every society has to draw a line somewhere on the spectrum between "traditional artisanal cheese" and "fake eggs made from industrial chemicals", if they don't want people to die from malnutrition or poisoning. But it's the ones that understand and maintain the value of foodcraft that don't end up with 70%+ obesity rates. © Distrust and Verify The parallels to LLM-driven software coding are not difficult to find. The craft of writing software is being threatened by a literal flood of cheap imitations. Open source software maintainers have been one of the first to feel the downsides. They already had a ton of difficulty finding motivated contributors and bringing them up to speed on the project's goals and engineering mindset. The last thing they needed was to receive slop-coded pull requests from contributors merely looking to cheat their way into having a credible GitHub resumé. Being on the receiving end of this is both demeaning and absurd, as the only thing the vibe-coder can do with the feedback you give them is paste it back into the tool that produced the errors in the first place. As a result, projects have closed down public contributions and dropped their bug bounties. Others just mock the posers and hope they go away. What this certainly isn't is helpful, clever, creative or productive. In day to day coding, working alongside vibe-coding co-workers has similar effects. While new employees might seem to get up to speed much quicker, in reality they're merely offloading those arduous first weeks to a bot, hoping no-one else notices. In the process, they'll inject run-of-the-mill mediocrity all over the place, when what you were really hoping for was their specific perspective. Anno 2026, if a new employee produces an extremely detailed PR with lots of explanation and comments, doubt every word. Experienced veterans who turn to AI are said to supposedly fare better, producing 10x or even 100x the lines of code from before. When I hear this, I wonder what sort of senior software engineer still doesn't understand that every line of code they run and depend on is a liability. One of the most remarkable things I've heard someone say was that AI coding is a great application of the technology because everything an agent needs to know is explained in the codebase. This is catastrophically wrong and absurd, because if it were true, there would be no actual coding work to do. It's also a huge tell. The salient difference here is whether an engineer has mostly spent their career solving problems created by other software, or solving problems people already had before there was any software at all. Only the latter will teach you to think about the constraints a problem actually has, and the needs of the users who solve it, which are always far messier than a novice would think. When software is seen as an end in itself, you end up with a massively over-engineered infrastructure cloud, when it could instead be running on a $10/month VPS, with plenty of money left for both backups and beer. Tools for Tools Engineers who know their craft can still smell the slop from miles away when reviewing it, despite the "advances" made. It comes in the form of overly repetitive code, unnecessary complexity, and a reluctance to really refactor anything at all, even when it's clearly stale and overdue. I've also observed several times now that even being senior, with years of familiarity, will not save them from vibe-coding some highly embarrassing goofs, and passing them on like an unpleasant fart. Trying to imagine what thought-process produced the odd work in question will quickly lead to the answer: none at all. It's not a co-pilot, it's just on auto-pilot. The same applies to vibe-coders themselves, and the reactions are largely predictable. The notion is being felt that slop code is bad code, full of bugs, with e.g. Microsoft's Co-pilot Discord recently banning the insult "Microslop". The user backlash was then framed as "spam" and even outright "harmful", demonstrating that the promise is often worth more than the actual result, and also, that the universe still has a sense of humor. Source M.C. Escher, "Print Gallery" – 1956 Less encouraging is that you'll see these tools referred to as "addicting" or even "the best friend you can have". While nerds being utterly drawn to computers is as old as the PC revolution itself, there doesn't seem to be an associated cambrian explosion of creativity and accomplishment to go with it. I can understand why outsiders would be impressed by it, what I don't understand is how so many insiders didn't stop and think about it. © What gets built with AI is really all the glue that's become necessary since said PC revolution, as software applications have gotten more closed, more distributed and more corporate. The options here for end-users are terrible. HTTP APIs don't make things open if every endpoint requires a barely documented JSON blob whose schema changes overnight. Slinging raw database dumps is also not viable, and is only used for disaster recovery. Software has largely rusted shut. Consider that many companies still primarily running on Excel. What's the Excel of JSON? There is none. So yeah, of course users think they need a machine to translate their intent into code so they can run it. Even then, what's the Jupyter notebooks of JSON? There's jq of course, but keep in mind that originally it was SQL that was framed as the solution that was going to free businesses and their workers from having to rely on dedicated tools. Look how that worked out... the more things