gen220

And another nontrivial one, in compound indices it’s really important that you order your columns in descending selectivity order. I.e. the column with the most unique values should go first.

In degenerate table/index situations, this could lead to index scans that are as slow as table scans, or not using an index at all! Especially common in SaaS schemas where you’re dealing with a tenancy key in many of the indicies: that tenancy key should almost always suffix the composite key not prefix it!!

If you have a horizontally scaled app (many 100s of API servers and async workers) you’re also probably going to need a connection pooling proxy like pgbouncer! with separate pools for separate connection configs (lower/higher timeouts, reader/writer). There’s a section on this that’s a bit of a stub right now, but IME tuning and configuring these connection poolers is pretty nontrivial and worth an expanded section!

I’ve seen this pointed out in other comments but I’d also strongly recommend expanding with a section on monitoring and alerting. One could write a blog post almost of this length just on monitoring :)

It's different now and children are being encouraged to transition. They aren't just told that some are naturally uncomfortable with their gender, but that conforming to a gender is abnormal. Way more are doing it than before, and even afterwards are committing suicide at high rates

The range of human (mis-)behavior is extremely wide, so I wouldn’t doubt that some doctors and patients are doing what you fear here. I don’t think we should form opinions on such a broad situation on the basis of a few extreme people and situations.

The question I would ask is would you rather have more people suffer from not having care, than some people suffer from receiving care that they later regret? The latter is something that’s incredibly sad, no doubt, but it’s an intractable and tragic side effect of offering major medical treatments and interventions in general; the “false positive” aspect is not unique to gender affirming care, either in its existence or its magnitude. (The politicizing of the false positive is, though, because gender in general is incredibly politicized).

Gender dysphoria solved by one-way-door gender affirming care is quite rare (there are many intermediary steps people can try and ultimately be helped with), but education about the issue and the availability of treatment helps people like my friend. I think it’s pretty unambiguously positive to universalize the availability of that care in the same way as any other form of healthcare and education, because it’s genuinely the only way some people can feel comfortable in their skin. And although there may be problems with the standard of care, the standard for care can only improve with time and experience.

I’m curious, do you personally know anybody who’s gone through gender-affirming-care?

For me it was a really confusing issue until I became close friends with someone whose childhood best friend is trans.

If he was born a decade earlier, he probably would have killed himself (this was the path he was on, which is incredibly tragic and all too common); the gender dysphoria invoked depression was unbearable.

Instead, he was able to work through therapy and medical care to understand his gender dysphoria and receive gender affirming care in his late teens.

Now (over a decade post treatment) he’s among the most cheerful people I’ve ever met. He inspires joy as a band teacher, is inspiringly happily married, and is raising a beautiful baby girl.

I often think about him when people talk about the issue in the abstract. The hundreds of children whose lives he’s impacted for the better, let alone the lives of his friends and family. Removing gender affirming care is implicitly saying you don’t want any of that to happen, because the logical conclusion of removing is people like him in a pit of depression and despair that often ends in suicide, all over an affliction that they did not choose.

This is where the “medically necessary” part of gender affirming care comes from.

I didn’t understand it before I knew him and his story so I don’t begrudge people who are in shows I used to walk in. But I’d encourage people to try to understand and lead with empathy and meet people where they are.

Disclosure that I work for Graphite->Cursor->??? :^)

FWIW, Origin is owned by the Graphite folks inside of Cursor: the project was announced by Tomas, Graphite co-founder, at Cursor's Compile conference today.

https://graphite.com/ if you're not familiar, is a tool built on top of GitHub!

Amazon is a large Anthropic shareholder (>5% of the cap table).

I think it’s impossible to interpret the actions of their executives here without considering this information.

Is there an explainer for people who are broadly familiar with the DB space? It sounds like you're building an equivalent to Vitesse for Postgres, but it's not super clear from the article (which I know is not the point of this, but still :) ).

Edit: It also might be interesting to point out how your solution differs from what the folks at Planetscale are building https://planetscale.com/neki

At this point I have a workflow that is fairly rote. I've yet to use a model newer than 4.6-1M-XHIGH that I trust to earn a higher ROI on that workflow, and not for lack of trying!

I personally don't believe in any sort of cabal (Occam's Razor hasn't let me down yet). Ultimately, I don't really care *why* they're wrong as much as I care *that* they have diverged from my rubber-meets-the-road measures of value.

That is concerning to me, because people are investing 100s of B's of capital based on the putative RoI putatively available to people like ourselves. When the benchmarks support this RoI thesis, but none of the anecdata does... that's really concerning!

