HNHacker News
TopNewBestAskShowJobs

gordonhart

1,171 karma · joined November 26, 2016

GitHub: https://github.com/gordonhart

Co-founder: https://kolena.com

Atlas of Space: https://atlasof.space

submissionscomments
gordonhart··on Times New American: A Tale of Two Fonts
A few HN commenting guidelines might apply here:

> Eschew flamebait. Avoid generic tangents. Omit internet tropes.

> Please don't use Hacker News for political or ideological battle. It tramples curiosity.

> Please don't comment about the voting on comments. It never does any good, and it makes boring reading.

gordonhart··on Kidnapped by Deutsche Bahn
Nothing here suggests that the author is American beyond his blog post being shared on an American website.
gordonhart··on How uv got so fast
`poetry install` on my dayjob’s monolith took about 2 minutes, `uv sync` takes a few seconds. Getting 2 minutes back on every CI job adds up to a lot of time saved
gordonhart··on Engineers who dismiss AI
As before, the big gap I still see is between engineers who set something up the right way and engineers who push code up without considering the bigger picture.

One nice change however is that you can guide the latter towards a total refactor during code review and it takes them a ~day instead of a ~week.

gordonhart··on GPT-5.2
> Makes no sense to me why this is the default.

You’re probably pretty far from the average user, who thinks “AI is so dumb” because it doesn’t remember what you told it yesterday.

gordonhart··on We're losing our voice to LLMs
My theory is that LLMs are accelerating [online] radicalization by commoditizing bland, HR-approved opinions. If you want to sound like a human on the internet, for better or for worse the easiest way is to say something that would make Anthropic’s safety team have a heart attack.
gordonhart··on Gemini 3
Matches my experience exactly. It's not the best at writing code but Gemini 2.5 Pro is (was) the hands-down winner in every other use case I have.

This was hard for me to accept initially as I've learned to be anti-Google over the years, but the better accuracy was too good to pass up on. Still expecting a rugpull eventually — price hike, killing features without warning, changing internal details that break everything — but it hasn't happened yet.

gordonhart··on Three kinds of AI products work
The best kind of businesses are the ones I don’t have to ask; they’ve already built a better product than what I would have asked for. That’s kinda the point the OP is making about chat vs a [good] dedicated interface.
gordonhart··on Ask HN: What Are You Working On? (Nov 2025)
Great game! The effort you put into animations and interactivity really pays off, especially when first learning how the game works.

This is a classic HN comment but I’d love a Thursday/Friday crossword difficulty equivalent in addition to the dailies which are a ~Monday.

gordonhart··on Why is Zig so cool?
Bundling a browser frontend together with your Python application.
gordonhart··on Leaving Meta and PyTorch
Fine-tuning is great for known, concrete use cases where you have the data in hand already, but how much of the industry does that actually cover? Managers have hated those use cases since the beginning of the deep learning era — huge upfront cost for data collection, high latency cycles for training and validation, slow reaction speed to new requirements and conditions.
gordonhart··on Leaving Meta and PyTorch
GP is likely saying that “building with AI” these days is mostly prompting pretrained models rather than training your own (using PyTorch).
gordonhart··on Kimi K2 Thinking, a SOTA open-source trillion-parameter reasoning model
It’s more fair to say they gave up after the Llama 4 disaster.
gordonhart··on Server DRAM prices surge 50% as AI-induced memory shortage hits hyperscalers
$1000 in 2010 is ~$1500 today — kids won't know these prices because the currency has been debased pretty rapidly in recent years.
gordonhart··on We need a clearer framework for AI-assisted contributions to open source
> As engineers it is our role to properly label our changes.

I've found myself wanting line-level blame for LLMs. If my teammate committed something that was written directly by Claude Code, it's more useful to me to know that than to have the blame assigned to the human through the squash+merge PR process.

Ultimately somebody needs to be on the hook. But if my teammate doesn't understand it any better than I do, I'd rather that be explicit and avoid the dance of "you committed it, therefore you own it," which is better in principle than in practice IMO.

gordonhart··on Gemini 3.0 spotted in the wild through A/B testing
We extensively benchmark frontier models at $DAYJOB and Gemini 2.5 is the uncontested king outside of a few narrow use cases. Tracks with the rumor that Google has the best pretraining and falls short only in tuning/alignment. Eagerly anticipating Gemini 3 as 2.5, while king of the hill, still has lots of room for improvement!