change, the more they stay the same. Is there a standard CRDT-like protocol for syncing editable graphs yet? Surprisingly, we haven't seen a return to native apps either. It turns out vibe-coding an Electron app is still preferable to vibe-coding on multiple platforms and delivering a tailored experience for each. So where is this famed 100x? If even Apple can't maintain proper form and iconography in their latest OS anymore, what chance does an AI trained on web-slop have? It says a lot about our industry, it just doesn't say much about engineering at all. Source J.M.W. Turner, "The Shipwreck" – 1805 And a Bottle of Rum Software engineers have largely jumped in without a life-jacket, but not every industry has been as eager. The frame of inevitability is just that, a frame, and one which you should question. Video games stand out as one market where consumers have pushed back effectively. Numerous titles have already apologized for unlabeled AI content and removed it. Platforms like Steam have clearly signposted policies about it, and tools exist to filter it out. That said, Steam's policy has been recently updated to exclude dev tools used for "efficiency gains", but which are not used to generate content presented to players. This isn't all that surprising, for two reasons. The first is that video games are a pure direct-to-consumer market with digital delivery. Gamers really do have all the choices on tap. When they don't like a game or its pricing model, it's the result of choices made by those specific producers. Other titles exist without those flaws, and those get promoted and bought instead. The taste-makers are gamers themselves, who demand transparency. But the second is that most video games are artistic, and bought for their specific artistic appeal. In art, copy-catting is frowned upon, as it devalues the original and steals the credit. Artists are rationally very sensitive to this, as part of the appeal of art is a creator's unique vision. The art is supposed to function as a personal proof-of-work, whose integrity must be preserved. The proper form of imitation is instead an homage, which respects and evolves an idea at the same time. This stands in stark contrast to code, which generally doesn't suffer from re-use at all, or may even benefit from it, if it's infrastructure. It also explains why open source projects are so particularly ill-suited to attracting talented, artistic creatives. The ethos of zero-cost sharing means any artistic design would be instantly pilfered and repurposed without its original context. Classic procedural generation is noteworthy here as a precedent, which gamers were already familiar with, because by and large it has failed to deliver. The promise of exponential content from a limited source quickly turns sour, as the main thing a procedural generator does is make the variety in its own outputs worthless. © No Man's Sky - 2016 version So it's no wonder artists would denounce generative AI as mass-plagiarism when it showed up. It's also no wonder that a bunch of tech entrepreneurs and data janitors wouldn't understand this at all, and would in fact embrace the plagiarism wholesale, training their models on every pirated shadow library they can get. Or indeed, every code repository out there. If the output of this is generic, gross and suspicious, there's a very obvious reason for it. The different training samples in the source material are themselves just slop for the machine. Whatever makes the weights go brrr during training. This just so happens to create the plausible deniability that makes it impossible to say what's a citation, what's a hallucination, and what, if anything, could be considered novel or creative. This is what keeps those shadow libraries illegal, but ChatGPT "legal". Labeling AI content as AI generated, or watermarking it, is thus largely an exercise in ass-covering, and not in any way responsible disclosure. It's also what provides the fig leaf that allows many a developer to knock-off for early lunch and early dinner every day, while keeping the meter running, without ever questioning whether the intellectual property clauses in their contract still mean anything at all. This leaves the engineers in question in an awkward spot however. In order for vibe-coding to be acceptable and justifiable, they have to consider their own output disposable, highly uncreative, and not worthy of credit. * * * If you ask me, no court should have ever rendered a judgement on whether AI output as a category is legal or copyrightable, because none of it is sourced. The judgement simply cannot be made, and AI output should be treated like a forgery unless and until proven otherwise. The solution to the LLM conundrum is then as obvious as it is elusive: the only way to separate the gold from the slop is for LLMs to perform correct source attribution along with inference. This would not just help with the artistic side of things. It would also reveal how much vibe code is merely just copy/pasted from an existing codebase, while conveniently omitting the original author, license and link. With today's models, real attribution is a technical impossibility. The fact that an LLM can even mention and cite sources at all is an emergent property of the data that's been ingested, and the prompt being completed. It can only do so when appropriate according to the current position in the text. There's no reason to think that this is generalizable, rather, it is far more likely that LLMs are merely good at citing things that are frequently and correctly cited. It's citation role-play. The implications of sourcing-as-a-requirement are vast. What does backpropagation even look like if the weights have to be attributable, and the forward pass auditable? You won't be able to fit that in an int4, that's for sure. Nevertheless, I think this would be quite revealing, as this is what "AI detection tools" are really trying to solve for backwards. It's crazy that the next big thing after the World Wide Web, and the Google-scale search engine to make use of it, was a technology that cannot tell you where the information comes from, by design. It's... sloppy. To stop the machines from lying, they have to cite their sources properly. And spoiler, so do the AI companies.

