This is a JVM infrastructure optimization, not a recommendation architecture breakthrough. At Netflix scale, most of the field runs GPU-accelerated ANN search (FAISS, ScaNN) for candidate retrieval.
From the sounds of it, they're looking for items that are different from their watch history among some candidate set, which vector dbs aren't optimized for.
First mover disadvantage: you need to invent a way of doing things (ANN) before any decent tools have been built. 10 years later the world is laughing at you for using Java to do vector math.
CAR-T works because blood is accessible. You pull the cells, engineer them outside the body, and reinfuse.
Astrocytes are fixed in brain tissue, so "single injection" means intracranial surgery.. the clinical bar for injecting anything into an Alzheimer's patient's brain is high enough that this is still a long way from a treatment.
1st, this is unlike CAR-T as there is no extraction, engineering on that extraction, and then reinjection.
2nd, the only injection is intravenous. It uses a kind of virus that has been specifically engineered to cross the blood-brain barrier. That virus has a payload which infects/alters astrocytes already inside the brain, and the astrocytes become aggressive at clearing amyloid plaques.
3rd, I agree that the road to a marketable therapeutic could be a long way off.
The legal question is a distraction. GPL was always enforced by economics: reimplementation had to cost more than compliance. At $1,100 for 94% API coverage, it doesn't. Copyleft was built for a world where clean-room rewrites were painful but they aren't anymore.
I don't think it's been established that clean-room rewrites are no longer painful. We don't know if chardet could have been rewritten so easily if the original code wasn't in the training set.
Nationalization assumes AGI creates a monopoly worth seizing. But if multiple labs converge simultaneously and price it toward marginal cost, there's nothing to nationalize. The question is whether governments coordinate before commoditization makes the whole debate moot. History suggests they don't: https://philippdubach.com/posts/is-ai-really-eating-the-worl...
Normally we'd just put the site into a spam filter, but in this case I'd rather assume that you want to use HN as intended and are not just treating this place as a vehicle for promotion, which is against the rules (https://news.ycombinator.com/newsguidelines.html).
Appreciate you flagging this directly rather than just filtering.
Spamming is the exact opposite of my intentions. After Tom told me I shouldn't post links to my site anymore because long-time users complained about it making the front page three times in a month, I really cut back on posting and tried to only comment on topics where I felt I had something of value to add. If you look at my profile, plenty of my comments have no links at all or link to other sources, and based on feedback like today (https://news.ycombinator.com/item?id=47265869), I thought I'd found the right balance.
83% win rate over industry professionals across 44 occupations.
I'd believe it on those specific tasks. Near-universal adoption in software still hasn't moved DORA metrics. The model gets better every release. The output doesn't follow. Just had a closer look on those productivity metrics this week: https://philippdubach.com/posts/93-of-developers-use-ai-codi...
This March 2026 blog post is citing a 2025 study based on Sonnet 3.5 and 3.7 usage.
Given that organization who ran the study [1] has a terrifying exponential as their landing page, I think they'd prefer that it's results are interpreted as a snapshot of something moving rather than a constant.
Not sure DORA is that much of an indictment. For "Change Failure Rate" for instance these are subject to tradeoffs. Organizations likely have a tolerance level for Change Failure Rate. If changes are failing too often they slow down and invest. If changes aren't failing that much they speed up -- and so saying "change failure rate hasn't decreased, obviously AI must not be working" is a little silly.
"Change Lead Time" I would expect to have sped up although I can tell stories for why AI-assisted coding would have an indeterminate effect here too. Right now at a lot of orgs, the bottle neck is the review process because AI is so good at producing complete draft PRs quickly. Because reviews are scarce (not just reviews but also manual testing passes are scarce) this creates an incentive ironically to group changes into larger batches. So the definition of what a "change" is has grown too.
Nike is a useful test case (1) here. Brand was the whole competitive moat for them and once athletic gear commoditized, then management spent five years cutting the things that sustain it: athlete relationships, premium positioning, product development. Each cut looked (somewhat) rational on its own but none of them were, taken together.
The question is, in this new software world order, how much do brands matter vs what they've done vs network effects. I could have Claude code shit out a Facebook or Twitter clone, or an Uber clone, and have none of the baggage of Cambridge Analytica, being owned by Elon Musk, or Travis kalanick of Greyball and S. Fowler legacy. An Uber driver-turned-dev could easily stand up a competitor and give way more money to the drivers simply by not having the overhead that Uber has with lawyers and executive salaries in this age of ChatGPT. Drivers will go to where there's riders and money, and riders will go to where there's drivers and cheaper rides. (and no drivers.) If someone needs an app idea to work on, it's the incumbents, without the suck. Facebook without "People just submitted it. I don't know why. They 'trust me'. Dumb fucks."