Edit: narrow use cases are roughly "true reasoning" (GPT-5) and Python script writing (the Claudes)

gordonhart··on Claude Skills
Agree — it's a big downside as a user to have more and more of these provider-specific features. More to learn, more to configure, more to get locked into.

Of course this is why the model providers keep shipping new ones; without them their product is a commodity.

gordonhart··on DOJ seizes $15B in Bitcoin from 'pig butchering' scam based in Cambodia
My first experience stepping off the plane in Cambodia was being scammed by the official issuing visas. It was $20, I gave him $50, and he didn't want to give any change back. Scams were the defining part of the tourist experience there.
gordonhart··on Ask HN: Has AI stolen the satisfaction from programming?
I largely agree. Thinking through the business requirements, hammering out the design, testing against the requirements, and reviewing the code were never the fun parts, but they were usually <50% of the job as an IC. Now that LLMs do the fun part (actually writing the code), those parts are all that’s left.

The job now feels quite different than the one I signed up for a decade+ ago. The only options I see are to accept that with a sigh or reject automation of the fun part and lose employability (worst case) or be nagged by anxiety that eventually that’ll happen.

gordonhart··on Removing these 50 objects from orbit would cut danger from space junk in half
To give an earnest answer, SpaceX deorbits its second stages, and the vast majority of those thousands of satellites are Starlinks which are in a quite low orbit (naturally decaying) and are intentionally deorbited when they reach EOL.
gordonhart··on Five years as a startup CTO: How, why, and was it worth it? (2024)
Solid definition. Being on the VC treadmill, and all of the “vibes” that come with that, is really what it boils down to.
gordonhart··on Building the heap: racking 30 petabytes of hard drives for pretraining
Backblaze reports an annual failure rate of 1.36% [0]. Since their cluster uses 2,400 drives, they would likely see ~32 failures a year (extra ~$4,000 annual capex, almost negligible).

[0] https://www.backblaze.com/cloud-storage/resources/hard-drive...

gordonhart··on If we can find information by asking GenAI, who needs the Web?
Browsing "the web" has been painful for years thanks to SEO and ads. Most people already spend their time in walled gardens, selectively jumping out to specific linked sites. Does AI fundamentally change this, or is it just a better front-end for the search engine?
gordonhart··on AlphaGenome: AI for better understanding the genome
Gemini 2.5 Pro is excellent. Top model in public benchmarks and soundly beat the alternatives (including all Claudes and that Chinese startup’s flagship) in my company’s internal benchmarks.

I’m no Google lover — in fact I’m usually a detractor due to the overall enshittification of their products — but denying that Gemini tops the pile right now is pure ignorance.

gordonhart··on Grok3 Launch [video]
_Could_ they have done the same thing with a tiny fraction of the money? Grok 3 benchmarks are SOTA for both base model and reasoning. By definition, nobody has been able to do the same thing with any amount of money (discounting o3 which has been teased but is unreleased). That may change in the future! But as of now this is the case.
gordonhart··on Grok3 Launch [video]
As a consumer, I'm just happy that base models are improving again after a ~quarter or more of relative stagnation (last big base model drop was Sonnet v2 in October). Many use cases can't use o1, r1, or o3[-mini] due to the additional reasoning latency.
gordonhart··on AI is stifling new tech adoption?
The main blocker to using the API with an alternative frontend is the cost.

Daily API usage can easily go above the $20/month subscription cost since output tokens are expensive and each new message reuses the whole message chain as input tokens. Especially true if you often upload images or documents.

gordonhart··on AI is stifling new tech adoption?
Claude is particularly bad about this, almost makes it unusable for my frontend use cases. I specify the exact tech stack in my prompt and it responds with a solution using whichever packages are available in its environment (Tailwind, shadcn/ui, etc.).

My request to model providers: the strength of your offering is its generality. Please let it be general-purpose and resist adding features (alignment, system prompts, custom stuff like Claude artifacts) that limit this.

gordonhart··on Show HN: Atlas of Space
There's a lot more out there that we just haven't seen, probably in all directions. Hard to spot stuff that far out from the sun.

The analog for cardinal directions is the direction of Vernal Equinox — the line in the ecliptic plane (Earth's orbit) formed from the Earth to the Sun at the March equinox. (there's an indicator for this in the corner of the app). This is usually treated as +x, with rotations within the ecliptic given relative to this axis. Then there's ecliptic north and ecliptic south that define +z and -z as the normal vector to the ecliptic plane.

← PreviousPage 6 of 6