Modern SSAO in a modern run-time Use.GPU 0.14 is out, so here's an update on my declarative/reactive rendering efforts. The highlights in this release are: dramatic inspector viewing upgrades a modern ambient-occlusion (SSAO/GTAO) implementation newly revised render pass infrastructure expanded shader generation for bind groups more use of generated WGSL struct types SSAO with Image-Based Lighting The main effect is that out-of-the-box, without any textures, Use.GPU no longer looks like early 2000s OpenGL. This is a problem every home-grown 3D effort runs into: how to make things look good without premium, high-quality models and pre-baking all the lights. Use.GPU's reactive run-time continues to purr along well. Its main role is to enable doing at run-time what normally only happens at build time: dealing with shader permutations, assigning bindings, and so on. I'm quite proud of the line up of demos Use.GPU has now, for the sheer diversity of rendering techniques on display, including an example path tracer. The new inspector is the cherry on top. A lot of the effort continues to revolve around mitigating flaws in GPU API design, and offering something simpler. As such, the challenge here wasn't just implementing SSAO: the basic effect is pretty easy. Rather, it brings with it a few new requirements, such as temporal accumulation and reprojection, that put new demands on the rendering pipeline, which I still want to expose in a modular and flexible way. This refines the efforts I detailed previously for 0.8. Good SSAO also requires deep integration in the lighting pipeline. Here there is tension between modularizing and ease-of-use. If there is only one way to assemble a particular set of components, then it should probably be provided as a prefab. As such, occlusion has to remain a first class concept, tho it can be provided in several ways. It's a good case study of pragmatism over purity. In case you're wondering: WebGPU is still not readily available on every device, so Use.GPU remains niche, tho it already excels at in-house use for adventurous clients. At this point you can imagine me and the browser GPU teams eyeing each other awkwardly from across the room: I certainly do. Inspector Gadget The first thing to mention is the upgraded the Use.GPU inspector. It already had a lot of quality-of-life features like highlighting, but the main issue was finding your way around the giant trees that Use.GPU now expands into. Old New Highlights show data dependencies The fix was filtering by type. This is very simple as a component already advertises its inspectability in a few pragmatic ways. Additionally, it uses the data dependency graph between components to identify relevant parents. This shows a surprisingly tidy overview with no additional manual tagging. For each demo, it really does show you the major parts first now. If you've checked it out before, give it another try. The layered structure is now clearly visible, and often fits in one screen. The main split is how Live is used to reconcile different levels of representation: from data, to geometry, to renders, to dispatches. These points appear as different reconciler nodes, and can be toggled as a filter. It's still the best way to see Live and Use.GPU in action. It can be tricky to grok that each line in the tree is really a plain function, calling other functions, as it's an execution trace you can inspect. It will now point you more in the right way, and auto-select the most useful tabs by default. The inspector is unfortunately far heavier than the GPU rendering itself, as it all relies on HTML and React to do its thing. At some point it's probably worth to remake it into a Live-native version, maybe as a 2D canvas with some virtualization. But in the mean time it's a dev tool, so the important thing is that it still works when nothing else does. Most of the images of buffers in this post can be viewed live in the inspector, if you have a WebGPU capable browser. SSAO Screen-space AO is common now: using the rendered depth buffer, you estimate occlusion in a hemisphere around every point. I opted for Ground Truth AO (GTAO) as it estimates the correct visibility integral, as opposed to a more empirical 'crease darkening' technique. It also allows me to estimate bent normals along the way, i.e. the average unoccluded direction, for better environment lighting. Hemisphere sampling This image shows the debug viz in the demo. Each frame will sample one green ring around a hemisphere, spinning rapidly, and you can hold ALT to capture the sampling process for the pixel you're pointing at. It was invaluable to find sampling issues, and also makes it trivial to verify alignment in 3D. The shader calls printPoint(…) and printLine(…) in WGSL, which are provided by a print helper, and linked in the same way it links any other shader functions. Bent