Have you considered that objectively difficult infants/toddlers/children exist? Children with O.D.D., for instance, show symptoms early, but diagnosis usually doesn't come until much later.

Perhaps the comments you came across online were from the parents of those kids.

-A parent of a very challenging child with Level II Autism

Have you ever met someone with a true addiction to food? I'm not talking about someone with a habitual craving for sweets. I'm talking about someone who consumes food compulsively like a chain-smoker; someone who, in the absence of whatever their favorites are, will consume and consume with little regard for what the food is: an entire jar of pickles, multiple pounds of grapes, a whole rotisserie chicken, et al.

I used to be one. I once ate six baked white onions¹ in one sitting before vomiting everywhere and rethinking my life.

I broke through naturally, but I wish GLP-1s had been prevalent at the time. Want to know what made breaking it so challenging?

  1. Unlike other addictions, you have to continue consuming this one or else you will die.

  2. Nearly every social event in the USA is tied in some way to food which means that you have to exercise willpower __constantly__ if you have a social life.

  3. People are more interested in shaming you than supporting you. Most want you to fail.

I'm mentioned on there, FWIW, toward the end of the CV page. I really wish I hadn't scared Mel away after my first e-mail exchange with him.

... and sufficiently interesting to more than a very small group of persons.

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

Laserdiscs are analog.

This goes hand-in-hand with the widespread death of belief in absolute truth in the US and other western nations.

If this technology were released during the height of the Monica Lewinsky scandal, I'd wager it would have had the impact most of us expected it to have, at least for a little while.

Yes.

> Did you give it H100 / H200 access to compile and then iterate?

Yes via Lambda.ai. Also, FWIW, I run claude with --dangerously-skip-permissions and codex with the equivalent flag.

> it does amazing kernel work (Codex-5.4)

Specifically with WGMMA + TMA?

---

Once TMA gets involved both Claude and Codex spin endlessly until they dump TMA for a slower fallback.

I've observed this with Claude-Code having Opus 4.6 reasoning set to medium, high, and max; "adaptive thinking" enabled and disabled; and I've made sure to max-out thinking tokens.

I've also observed this with Codex GPT-5.4 in addition to GPT-5.3-Codex with reasoning efforts from medium to xhigh.

---

I've also observed this on the web, as mentioned in my OP, with GPT-5.4pro (Extended Pro), Gemini3-DeepThink, and Opus 4.6.

If you're working on something not truly novel, sure.

If you're using LLMs to assist in e.g. Mathematics work on as-yet-unproven problems, then this is hardly the case.

Hell, if we just stick to the software domain: Gemini3-DeepThink, GPT-5.4pro, and Opus 4.6 perform pretty "meh" writing CUDA C++ code for Hopper & Blackwell.

And I'm not talking about poorly-spec'd problems. I'm talking about mapping straightforward mathematics in annotated WolframLanguage files to WGMMA with TMA.

Ha, I do the same thing!

You're conflating an algorithm in N or Z with inherent meaning.

Let's shift over to R: Expand e*pi to repeated addition.

Think exponentiation is "repeated multiplication"? Try 2^pi.

This is a very "New World" perspective which won't cut it for the "Old World" until their countries hit their own crisis points.

Partially. Ironing/steaming is only partially automated. Folding/hanging is not.

Not GP, but I'll be forever thankful to have been able to make my career focused on embedded software.

In my line of work there's nothing to view because there's no visual component at all. If my user(s) "see" the results of my work, then it means I've catastrophically fucked up.

I spend 90% of my time working in vim within XTerm.

The closest I get to UI/UX is a UART debugging interface.

Mathematics is hardly an edge case, but SOTA models differ wildly in their ability to write proofs for unsolved problems.

Models also differ wildly in tasks like decompilation for reverse engineering.

Also, so far, the only model I've found which can competently write PTX for SM100 CUDA devices is GPT-5.4pro, but I'm willing to admit that this is more of an edge case than the aforementioned.

AFAICT, the extent to which someone finds models interchangeable is inversely proportional to the novelty of their work.

I developed a fondness for 1970s interior decor/styling even though I was born in 1988 because most of the places in my town, such as the library, were last renovated during that time.

Also, many people in my life, such as uncles & aunts, were still living in the homes they purchased in the 1970s and some design choices just can't be easily/cheaply changed.

I grew up within and around a ghost of 1970s architecture and design. As an adult I wound up moving into a suburb built in 1968 for this reason.

It's less nostalgia and more like a vague sense of familiarity that you can only scratch the surface of in your mind.