HNHacker News
TopNewBestAskShowJobs

mattxxx

355 karma · joined September 30, 2014

x_x
submissionscomments
mattxxx··on AI engineers claim new algorithm reduces AI power consumption by 95%
That's interesting.

Obviously, energy cost creates a barrier to entry, so reduction of cost reduces the barrier to entry... which adds more players... which increases demand.

mattxxx··on Counterintuitive Properties of High Dimensional Space (2018)
Yea - high dimensional spaces are weird and hard to reason about... and we're working very frequently in them, especially when dealing with ML.
mattxxx··on Scrum's “Product Owner” Problem
The article opens with:

> In Scrum, Product Owners have sole authority over the Product Backlog; they control what to build and when to build it. Engineers are merely ticket-takers, implementing one instruction after another.

I think this immediately starts the article off in the wrong place. An Engineer that is a "ticket-taker" is relegating themselves to that role; they have more agency than that.

If you're able to cogently explain why something needs to done, and are additionally weighing product value while doing that, you will be heard. Stakeholders are not the problem here.

mattxxx··on Dito – an advanced Layer 7 reverse proxy server written in Go
^ This.

Before seeing this here, I went down a rabbit hole on why-anyone-cares about the OSI model, especially as a descriptor for their golang project. It seems to be just a classification that one person found useful, and people treat like an interesting thing.

Separately, we need more deprogrammers in the world.

mattxxx··on The irrational hungry judge effect revisited (2023)
> Is it difficult to believe that a judge would be able to predict at least some outcomes from a single paragraph?

It's difficult to believe that the decision should be made quickly. When making decisions about the trajectory of someone's life, they should be made with care. Having a system that removes snap judgements and bias is important, and I would say that even a quick-scan and re-ordering of documents is a form of bias.

mattxxx··on Deactivating Facebook for just a few weeks reduces belief in fake news
I would love if there was a Nielsen-rating equivalent for fake news outlets.

Having worked in AdTech and Pulisher Analytics for about a decade, I know that orgs are measuring fake news viewability and content type + quality, so it'd just be nice to see a company making that digestible for people.

mattxxx··on Show HN: gpudeploy.com – "Airbnb" for GPUs
Doesn't this just raise the price of gpu's, and further encourage people to hoard them?
mattxxx··on Timeline of the xz open source attack
A service that measured "credibility" or "activity" of a username/email could be really useful here. At least, it would be a leading indicator that something _might_ be up. In particular the aside here about the email addresses are suspect:

https://research.swtch.com/xz-timeline#jia_tan_becomes_maint...

would be useful info for Lasse Collins before taking the pressure campaign seriously.

mattxxx··on Banning open weight models would be a disaster
Agreed here - Stopping general access is going to not have any impact on well-funded criminals and bad-acting governments.
mattxxx··on Barcelona’s Sagrada Familia will be completed in 2026
Here's a really cool article I read a while back on using computers and geometry to finish Gaudi's work:

https://www.researchgate.net/profile/Gilfranco-Alves/publica...

It's interesting using computational tools to finish something in spirit, that started long before their mainstreaming.

mattxxx··on Game of Life, simulating itself, infinitely zoomable
very beautiful application of games of life as computational tools, simulations of simulations of simulations.

you're hired

mattxxx··on Grok
I respect the openness here! This is the future that I want to see
mattxxx··on How AI is disrupting the demand for software engineers: data from 20M jobs
Larger tech companies are likely betting on being able to replace a lot of coding jobs with ML. Additionally, I think it's likely that we're indeed even training AI to replace Engineers with code generation tools. That is, code-generation tools offer an instant feedback loop on "Is this model able to write acceptable code?"

Nvidia's CEO openly stated AI will replace coding: https://techwireasia.com/03/2024/nvidias-ceo-jensen-huang-ai...

mattxxx··on The worst possible way to be rejected for a job
Stories like this confirm my opinion about how much bad management there is out there.
mattxxx··on Stable pseudonyms create a more civil environment than real names: study (2021)
Agreed here. Using a real name is a deal breaker for a lot of people, and as a result you lose their contributions, which hypothetically could be more informed.
mattxxx··on I is for Intent
Is it hard to teach "Intent", because the treatise here isn't well thought out?

The article starts with building a straw person, Stanley (who is weirdly homophobic for some reason?), and doesn't really articulate "here's how you code with intent" and "why coding with intent is important". Instead, it tries to tear apart people who don't use intent.

mattxxx··on I is for Intent
I know. This author is writing from a pretty weird place, especially if they're first categorizing fictional "Stanley" as backend Engineer.
mattxxx··on In 2024, please switch to Firefox
Use Firefox if only to put up a decent fight against Chrome's massive play to control your data and the ad ecosystem.
mattxxx··on Against Superficial Simplicity (2022)
I think the article would more strongly articulate its thesis if it offered some suggestions to improve the current notification system.
mattxxx··on Pragmatic Versioning – An Alternative to Semver
I think "BIGRELEASE" is not accurate representation of how semver is used. I always understood it as something that introduces a non-backwards compatible change in the API.