normal and occlusion samples SSAO is expensive, and typically done at half-res, with heavy blurring to hide the sampling noise. Mine is no different, though I did take care to handle odd-sized framebuffers correctly, with no unexpected sample misalignments. It also has accumulation over time, as the shadows change slowly from frame to frame. This is done with temporal reprojection and motion vectors, at the cost of a little bit of ghosting. Moving the camera doesn't reset the ambient occlusion, as long as it's moving smoothly. Motion vectors example Accumulated samples As Use.GPU doesn't render continuously, you can now use <Loop converge={N}> to decide how many extra frames you want to render after every visual change. Reprojection requires access to the last frame's depth, normal and samples, and this is trivial to provide. Use.GPU has built-in transparent history for render targets and buffers. This allows for a classic front/back buffer flipping arrangement with zero effort (also, n > 2). Depth history You bind this as virtual sources, each accessing a fixed slot history[i], which will transparently cycle whenever you render to its target. Any reimagined GPU API should seriously consider buffer history as a first-class concept. All the modern techniques require it. IGN Rather than use e.g. blue noise and hope the statistics work out, I chose a very precise sampling and blurring scheme. This uses interleaved gradient noise (IGN), and pre-filters samples in alternating 2x2 quads to help diffuse the speckles as quickly as possible. IGN is designed for 3x3 filters, so a more specifically tuned noise generator may work even better, but it's a decent V1. Reprojection often doubles as a cheap blur filter, creating free anti-aliasing under motion or jitter. I avoided this however, as the data being sampled includes the bent normals, and this would cause all edges to become rounded. Instead I use a precise bilateral filter based on depth and normal, aided by 3D motion vectors. This means it knows exactly what depth to expect in the last frame, and the reprojected samples remain fully aliased, which is a good thing here. The choice of 3D motion vectors is mainly a fun experiment, it may be an unnecessary luxury. Detail of accumulated samples The motion vectors are based only on the camera motion for now, though there is already the option of implementing custom motion shaders similar to e.g. Unity. For live data viz and procedural geometry, motion vectors may not even be well-defined. Luckily it doesn't matter much: it converges fast enough that artifacts are hard to spot. The final resolve can then do a bilateral upsample of these accumulated samples, using the original high-res normal and depth buffer: Upscaled and resolved samples, with overscan trimmed off Because it's screen-space, the shadows disappear at the screen edges. To remedy this, I implemented a very precise form of overscan. It expands the framebuffer by a constant amount of pixels, and expands the projectionMatrix to match. This border is then trimmed off when doing the final resolve. In principle this is pixel-exact, barring GPU quirks. These extra pixels don't go to waste either: they can get reprojected into the frame under motion, reducing visible noise significantly. In theory this is very simple, as it's a direct scaling of [-1..1] XY clip space. In practice you have to make sure absolutely nothing visual depends on the exact X/Y range of your projectionMatrix, either its aspect ratio or in screen-space units. This required some cleanup on the inside, as Use.GPU has some pretty subtle scaling shaders for 2.5D and 3D points and lines. I imagine this is also why I haven't seen more people do this. But it's definitely worth it. Overall I'm very satisfied with this. Improvements and tweaks can be made aplenty, some performance tuning needs to happen, but it looks great already. It also works in both forward and deferred mode. The shader source is here. Render Buffers & Passes The rendering API for passes reflects the way a user wants to think about it, as 1 logical step in producing a final image. Sub-passes such as shadows or SSAO aren't really separate here, as the correct render cannot be finished without it. The main entry point here is the <Pass> component, representing such a logical render pass. It sits inside a view, like an <OrbitCamera>, and has some kind of pre-existing render context, like the visible canvas. <Pass lights ssao={{ radius: 3, indirect: 0.5 }} overscan={0.05} > ... </Pass> You can sequence multiple logical passes to add overlays with overlay: true, or even merge two scenes in 3D using the same Z-buffer. Inside it's a declarative recipe that turns a few flags and options into the necessary arrangement of buffers and passes required. This uses the alt-Live syntax use(…) but you can pretend that's JSX: const resources = [ use(ViewBuffer, options), lights ? use(LightBuffer, options) : null, shadows ? use(ShadowBuffer, options) : null, picking ? use(PickingBuffer, options) : null, overscan ? use(OverscanBuffer, options) : null, ...