Re: academics, I don't think any of the data academics have access to are good proxies for the work real people are doing. And for the data that are good proxies, the model labs certainly have access to the same data, and therefore the benchmark performance against those data is irrelevant.

I would encourage you to look into the open evals of some of these benchmarks (find one that actually is open-data, this is itself a good challenge), read the results generated and assess them for yourself.

This is what myself and my coworkers (and many other people in this thread) are doing on a daily basis with real stakes and real tasks – which these benchmarks are all aiming to be a proxy for. There's a real, tangible [cost]benefit to [not] using the highest-ROI models and harnesses.

The people with real incentives and skin in the game are telling you that the data diverges from "the data".

I don't mind if you don't take it seriously, our jobs are more important to us than a benchmark is.

But I wouldn't opt-out of using your own eyes and the eyes of others so easily, especially when there are literally hundreds of billions of dollars in invested capital with an interest in a certain outcome... this is how you end up in "Emperor's New Clothes" situations.

Actually anecdata I gather on my job from myself and coworkers is the only benchmark I trust anymore, because it so heavily diverges from the “benchmarks”.

at risk of quoting myself... :)

By offering frontier inference closer to cost *and* open-sourcing everything that's sub-frontier

It's two prongs! One prong is that their frontier inference pricing is significantly cheaper/closer-to-at-cost as Anthropic's.

The subject of this thread is the other prong: offering compelling models that are sub-frontier and self-hostable.

Self-hosting models and at-cost frontier models are the high-end and low-end disruptions, respectively, to Ant/OAI/etc.'s business models.

A big part of the frontier labs abilities to charge 80% gross margins on inference is having the cornered resource of frontier models.

If that inference becomes popular and valuable enough that those companies make billions of dollars in profit, those companies could use that profit to fund the building of alternative products and platforms that dis-intermediate google's relationship with the customer.

Google already has an 80% gross margin business, the biggest one in the world. Everybody wants a slice of it.

By offering frontier inference closer to cost and open-sourcing everything that's sub-frontier, they're commoditizing frontier labs' models, which inhibits their ability to durably make high gross margins on inference.

It's a strategic play.

How many failed foundation model training run cycles do you think these companies can tank before the bubble pops and deepseek/etc. catch up to frontier quality?

If Ant, OAI, etc. aren't able to make 20-30% improvements on Opus 4.6 in 2026, does the music stop playing altogether? It seems like they'd lose their ability to charge >10% gross margin on inference in a span of 3-6 months.

It’s not easy to buy such a large tranche of shares at a fixed and fair price in a single transaction!

Both parties get something they want this transaction. Alphabet gets the Berkshire halo effect and a guaranteed buyer of $10 billion worth of equities, Berkshire gets a large tranche of equity at a price they believe is fair.

I think they view Alphabet as their next Apple, and a relatively safe place to ride out whatever happens with AI: Alphabet is fairly well positioned for the upturn or the downturn, especially now with this expanded warchest of cash.

why don't we try to protect human dignity and move towards a more humane future?

I hear and have a lot respect for what you’re saying, but I’d like to propose that we thoroughly explore every other alternative first, just to make sure we aren’t missing out on something bigger and better and leaving anything on the table.

Sigh.

Of course, I’m not trying to dismiss gains from harness, actually the opposite.

But the narrative that 4.Y is an improvement over 4.X is essential to keep the model training music playing.

If 90+% of the gains come from the harness, how can you continue to justify spending billions of dollars on training and an 80% gross margin on inference on the latest model? (Reportedly what Anthropic commands on the top tier of their frontier model API billing).

So differentiating between the two (what I’m trying to do here) is really consequential!

Eh, maybe for the more luxurious properties? But plenty of landlords are operating on tight margins and they’re not legally allowed to raise rent by more than some measure of inflation reported by the state each year.

But you’re right, the tax would have to be much more punitive to crossover into the red.

If it does make it more challenging to justify the business of being a landlord, I’m all for it though. Steps towards the end goal of more New Yorkers who want to owning their primary residence.

As a New Yorker I'm thrilled. LVT/Landlording Tax next pls :)

Edit: Actually, as a property tax of nonprimary residences, is this not also effectively also a Landlording tax? Will my landlord's tax bill go up because he's not residing in my building, if my building is above the threshold assessed value of $1mm? Or are >$1mm "multi-family homes" (significant % of housing of New Yorkers in BK/Queens) exempt and this only applies to condos?

I'm curious to poll HN on this issue. Do you feel like we've had meaningful/noticeable gains in terms of your programming workflows between 4.5 and 4.7?

