Hacker News: waffletower

New comment by waffletower in "GPT-5.5"

waffletower — Fri, 24 Apr 2026 18:58:53 +0000

Neuro-modulation is an extremely interesting idea for generative diffusion models.

New comment by waffletower in "GPT-5.5"

waffletower — Fri, 24 Apr 2026 18:49:42 +0000

Here's a tautology: slacking, consciously refusing to engage agency, requires consciousness and agency. A model can't slack without them.

New comment by waffletower in "Palantir employees are starting to wonder if they're the bad guys"

waffletower — Thu, 23 Apr 2026 18:15:24 +0000

The company also chose to name itself after a fantasy scrying device corrupted by evil. There might be an ounce of self-fulfilling prophecy here.

New comment by waffletower in "Clojure: Transducers"

waffletower — Tue, 21 Apr 2026 19:31:56 +0000

It isn't messy in Clojure

New comment by waffletower in "Clojure: Transducers"

waffletower — Tue, 21 Apr 2026 19:27:06 +0000

Babashka is definitely innovative and useful

New comment by waffletower in "Clojure: Transducers"

waffletower — Tue, 21 Apr 2026 19:21:59 +0000

A not too contrived example: (require '[net.cgrand.xforms :as x]) (into {} (x/by-key :name :size (comp (x/reduce +) (map str))) example-mapseq)

New comment by waffletower in "Clojure: Transducers"

waffletower — Tue, 21 Apr 2026 19:14:36 +0000

I am a fan of Christophe Grand's xforms library -- https://github.com/cgrand/xforms -- I find the transducer nexus function, by-key, to be particularly useful for eliminating clojure.core destructuring dances when one needs group-by with post-processing.

New comment by waffletower in "Clojure: Transducers"

waffletower — Tue, 21 Apr 2026 19:11:16 +0000

That is a bit reductive. You can consider these implementations in other languages: https://github.com/hypirion/haskell-transducers -- https://github.com/ruuda/transducers

New comment by waffletower in "Car Owners Are Revolting over Tesla's Self-Driving Promises"

waffletower — Mon, 20 Apr 2026 15:47:57 +0000

I am not revolting; I shower everyday. Related story without paywall: https://www.edgen.tech/news/post/tesla-faces-growing-backlas...

New comment by waffletower in "It's OK to compare floating-points for equality"

waffletower — Sat, 18 Apr 2026 16:43:46 +0000

This is highly reductive, "they are meant to work with physical quantities", but agree that the applicability of an epsilon is entirely situational.

New comment by waffletower in "Everything we like is a psyop?"

waffletower — Fri, 17 Apr 2026 15:41:05 +0000

Taqueria style chile de arbol salsa roja is not a psyop. So I don't like the title. But I enjoy the examples in the article -- I am only Geese adjacent, and much closer to actual Canadian Geese on a daily basis, but my daughter has been critical of Geese hype.

New comment by waffletower in "Claude Opus 4.7"

waffletower — Thu, 16 Apr 2026 16:24:32 +0000

Even neuronal density is simplistic, and the dimension of size alone doesn't consider that.

New comment by waffletower in "The buns in McDonald's Japan's burger photos are all slightly askew"

waffletower — Thu, 16 Apr 2026 16:19:03 +0000

Hey! That's not McDonald's! That's マクドナルド！「Makudonarudo」

New comment by waffletower in "Apple's accidental moat: How the "AI Loser" may end up winning"

waffletower — Mon, 13 Apr 2026 20:41:32 +0000

I wouldn't characterize Apple's AI strategy as "smart". "Accidental" is a perfect descriptor here. "Apple Intelligence" and "Liquid Glass" show they are asleep at the wheel. I wrote an email to Tim last year imploring him to leverage Apple Silicon and its unified memory for private AI. I didn't tell him that I had dumped 95% of my Apple shares.

New comment by waffletower in "Afrika Bambaataa has died"

waffletower — Fri, 10 Apr 2026 19:06:36 +0000

"Bruce Dickinson" would definitely say that there wasn't enough cowbell

New comment by waffletower in "Study found that young adults have grown less hopeful and more angry about AI"

waffletower — Thu, 09 Apr 2026 18:25:59 +0000

Older U.S. taxpayers approaching retirement will bristle at word combinations such as "entitlements to retirees" as they have put enormous capital, which can clearly be summed from their large stack of W2s, into Social Security. Also, there are large segments of "retirement age" people who simply can't afford to retire.

New comment by waffletower in "Clojure: The Documentary, official trailer [video]"

waffletower — Thu, 09 Apr 2026 18:17:01 +0000

I agree -- all modern frontier models I have tested generate Clojure very well.

New comment by waffletower in "Project Glasswing: Securing critical software for the AI era"

waffletower — Wed, 08 Apr 2026 17:18:48 +0000

My comment is a completely unsubstantiated conspiracy theory: the choice of model name, Mythos, seems out of character for Anthropic models, and one can easily wonder if the model truly exists as the name suggests. It could instead be a symbolic model used by colluding companies (and perhaps even governments) to establish a reference limit upon what models will be publicly accessible, period. Probably a terrible theory as it could spell doom for frontier model developing companies' business models -- setting the bar already would likely commodify LLMs via open source models quite quickly. But the name "Mythos" is such a strange choice for this model and the circumstances surrounding its release.

New comment by waffletower in "System Card: Claude Mythos Preview [pdf]"

waffletower — Wed, 08 Apr 2026 17:07:36 +0000

a reference to the Twilight Zone episode no doubt: https://en.wikipedia.org/wiki/To_Serve_Man_(The_Twilight_Zon...

New comment by waffletower in "A Rave Review of Superpowers (For Claude Code)"

waffletower — Fri, 03 Apr 2026 16:02:49 +0000

In my experience the plan is plastic and responsive to iteration through dialogue. I don't think too much structure here is going to be desired by all users. But the "sounds good, but needs these edits" option can save a small amount of interaction. I often use the tab command that exists to modify the yes and no options. If you tab modify "Yes", I believe the plan will be executed and addendums from your additional prompting will be made afterward. By using tab modified "No", you can be clear about what parts of the plan are to be kept and what is to be immediately adjusted. Tab modified "No", definitely works as the "sounds good" option, but perhaps requires slightly more typing than you would like. "No, use most of this, but change..." I don't think there is a glaring lack here at all.