Because looking at Truth Social and Gab, people do adopt brands as part of their identity; and Uber but for drivers, or Facebook, without the spying, are trivial to make the software side of things on now. The fact that we haven't seen a dozen Uber competitors spring up is a testament to the fact that branding is a helluva moat. It's impossible to put a dollar value on it, but ChatGPT has no moat, except that it's Chat-fucking-GPT. The original chatbot and no matter how good Claude gets, it'll never be the original.
Some of them will. And I suspect the set of markets in which they do will only increase—traditional SWE is probably dying, hard as that is to accept. But the fundamentals of engineering and business are nowhere close to going away. And those are the actually-hard parts of business.
> I could have Claude code shit out a... clone, and have none of the baggage...
Social media is different because of the network effects. Taking hold requires being there to catch refugees when something else collapses, and being sensitive to the fact of that collapse. It seems like quite a bit of luck is often required.
But you could also, you know, actually try to differentiate yourself. Figure out actual things that you'd want those sites to do differently, and have an opinion on which different approach to take.
And I especially wouldn't try with Twitter, because there are already at least two major competitors, and the people who have fled to them, in broad terms, seem to have done so based more on internal social strife than on any technical dissatisfaction. (Despite all the seemingly obvious technical issues to complain about!)
> I could have Claude code shit out a Facebook or Twitter clone, or an Uber clone
No, you couldn’t. At best you’d turn out a video game simulating Uber. The idea that all of the business is in its software seems to be one Silicon Valley perennially unlearns.
As long as we agree that code generation is capable of creating a video game simulating Uber, we're on the same page. The fact that there's more to a business than a flashy bit of software is exactly my point! The idea that Silicon Valley is one dude who doesn't get it is as old as the valley itself.
I've found the newer generation of founders understand that. The issue is they don't use HN anymore.
I've noticed a significant tone and demographic shift on the site over the past 2-3 years with more Western Europeans and Midwesterners and fewer Bay Area+NYC users, and fewer decisionmakers or decisionmaking adjacent people using the site.
And the deeply technical types who used HN largely shifted to lobste.rs.
Karrot_Kream (another longtime HN user) identified this shift as well [0]
The Bay Area/NYC and founder adjacent folks I know are mostly on group chats and messaging apps. Fundamentally the signal on this site is too low to be of use in those conversations. It's difficult to discuss decisionmaking in a forum full of random Fortune 100 employee deep in a torrid company hierarchy complaining about what their management hierarchy wants them to do.
There's also a tension between the increasing "community building" happening on HN and the Bay Area/NYC crowd. A lot of them have an extant community largely based on in-person relationships. The more HN builds its own community, the more you alienate this set of people. In other words, Slashdotification is happening more and more to HN where a set of very online tech people who don't really make decisions generate most of the chatter on this site.
The younger generation of founders meets in-person and uses iMessage and Instagram. The older generation meets in-person and uses iMessage, Signal, or WhatsApp.
The reality is, most people are in-person now and conversations that were happening on HN because of the pandemic are now being done offline.
Blind is toxic, but at least the users are cynically realistic.
Your taste buds prefer the flavor, not the brand. If they changed their name to "Caramel Diet Fanta" but kept the recipe identical, you'd still enjoy the taste.
The New Coke brand failed because people didn't like the taste, not the other way around.
New Coke is a very interesting counterpoint to the brand focus, but on the other hand, they did at the time make a very big push of it being "New" Coke. Hard to tell what would've happened if they had just swapped out the formula.
I drink Diet Coke, which is basically the same formula that became New Coke with chemical sludge instead of sugar, and it tastes pretty good to my tongue to the point where I drink it over Coke Zero, the one closer to "the real thing".
Every time I drink a Coke (or any other soft drink), the brand's baggage (good and bad) is present. Unless you're doing a blind taste test, it's impossible to avoid that.
Blocklists assume you can separate malicious infrastructure from legitimate infrastructure. Once phishing moves to Google Sites and Weebly that model just doesn't work.
I read it as Jones describing intent, not mechanics. The harder version of this argument isn't that companies want to replace workers, it's that even when AI genuinely augments productivity, the goalposts shift and you get displacement anyway. I wrote about that dynamic: https://philippdubach.com/posts/does-ai-mean-the-demand-on-l... The conclusion was that AI was supposed to free us, but inescapability might be closer to the truth.
Nvidia sells chips to whoever wins, so investing in a specific lab creates downside with no real upside. The more interesting read is whether Huang sees model providers compressing toward commodity pricing. I wrote about why that layer is structurally squeezed: https://philippdubach.com/posts/is-ai-really-eating-the-worl...
Classified multi-year contracts and government-funded compute are hard to walk away from when you're burning cash at that rate. Defense economics always do this to companies. Same thing that consolidated the primes in the 90s.
reply