(ssao ? [ use(NormalBuffer, options), use(MotionBuffer, options), ] : []), ssao ? use(SSAOBuffer, options) : null, ]; const resolved = passes ?? [ normals ? use(NormalPass, options) : null, motion ? use(MotionPass, options) : null, ssao ? use(SSAOPass, options) : null, shadows ? use(ShadowPass, options) : null, use(DEFAULT_PASS[viewType], options), picking ? use(PickingPass, options) : null, debug ? use(DebugPass, options) : null, ] e.g. The <SSAOBuffer> will spawn all the buffers necessary to do SSAO. Notice what is absent here: the inputs and outputs. The render passes are wired up implicitly, because if you had to do it manually, there would only be one correct way. This is the purpose of separating the resources from the passes: it allows everything to be allocated once, up front, so that then the render passes can connect them into a suitable graph with a non-trivial but generally expected topology. They find each other using 'well-known names' like normal and motion, which is how it's done in practice anyway. Render passes in the inspector This reflects what I am starting to run into more and more: that decomposed systems have little value if everyone has to use it the same way. It can lead to a lot of code noise, and also tie users to unimportant details of the existing implementation. Hence the simple recipe. But, if you want to sequence your own render exactly, nothing prevents you from using the render components à la carte: the main method of composition is mounting reactive components in Live, like everything else. Your passes work exactly the same as the built-in ones. I make use of the dynamicism of JS to e.g. not care what options are passed to the buffers and passes. The convention is that each should be namespaced so they don't collide. This provides real extensibility for custom use, while paving the cow paths that exist. It's typical that buffers and passes come in matching pairs. However, one could swap out one variation of a <FooPass> for another, while reusing the same buffer type. Most <FooBuffer> implementations are themselves declarative recipes, with e.g. a <RenderTarget> or two, and perhaps an associated data binding. All the meat—i.e. the dispatches—is in the passes. It's so declarative that there isn't much left inside <Renderer> itself. It maps logical calls into concrete ones by leveraging Live, and that's reflected entirely in what's there. It only gathers up some data it doesn't know details about, and helps ensure the sequence of compute before render before readback. This is a big clue that renderers really want to be reactive run-times instead. Bind Group Soup Use.GPU's initial design goal was "a unique shader for every draw call". This means its data binding fu has mostly been applied to local shader bindings. These apply only to one particular draw, and you bind the data to the shader at the same time as creating it. This is the useShader hook. There is no separation where you first prepare the binding layout, and as such, you use it like a deferred function call, just like JSX. // Prepare to call surfaceShader(matrix, ray, normal, size, ...) const getSurface = useShader(surfaceShader, [ matrix, ray, normal, size, insideRef, originRef, sdf, palette, pbr, ...sources ], defs); Shader and pipeline reuse is handled via structural hashing behind the scenes: it's merely a happy benefit if two draw calls can reuse the same shader and pipeline, but absolutely not a problem if they don't. As batching is highly encouraged, and large data sets can be rendered as one, the number of draw calls tends to be low. All local bindings are grouped in two bind groups, static and volatile. The latter allows for the transparent history feature, as well as just-in-time allocated atlases. Static bindings don't need to be 100% static, they just can't change during dispatch or rendering. WebGPU only has four bind groups total. I previously used the other two for respectively the global view, and the concrete render pass, using up all the bind groups. This was wasteful but an unfortunate necessity, without an easy way to compose them at run-time. Bind Group: #0 #1 #2 #3 Use.GPU 0.13 View Pass Static Volatile Use.GPU 0.14 Pass Static Volatile Free This has been fixed in 0.14, which frees up a bind group. It also means every render pass fully owns its own view. It can pick from a set of pre-provided ones (e.g. overscanned or not), or set a custom one, the same way it finds buffers and other bindings. Having bind group 3 free also opens up the possibility of a more traditional sub-pipeline, as seen in a traditional scene graph renderer. These can handle larger amounts of individual draw calls, all sharing the same shader template, but with different textures and parameters. My goal however is to avoid monomorphizing to this degree, unless it's absolutely necessary (e.g. with the lighting). This required upgrading the shader linker. Given e.g. a static binding snippet such as: use '@use-gpu/wgsl/use/types'::{ Light }; @export struct LightUniforms { count: u32, lights: array<Light>, }; @group(PASS) @binding(1) var<storage> lightUniforms: LightUniforms; ...you can import it in Typescript like any other shader module, with the @binding as an attribute to be linked. The shader linker will understand struct types like LightUniforms with array<Light> fully now, and is able to produce e.g. a correct minimum binding size for types that cross module boundaries. The ergonomics of useShader have been replicated here, so that useBindGroupLayout takes a set of these and prepares them into a single static bind group, managing e.g. the shader stages for you. To bind data to the bind group, a render pass delegates via useApplyPassBindGroup: this allows the source of the data to be modularized, instead of requiring every pass to know about every possible binding (e.g. lighting, shadows, SSAO, etc.). That is, while there is a separation between bind group layout and data binding, it's lazy: both are still defined in the same place. The binding system is flexible enough end-to-end that the SSAO can