My 2¢, I personally feel like all of the productivity gains since 4.5's release (in November 2025!!) have come from improvements to the harnesses (cc, cursor cli, codex, opencode, whatever) AND from the context window expansion from 200k to 1M.

But the actual "raw" intelligence of the model / ability to make good decisions feels like it has plateaued since 4.5. 4.6 was maybe a small improvement, but hard to differentiate from in-context-learning with the 1M window. 4.7 if anything felt like a regression in wisdom for me and my coworkers, with it consistently making worse/lazier decisions.

One tangent, I believe sev-0 is actually "critical" (at least as how I'm used to reading it), and the higher you go the less critical something is.

IMO as a github-watcher, I think they changed their definition of what constitutes a sev-0 between sev-1 for the better. In particular, they had a few "sev-1"'s around the turn of the year that would be classified as sev-0's if they happened today.

Pre-4/22 GitHub sev-1 was a normal SaaS company's sev-0, imo. So I think their new system is more reflective of reality. My guess is that a few of their big customers bullied them to have more accurate SEV categorization.

FWIW, I'm not convinced that chart is necessarily an accurate representation of pre-acquisition reality. It would really surprise me if GitHub did not have a single sev-0 pre-acquisition, but it wouldn't surprise me if they were not formally captured and reported in a format that would make its way into their current status page's database.

FWIW, I'm not convinced that chart is necessarily an accurate representation of pre-acquisition reality. It would really surprise me if GitHub did not have a single sev-0 pre-acquisition, but it wouldn't surprise me if they were not formally captured and reported in a format that would make its way into their current status page's database.

It's a streak for continuous uptime, and yeah it is fairly depressing to imagine overseeing that :/

Oh wow, I'm in the position to be able to give a peek behind the curtain of something (validly!!) critiqued as AI slop! Exciting.

I originally made the core data functionality of this site for myself because I was curious what the uptime stats for each service were (I build something that heavily depends on GitHub), and to viz the distribution/severity of those incidents, again per-service, over time.

It involved a lot of back-and-forth, and is not a one-shotter; maybe closer to 40-50 shots over maybe ~10 hours of human time. A couple memorable things that made it complicated, irrespective of the UI: sneaky bugs around double-counting time for overlapping incidents, no GitHub API for incidents so you need to puppeteer-scrape the backlog of incidents to get historical data. Although, you all are right to call out that the CSS was three shots, though, and it shows :) I thought it looked so cool in ~January 2026 and now it gives me the ick, too!

For people who are curious about how much direction went into the information architecture/presentation, it was fairly substantial. I wanted a contribution graph style viz and it took many turns to get it working the way I wanted. The swimlane viz for selected-day-incident visualization was also me, because I love swimlane graphs.

I ended up sharing it with some folks and they wanted to reference it, so I put it on a website. So it's jokey for sure, but I take my jokes seriously! I'm grateful that people have feedback on how it can better functionally and visually :)

https://isgithubcooked.com

Normally I defend GH in the comments of these incidents but it’s been an impressively bad month by their standards, even when you filter for critical components filter out sev-2’s and 3’s.

I think of debt-avoidance sort of like teetotaling: I understand where it comes from and empathize with it but I tend to agree with you that total/dogmatic avoidance feels unnecessary and maybe even deleterious in the limit.

Like alcohol or drugs, debt can easily be abused, and there is no shortage of people and corporations waiting to make a profit from selling you debt, alcohol and drugs in difficult or joyous circumstances.

Using debt as a tool requires a degree of "know thyself" wisdom and financial literacy that many people struggle to possess in their best times, let alone hold onto in their worst times. So the "overcorrecting" edicts ("avoid debt like the plague") probably do more harm than good, because most people don't care about the finer/nuanced details of these things and want simple rules to follow through good times and (especially) bad times.

The motivation behind the statement is all about avoiding ruin, not maximizing opportunities or even happiness. They're different goals but it's easy to confuse one for the other.

As a new yorker who loves pizza and could talk about it for hours, the median pizza place in naples is way better than the median pizza place in new york. :P

"Italy" as a whole, I make no claims on.

It’s a very fractured and heterogeneous landscape where your own perspective will be warped by your personal experience.

Anthropic has a lot of the market share and dominates the mind share, but each of Codex, Devin, Cursor, Claude, et al have significantly more market usage than they had 6 months ago and each are likely still growing very quickly based on publicly-reported information.

FWIW, SpaceX (parent company of xAI) has an option to acquire Cursor for $60B that expires 7 days after their imminent IPO.

Wow, OpenAI is a stainless customer right?