This is so that you can specify ranges of api versions that your program could be built with. For reference, here's where I took my understanding: https://semver.org/

mattxxx··on Should you split that file?
Hm this is a good idea. Or similarly, be able to make a "meta file" that maybe contains every function that was used in a chain... Cool thought
mattxxx··on Code is run more than read
I generally agree with the thesis here that "users and business are under-represented by people that are building and maintaining software", but the greater-than comparison over simplifies it, right?

It's something like "Net revenue" and "net user happiness" trump "code prettiness". Additionally, in the "maintainer > user" scenario, is it all maintainers? Is it one particular maintainer that has strong opinions (maybe which don't reflect what the business cares about)?

I'd suggest to replace the inequality mentality with an optimization problem, like: Find a version of the code that maximizes user happiness and minimizes maintenance cost.

If there are two solutions that give roughly the same values for the above, then accept either.

mattxxx··on The Dunning-Kruger effect is autocorrelation
Wait... but what if this is DK? What if my comment is DK??
mattxxx··on Don't use DISTINCT as a "join-fixer"
Agreed - I think SQL is great, but the JOIN operator has been problematic for me with analytic queries and queries where you want to enrich data across different tables.

I think the problem is fundamentally that if you take a record, and split it across 2 tables, then there's no easy operation to merge them and guarantee distinctness.

Separately, even though there are some %-breakdowns in the article, I don't think the `DISTINCT` sort is so much of a problem in most cases. Usually the actually joining is the problem for any truly large dataset, so while I trust the diagrams shown, I think those %'s change dramatically based on the underlying data.

mattxxx··on Chrome now tracks users and shares a “topic” list with advertisers
Glad this is getting press!

Google was playing 3D chess when they started developing Chrome, and when they started making moves to strip away things like user agent strings, they were really just making the final moves of a campaign set years before to create a walled garden of ads data.

mattxxx··on How to communicate when trust is low without digging yourself into a deeper hole
Note that this contrasts with BLUF, a good standard for direct, factual communication: https://en.wikipedia.org/wiki/BLUF_(communication)

In particular, the parts where you are adding buffers + prefaces ahead of the important content.

That said, non-violent communication is super important, so learning to retract from more direct, factual styles in exchange for a style that make people feel understood and respected as a colleague is important.

Tense situations usually evolve because both parties are adding slight amounts of intensity/unhappiness to the situation iteratively, and the only way to get them back to peaceful is by making commitment to making _the next interaction_ some amount less tense/hostile. Do that repeatedly and you'll get somewhere.

PS: Also Stoicism helps especially with these things... great, practical text on it: "The Practicing Stoic: A Philosophical User's Manual" by Farnsworth

mattxxx··on Every phone should be able to run personal websites
I really like this in a cyberpunk-kind-of-way, but think it's impractical from a data usage perspective + the fact that your ip should change frequently on any mobile device.

In the abstract, I feel like editing a file on your phone vs. editing a file somewhere "in the cloud" isn't so different if the interface makes it indistinguishable, so I don't know if you get any real benefit to toting around your webserver...

but I like the notion of a homegrown, self-managed webserver that you keep in your pocket.

mattxxx··on Ask HN: How do you exercise?
Muay Thai - 5 hours a week. Great aerobic exercise; get flexible; I highly recommend that anyone (who hasn't tried it) give it a go!
mattxxx··on Israel deploys remote-controlled robotic guns
A few thoughts:

- Operation of the gun can be used as training data for AI

- Eventually we won't know when the gun becomes completely guided by AI

mattxxx··on SQLite or PostgreSQL? It's Complicated
Some context: I've worked with massive PostgreSQL db's, containing millions of daily records spanning multiple years (allowing for potentially massive numbers of returned rows). I also exclusively MVP simple db architectures in SQLite (think: no massive joins or need to partition data).

Best takeaways here:

- SQLite kills it. It appears untuned PostgreSQL is only 2 seconds (???) faster than untuned SQLite; always try to kickoff a project with SQLite

- PostgreSQL has a performance cliff when query responses leave memory. This is extra true when an index no longer fits in memory. worker_mem is critical for pushing back this performance cliff

Side note: Tuning postgres can be really painful, but it's also a really dope tried-and-true db.

← PreviousPage 2 of 6Next →