e.g. be applied to the voxel raytracer from @use-gpu/voxel with zero effort required, as it also uses the shaded technique (with per fragment depth). It has a getSurface(...) shader function that raytraces and returns a surface fragment. The SSAO sampler can just attach its occlusion information to it, by decorating it in WGSL. WGSL Types Worth noting, this all derives from previous work on auto-generated structs for data aggregation. It's cool tech, but it's hard to show off, because it's completely invisible on the outside, and the shader code is all ugly autogenerated glue. There's a presentation up on the site that details it at the lower level, if you're curious. The main reason I had aggregation initially was to work around the 8 storage buffers limit in WebGPU. The Plot API needed to auto-aggregate all the different attributes of shapes, with their given spread policies, based on what the user supplied. This allows me to offer e.g. a bulk line drawing primitive where attributes don't waste precious bandwidth on repeated data. Each ends up grouped in structs, taking up only 1 storage buffer, depending on whether it is constant or varying, per instance or per vertex: <Line // Two lines positions={[ [[300, 50], [350, 150], [400, 50], [450, 150]], [[300, 150], [350, 250], [400, 150], [450, 250]], ]} // Of the same color and width color={'#40c000'} width={5} /> <Line // Two lines positions={[ [[300, 250], [350, 350], [400, 250], [450, 350]], [[300, 350], [350, 450], [400, 350], [450, 450]], ]} // With color per line color={['#ffa040', '#7f40a0']} // And width per vertex widths={[[1, 2, 2, 1], [1, 2, 2, 1]} /> This involves a comprehensive buffer interleaving and copying mechanism, that has to satisfy all the alignment constraints. This then leverages @use-gpu/shader's structType(…) API to generate WGSL struct types at run-time. Given a list of attributes, it returns a virtual shader module with a real symbol table. This is materialized into shader code on demand, and can be exploded into individual accessor functions as well. Hence data sources in Use.GPU can now have a format of T or array<T> with a WGSL shader module as the type parameter. I already had most of the pieces in place for this, but hadn't quite put it all together everywhere. Using shader modules as the representation of types is very natural, as they carry all the WGSL attributes and GPU-only concepts. It goes far beyond what I had initially scoped for the linker, as it's all source-code-level, but it was worth it. The main limitation is that type inference only happens at link time, as binding shader modules together has to remain a fast and lazy op. Native WGSL types are somewhat poorly aligned with the WebGPU API on the CPU side. A good chunk of @use-gpu/core is lookup tables with info about formats and types, as well as alignment and size, so it can all be resolved at run-time. There's something similar for bind group creation, where it has to translate between a few different ways of saying the same thing. The types I expose instead are simple: TextureSource, StorageSource and LambdaSource. Everything you bind to a shader is either one of these, or a constant (by reference). They carry all the necessary metadata to derive a suitable binding and accessor. That said, I cannot shield you from the limitations underneath. Texture formats can e.g. be renderable or not, filterable or not, writeable or not, and the specific mechanisms available to you vary. If this involves native depth buffers, you may need to use a full-screen render pass to copy data, instead of just calling copyTextureToTexture. I run into this too, and can only provide a few more convenience hooks. I did come up with a neat way to genericize these copy shaders, using the existing WGSL type inference I had, souped up a bit. This uses simple selector functions to serve the role of reassembling types. It's finally given me a concrete way to make 'root shaders' (i.e. the entry points) generic enough to support all use. I may end up using something similar to handle the ordinary vertex and fragment entry points, which still have to be provided in various permutations. * * * Phew. Use.GPU is always a lot to go over. But its à la carte nature remains and that's great. For in-house use it's already useful, especially if you need a decent GPU on a desktop anyway. I have been using it for some client work, and it seems to be making people happy. If you want to go off-road from there, you can. It delivers on combining low-level shader code with its own stock components, without making you reinvent a lot of the wheels. Visit usegpu.live for more and to view demos in a WebGPU capable browser. PS: I upgraded the aging build of Jekyll that was driving this blog, so if you see anything out of the ordinary, please let me know.

