Hacker News: jbellis

New comment by jbellis in "It's time to stop doing code reviews"

jbellis — Fri, 14 Aug 2026 20:44:20 +0000

Because there are better ways now to solve the problems that code review exists to solve.

It's time to stop doing code reviews

jbellis — Fri, 14 Aug 2026 20:44:20 +0000

Article URL: https://blog.brokk.ai/its-time-to-rip-off-the-band-aid-and-stop-performing-code-reviews/

Comments URL: https://news.ycombinator.com/item?id=49304343

Points: 4

# Comments: 4

New comment by jbellis in "DeepSeek Harness"

jbellis — Thu, 13 Aug 2026 14:33:53 +0000

And that's it, that's the last lab releasing models worth coding with that didn't have a first party harness that its models are trained to use.

New comment by jbellis in "Amazon backs power plant that may become top source of US climate pollution"

jbellis — Mon, 10 Aug 2026 22:20:27 +0000

I'm so old I can remember when ars technica published actual technical content.

New comment by jbellis in "Docker Sandboxes – Disposable, isolated sandboxes for AI agents"

jbellis — Mon, 10 Aug 2026 06:39:14 +0000

Why less workflow friction?

New comment by jbellis in "Qwen3.8-Max: A New Bar for Coding and Cowork"

jbellis — Mon, 03 Aug 2026 05:53:37 +0000

The only scenario is if you have enough work to do batch inference. Using a tiny fraction of GPU capacity to decode a single request at a time just doesn't make sense, as you say.

New comment by jbellis in "Qwen3.8-Max: A New Bar for Coding and Cowork"

jbellis — Mon, 03 Aug 2026 05:50:14 +0000

Yes, look up REAP.

New comment by jbellis in "Patreon laying off 20% of staff"

jbellis — Fri, 24 Jul 2026 14:05:28 +0000

Apparently this is an unpopular opinion on HN, but a job isn't an output that companies produce, it's an agreement to compensate an employee in exchange for services, and just as the CEO shouldn't take it personally when one leaves to go work elsewhere, nobody needs to act like it's some kind of invidious social contract violation when the company no longer needs some of those services. In other words: resignation and termination are ordinary exercises of each's right to reallocate its resources.

(I am making a normative claim, not a descriptive one; I am aware that that both CEOs and laid-off employees do in fact take it personally at times.)

New comment by jbellis in "Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber"

jbellis — Tue, 21 Jul 2026 19:25:32 +0000

It's roughly equal on price and intelligence as GLM 5.2 while being ~8x faster.

New comment by jbellis in "German AI consortium releases Soofi S, an open 30B model that tops benchmarks"

jbellis — Fri, 17 Jul 2026 17:27:52 +0000

I guess you didn't read ellie's reply directly underneath that?

New comment by jbellis in "German AI consortium releases Soofi S, an open 30B model that tops benchmarks"

jbellis — Thu, 16 Jul 2026 19:26:49 +0000

It tops benchmarks because it uses them in its training data. https://x.com/eliebakouch/status/2077425801633427919

New comment by jbellis in "Data centers have hiked electricity prices on the public by $23B"

jbellis — Wed, 15 Jul 2026 03:03:43 +0000

This just isn't true. On balance, data centers are turning out to be more like the "anchor tenant" of the power grid, financing improvements for everyone.

Overview article with links to actual studies: https://cityjournal.substack.com/p/data-centers-arent-raisin...

New comment by jbellis in "Ternlight – 7 MB embedding model that runs in browser (WASM)"

jbellis — Tue, 07 Jul 2026 20:32:18 +0000

voyage 4 nano is sota at the next size up

and if you really want the best teacher models it's probably the voyage commercial APIs

New comment by jbellis in "Ternlight – 7 MB embedding model that runs in browser (WASM)"

jbellis — Tue, 07 Jul 2026 03:58:44 +0000

FWIW -- Granite r2 small is a 30M model, still small enough to run on CPU, and a good baseline for fine tunes.

New comment by jbellis in "Why low-latency Java still requires discipline?"

jbellis — Mon, 06 Jul 2026 16:23:32 +0000

Yes, disappointing given that Chronicle actually does have legit expertise here. (Chronicle Map is not very well-known even in the Java space but it's by far the best larger-than-memory Map available.)

New comment by jbellis in "Jamesob's guide to running SOTA LLMs locally"

jbellis — Fri, 03 Jul 2026 16:12:48 +0000

That's a reasonable option, just be aware that you get about 1/3 as much memory bandwidth with the M5 Pro, or 2/3 with the M5 Max [now you're at $4100 for the lowest-end]. So both your prefill (flops-bound, M5 has a lot less) and decode (bw-bound) will be slower.

New comment by jbellis in "Project Valhalla, Explained: How a Decade of Work Arrives in JDK 28"

jbellis — Fri, 19 Jun 2026 09:25:28 +0000

It's no mystery. https://en.wikipedia.org/wiki/Anders_Hejlsberg

New comment by jbellis in "The Token Compression Illusion: Why I'm Skeptical of RTK"

jbellis — Thu, 18 Jun 2026 22:51:36 +0000

I feel bad that I wasted my time reading this.

On the points in the article:

1. Yes, "gain" is a vanity metric but it's harmless, nobody is being "fooled" here.

2. This could be a problem in principle, sure, but unless you're actually vetting bug reports you're just spreading FUD.

3. Again, do you have any reason to believe that the thousands of devs using rtk are silently tanking their performance without noticing? here's a thought: instead of reporting that SOMEONE SHOULD MEASURE THIS, you could, you know, measure it yourself.

4. Good lord, what is this doing in a purportedly technical article?

5. Yes, this is inherent in the problem domain, again, nobody is being "fooled".

Yes, I'm grumpy; reading this article was a waste of time.

Bias: had my first RTK pr accepted today, so I guess I probably know more about it than this guy who got offended by "gain" and spit out the first thoughts that came to mind.

New comment by jbellis in "Looking Forward to Postgres 19: Query Hints"

jbellis — Tue, 09 Jun 2026 00:44:25 +0000

man, Tom Lane has hated query hints for literally decades

did he finally come around?

New comment by jbellis in "MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second"

jbellis — Mon, 08 Jun 2026 17:13:33 +0000

it is hard to understand what the actually meaningful innovations are here / what TileRT is bringing to the table.

- dflash: new-ish but February is ancient by the standards of the pace of AI innovation lately, I guess applying it to a 1T model is new-ish in the sense that the dflash researchers don't have the hw budget to prove that out - persistent engine kernel: this is like CUDA 101 - warp specialization: I think this just means "keep different gpu resources all busy w/ pipelining" which is CUDA 201, some of it is even baked into pytorch now - MXFP4 QAT: not new - TileRT: hard to tell what this actually does, there's a PyPi wheel with support for DS 3.2 and GLM 5 but binary only