Hacker News: ekelsen

New comment by ekelsen in "F-35 is built for the wrong war"

ekelsen — Tue, 21 Apr 2026 02:38:10 +0000

https://www.baen.com/Chapters/1439133476/1439133476___5.htm

Arthur c Clarke's short story, "Superiority," describes this dynamic perfectly.

New comment by ekelsen in "We Built It with Slide Rules. Then We Forgot How"

ekelsen — Thu, 02 Apr 2026 01:31:59 +0000

Read the entries in the blog that predate AI, the style is very clearly different.

I would be shocked if someone consciously or unconsciously adopted AI style so perfectly so quickly. Changing your style is not easy and if you're capable of it, probably this is not the style you'd pick.

New comment by ekelsen in "We Built It with Slide Rules. Then We Forgot How"

ekelsen — Thu, 02 Apr 2026 01:21:14 +0000

I'm willing to bet hard money.

I went to some of the oldest writing on the blog, the style is completely different.

New comment by ekelsen in "We Built It with Slide Rules. Then We Forgot How"

ekelsen — Thu, 02 Apr 2026 01:20:33 +0000

It's absolutely a discussion of the article. It's literally about how it was written.

New comment by ekelsen in "We Built It with Slide Rules. Then We Forgot How"

ekelsen — Wed, 01 Apr 2026 17:01:21 +0000

Ahhhh the AI writing! The goggles, they do nothing!

Maybe the author should be more worried about AI allowing us to be lazy and forgetting how to write.

New comment by ekelsen in "Building an FPGA 3dfx Voodoo with Modern RTL Tools"

ekelsen — Mon, 23 Mar 2026 18:16:14 +0000

I'm fairly sure not because I have proof, but because of all the "not this, but that!" clauses.

If you spend time generating text with LLMs, there is a style that you learn to recognize pretty quickly.

Also, to be clear -- I'm not saying that we shouldn't use LLMs to help us produce the best text/prose we can -- but letting them just generate a lot of the text doesn't led to the best outcome imo.

New comment by ekelsen in "Building an FPGA 3dfx Voodoo with Modern RTL Tools"

ekelsen — Sun, 22 Mar 2026 15:18:43 +0000

The project is cool, but the LLM generated blog bothers my brain.

New comment by ekelsen in "Dario Amodei – "We are near the end of the exponential" [video]"

ekelsen — Fri, 13 Feb 2026 18:31:48 +0000

Same thing was true of his interview with Tony Blair. It was such a night and day difference between the two. Tony's skill, knowledge and polish saved the interview and made it enjoyable despite the interviewer.

New comment by ekelsen in "My iPhone 16 Pro Max produces garbage output when running MLX LLMs"

ekelsen — Mon, 02 Feb 2026 15:16:33 +0000

As the comment below notes, the language should means it is recommended, but not required. And there are indeed platforms that do not implement the recommendation.

New comment by ekelsen in "My iPhone 16 Pro Max produces garbage output when running MLX LLMs"

ekelsen — Mon, 02 Feb 2026 15:05:51 +0000

Thanks for the direct evidence that the output NaN is not required to be one of the input NaNs.

New comment by ekelsen in "My iPhone 16 Pro Max produces garbage output when running MLX LLMs"

ekelsen — Mon, 02 Feb 2026 07:10:39 +0000

I also don't have access to the spec, but the people writing Rust do and they claim this: "IEEE makes almost no guarantees about the sign and payload bits of the NaN"

https://rust-lang.github.io/rfcs/3514-float-semantics.html

See also this section of wikipedia https://en.wikipedia.org/wiki/NaN#Canonical_NaN

"On RISC-V, most floating-point operations only ever generate the canonical NaN, even if a NaN is given as the operand (the payload is not propagated)."

And from the same article:

"IEEE 754-2008 recommends, but does not require, propagation of the NaN payload." (Emphasis mine)

I call bullshit on the statement "specifically binary operations combining two NaN inputs must result in one of the input NaNs." It is definitely not in the spec.

New comment by ekelsen in "My iPhone 16 Pro Max produces garbage output when running MLX LLMs"

ekelsen — Mon, 02 Feb 2026 04:34:28 +0000

Can you show me where in the ieee spec this is guaranteed?

My understanding is the exact opposite - that it allows implementations to return any NaN value at all. It need not be any that were inputs.

It may be that JavaScript relies on it and that has become more binding than the actual spec, but I don't think the spec actually guarantees this.

Edit: actually it turns out nan-boxing does not involve arithmetic, which is why it works. I think my original point stands, if you are doing something that relies on how bit values of NaNs are propagated during arithmetic, you are on shaky ground.

New comment by ekelsen in "My iPhone 16 Pro Max produces garbage output when running MLX LLMs"

ekelsen — Mon, 02 Feb 2026 02:55:13 +0000

I guess at the bit level, but not at the level of computation? Anything that relies on bit patterns of nans behaving in a certain way (like how they propagate) is in dangerous territory.

New comment by ekelsen in "My iPhone 16 Pro Max produces garbage output when running MLX LLMs"

ekelsen — Mon, 02 Feb 2026 02:22:47 +0000

I think it commutes even when one or both inputs are NaN? The output is always NaN.

New comment by ekelsen in "Dude, where's my supersonic jet?"

ekelsen — Tue, 06 Jan 2026 22:20:44 +0000

private jets already solve these problems.

New comment by ekelsen in "Dude, where's my supersonic jet?"

ekelsen — Tue, 06 Jan 2026 22:20:18 +0000

But for the private jet market, the reduction would be huge. They're already paying a premium to save time. The top end will pay an even larger premium to save even more time.

I agree with you that for commercial, anything other than super long haul (which is technically very hard), the time saving advantages are much less compelling.

Databricks Introduces OfficeQA Benchmark for Agents

ekelsen — Tue, 09 Dec 2025 18:08:08 +0000

Article URL: https://www.databricks.com/blog/introducing-officeqa-benchmark-end-to-end-grounded-reasoning

Comments URL: https://news.ycombinator.com/item?id=46208290

Points: 3

# Comments: 0

New comment by ekelsen in "Horses: AI progress is steady. Human equivalence is sudden"

ekelsen — Tue, 09 Dec 2025 04:47:10 +0000

sometimes not nearly so pleasant for them.

New comment by ekelsen in "The last-ever penny will be minted today in Philadelphia"

ekelsen — Wed, 12 Nov 2025 22:23:34 +0000

Of course that's true, that doesn't mean you should.

New comment by ekelsen in "The last-ever penny will be minted today in Philadelphia"

ekelsen — Wed, 12 Nov 2025 19:27:03 +0000

During the French Revolution, they tried to make a right angle have 100 degrees and even recomputed all new trig tables for this new standard. It obviously did not catch on :)

https://en.wikipedia.org/wiki/Gradian