Keeping up appearances in tech The word "rant" is used far too often, and in various ways. It's meant to imply aimless, angry venting. But often it means: Naming problems without proposing solutions, this makes me feel confused. Naming problems and assigning blame, this makes me feel bad. I saw a remarkable pair of tweets the other day. In the wake of the outage, the CEO of CrowdStrike sent out a public announcement. It's purely factual. The scope of the problem is identified, the known facts are stated, and the logistics of disaster relief are set in motion. Millions of computers were affected. This is the equivalent of a frazzled official giving a brief statement in the aftermath of an earthquake, directing people to the Red Cross. Everything is basically on fire for everyone involved. Systems are failing everywhere, some critical, and quite likely people are panicking. The important thing is to give the technicians the information and tools to fix it, and for everyone else to do what they can, and stay out of the way. In response, a communication professional posted an 'improved' version: Credit where credit is due, she nailed the style. 10/10. It seems unobjectionable, at first. Let's go through, shall we? Opposite Day First is that the CEO is "devastated." A feeling. And they are personally going to ensure it's fixed for every single user. This focuses on the individual who is inconvenienced. Not the disaster. They take a moment out of their time to say they are so, so sorry a mistake was made. They have let you and everyone else down, and that shouldn't happen. That's their responsibility. By this point, the original statement had already told everyone the relevant facts. Here the technical details are left to the imagination. The writer's self-assigned job is to wrap the message in a more palatable envelope. Everyone will be working "all day, all night, all weekends," indeed, "however long it takes," to avoid it happening again. I imagine this is meant to be inspiring and reassuring. But if I was a CrowdStrike technician or engineer, I would find it demoralizing: the boss, who will actually be personally fixing diddly-squat, is saying that the long hours of others are a sacrifice they're willing to make. Plus, CrowdStrike's customers are in the same boat: their technicians get volunteered too. They can't magically unbrick PCs from a distance, so "until it's fully fixed for every single user" would be a promise outsiders will have to keep. Lovely. There's even a punch line: an invitation to go contact them, the quickest way linked directly. It thanks people for reaching out. If everything is on fire, that includes the phone lines, the inboxes, and so on. The most stupid thing you could do in such a situation is to tell more people to contact you, right away. Don't encourage it! That's why the original statement refers to pre-existing lines of communication, internal representatives, and so on. The Support department would hate the CEO too. Root Cause If you're wondering about the pictures, it's Hyacinth Bucket, from 90s UK sitcom Keeping Up Appearances, who would always insist "it's pronounced Bouquet." Hyacinth's ambitions always landed her out of her depth, surrounded by upper-class people she's trying to impress, in the midst of an embarrassing disaster. Her increasingly desperate attempts to save face, which invariably made things worse, are the main source of comedy. Try reading that second statement in her voice. I’m devastated to see the scale of today’s outage and will be personally working on it together with our team until it’s fully fixed for every single user. But I wanted to take a moment to come here and tell you that I am sorry. People around the world rely on us, and incidents like this can’t happen. This came from an error that ultimately is my responsibility. I can hear it perfectly, telegraphing Britishness to restore dignity for all. If she were in tech she would give that statement. It's about reputation management first, projecting the image of competence and accountability. But she's giving the speech in front of a burning building, not realizing the entire exercise is futile. Worse, she thinks she's nailing it. If CrowdStrike had sent this out, some would've applauded and called it an admirable example of wise and empathetic communication. Real leadership qualities. But it's the exact opposite. It focuses on the wrong things, it alienates the staff, and it definitely amplifies the chaos. It's Monty Python-esque. Apologizing is pointless here, the damage is already done. What matters is how severe it is and whether it could've been avoided. This requires a detailed root-cause analysis and remedy. Otherwise you only have their word. Why would that re-assure you? The original restated the company's mission: security and stability. Those are the stakes to regain a modicum of confidence. You may think that I'm reading too much into this. But I know the exact vibe on an engineering floor when the shit hits the fan. I also know how executives and staff without that experience end up missing the point entirely. I once worked for a Hyacinth Bucket. It's not an anecdote, it's allegory. They simply don't get the engineering mindset, and confuse authority with ownership. They step on everyone's toes without realizing, because they're constantly wearing clown shoes. Nobody tells them. Softness as a Service The change in style between #1 and #2 is really a microcosm of the conflict that has been broiling in tech for ~15 years now. I don't mean the politics, but the shifting of norms, of language and behavior. It's framed as a matter of interpersonal style, which needs to be welcoming and inclusive. In practice this means they assert or demand that style #2 be the norm, even when #1 is advisable or required. Factuality is seen as deficient, improper and primitive. It's a form of doublethink: everyone's preference is equally valid, except yours, specifically. But the difference is not a preference. It's about what actually works and what doesn't. Style #1 is aimed at the people who have to fix it. Style #2 is aimed at the people who can't do anything until it's fixed. Who should they be reaching out to? In #2, communication becomes an end in itself, not a means of conveying information. It's about being seen saying the words, not living them. Poking at the statement makes it fall apart. When this becomes the norm in a technical field, it has deep consequences: Critique must be gift-wrapped in flattery, and is not allowed to actually land. Mistakes are not corrected, and sentiment takes precedence over effectiveness. Leaders speak lofty words far from the trenches to save face. The people they thank the loudest are the ones they pay the least. Inevitably, quiet competence is replaced with gaudy chaos. Everyone says they're sorry and responsible, but nobody actually is. Nobody wants to resign either. Sound familiar? Cope and Soothe The elephant in the room is that #1 is very masculine, while #2 is more feminine. When you hear "women are more empathetic communicators", this is what it means. They tend to focus on the individual and their relation to them, not the team as a whole and its mission. Complaints that tech is too "male dominated" and "notoriously hostile to women" are often just this. Tech was always full of types who won't preface their proposals and criticisms with fluff, and instead lean into autism. When you're used to being pandered to, neutrality feels like vulgarity. The notable exceptions are rare and usually have an exasperating lead up. Tech is actually one of the most accepting and egalitarian fields around. The maintainers do a mostly thankless job. "Oh so you're saying there's no misogyny in tech?" No I'm just saying misogyny doesn't mean "something 1 woman hates". The tone is really a distraction. If someone drops an analysis, saying shit or get off the pot, even very kindly and patiently, some will still run away screaming. Like an octopus spraying ink, they'll deploy a nasty form of #2 as a distraction. That's the real issue. Many techies, in their naiveté, believed the cultural reformers when they showed up to gentrify them. They obediently branded heretics like James Damore, and burned witches like Richard Stallman. Thanks to racism, words like 'master' and 'slave' are now off-limits as technical terms. Ironic, because millions of computers just crashed because they worked exactly like that. The cope is to pretend that nothing has truly changed yet, and more reform is needed. In fact, everything has already changed. Tech forums used to be crucibles for distilling insight, but now they are guarded jealously by people more likely to flag and ban than strongly disagree. I once got flagged on HN because I pointed out Twitter's mass lay-offs were a response to overhiring, and that people were rooting for the site to fail after Musk bought it. It suggested what we all know now: that the company would not implode after trimming the dead weight, and that they'd never forgive him for it. Diversity is now associated with incompetence, because incompetent people have spent over a decade reaching for it as an excuse. In their attempts to fight stereotypes, they ensured the stereotypes came true. Bait and Snitch The outcry tends to be: "We do all the same things you do, but still we get treated differently!" But they start from the conclusion and work their way backwards. This is what the rewritten statement does: it tries to fix the relationship before fixing the problem. The average woman and man actually do things very differently in the first place. Individual men and women choose. And others respond accordingly. The people who build and maintain the world's infrastructure prefer the masculine style for a reason: it keeps civilization running, and helps restore it when it breaks. A disaster announcement does not need to be relatable, it needs to be effective. Furthermore, if the job of shoveling shit falls on you, no amount of flattery or oversight will make that more pleasant. It really won't. Such commentary is purely for the benefit of the ones watching and trying to look busy. It makes it worse, stop pretending otherwise. There's little loyalty in tech companies nowadays, and it's no surprise. Project and product managers are acting more like demanding clients to their own team, than leaders. "As a user, I want..." Yes, but what are you going to do about it? Do you even know where to start? What's perceived as a lack of sensitivity is actually the presence of sensibility. It's what connects the words to the reality on the ground. It does not need to be improved or corrected, it just needs to be respected. And yes it's a matter of gender, because bashing men and masculine norms has become a jolly recreational sport in the overculture. Mature women know it. It seems impossible to admit. The entire edifice of gender equality depends on there not being a single thing men are actually better at, even just on average. Where men and women's instincts differ, women must be right. It's childish, and not harmless either. It dares you to call it out, so they can then play the wounded victim, and paint you as the unreasonable asshole who is mean. This is supposed to invalidate the argument. * * * This post is of course a giant cannon pointing in the opposite direction, sitting on top of a wall. Its message will likely fly over the reformers' heads. If they read it at all, they'll selectively quote or paraphrase, call me a tech-bro, and spool off some sentences they overheard, like an LLM. It's why they adore AI, and want it to be exactly as sycophantic as them. They don't care that it makes stuff up wholesale, because it makes them look and feel competent. It will never tell them to just fuck off already. Think less about what is said, more about what is being done. Otherwise the next CrowdStrike will probably be worse.