HNHacker News
TopNewBestAskShowJobs

conradkay

303 karma · joined December 31, 2018

submissionscomments
conradkay··on GPT-5.5
Was 4.7 distilled off Mythos (which got 77.8%)? Interesting how mythos got 82% on terminal-bench 2.0 compared to 82.7% for GPT-5.5.

Also notice how they state just for SWE-Bench Pro: "*Anthropic reported signs of memorization on a subset of problems"

conradkay··on Introducing GPT-5.5
They say latency is the same as 5.4 and 5.5 is served on GB200 NVL72, so I assume 5.4 was served on hopper.
conradkay··on Kimi K2.6: Advancing open-source coding
Codex has a lot better limits, and 5.5 will be out soon
conradkay··on Claude Design
https://htmx.org/essays/locality-of-behaviour/

I'm not a big tailwind fan, but keeping styling in a separate file feels like a net negative

conradkay··on Claude Design
If they wanted couldn't they do something like RLHF? Instead of humans picking the best of 2 text outputs, they pick the best rendered design
conradkay··on ChatGPT Pro now starts at $100/month
Codex had 2x usage until April 1, I think when that ended there were a lot more people (like myself) who were fine on $20 but now want more usage
conradkay··on ChatGPT Pro now starts at $100/month
Anthropic is the exact same way, I think they're just trying to avoid having 5 different subscription tiers visible. Probably needing 20x is very niche
conradkay··on ChatGPT Pro now starts at $100/month
Yeah it's probably a bit better overall. 5.4 is a month newer than Opus 4.6

My guess is that 5.5 will come out soon and be significantly better so you'd want to be using Codex then, but then when Opus 5 comes out probably back to claude code

Also 5.4 has fast mode, and higher usage limits since it's cheaper

conradkay··on Muse Spark: Scaling Towards Personal Superintelligence
Stock up 9% today, very pleasant for Zuck if you do the math on his net worth :)
conradkay··on Muse Spark: Scaling towards personal superintelligence
It doesn't seem benchmaxxed, ARC AGI 2 score is quite bad (42.5%, GPT 5.4 is 76.1%) and coding is okay. But maybe this is the best Meta can do even benchmaxxing

The impressive part is multimodality, very plausible since there's less focus there by other labs (especially Anthropic)

conradkay··on Taste in the age of AI and LLMs
I don't think any human would make that table in the middle
conradkay··on Project Glasswing: Securing critical software for the AI era
I guess so, but there's a ton of buffer overflow vulnerabilities in the wild, and ostensibly it wasn't detected by static analysis

The red team post goes over some more impressive finds, and says that there's hundreds more they can't disclose yet: https://red.anthropic.com/2026/mythos-preview/

conradkay··on Project Glasswing: Securing critical software for the AI era
From the red team post https://red.anthropic.com/2026/mythos-preview/

"Most security tooling has historically benefitted defenders more than attackers. When the first software fuzzers were deployed at large scale, there were concerns they might enable attackers to identify vulnerabilities at an increased rate. And they did. But modern fuzzers like AFL are now a critical component of the security ecosystem: projects like OSS-Fuzz dedicate significant resources to help secure key open source software.

We believe the same will hold true here too—eventually. Once the security landscape has reached a new equilibrium, we believe that powerful language models will benefit defenders more than attackers, increasing the overall security of the software ecosystem. The advantage will belong to the side that can get the most out of these tools. In the short term, this could be attackers, if frontier labs aren’t careful about how they release these models. In the long term, we expect it will be defenders who will more efficiently direct resources and use these models to fix bugs before new code ever ships. "

conradkay··on Project Glasswing: Securing critical software for the AI era
Well, in a slightly indirect manner. Claude is writing a ton of code, and therefore creating a lot of security vulnerabilities.
conradkay··on Project Glasswing: Securing critical software for the AI era
For what it's worth Anthropic explicity denies that. "To state it plainly: We never reduce model quality due to demand, time of day, or server load"

Also can see https://marginlab.ai/trackers/claude-code/

It's very interesting to me how widespread this conception is. Maybe it's as simple as LLM productivity degrading over time within a project, as slop compounds.

Or more recently since they added a 1m context window, maybe people are more reckless with context usage

conradkay··on Project Glasswing: Securing critical software for the AI era
That generally makes sense to me, but I wonder if it's different when the attacker and defender are using the same tool (Mythos in this case)

Maybe you just spend more on tokens by some factor than the attackers do combined, and end up mostly okay. Put another way, if there's 20 vulnerabilities that Mythos is capable of finding, maybe it's reasonable to find all of them?

conradkay··on Project Glasswing: Securing critical software for the AI era
Going off the recent biography of Demis Hassabis (CEO/co-founder of Deepmind, jointly won the Nobel Prize in Chemistry) it seems like he's very concerned about it as well
conradkay··on Project Glasswing: Securing critical software for the AI era
I would've basically agreed with you until I'd seen this talk: https://www.youtube.com/watch?v=1sd26pWhfmg

Maybe a bad example since Nicholas works at Anthropic, but they're very accomplished and I doubt they're being misleading or even overly grandiose here

See the slide 13 minutes in, which makes it look to be quite a sudden change

conradkay··on Project Glasswing: Securing critical software for the AI era
permanent underclass has arrived :(
conradkay··on Project Glasswing: Securing critical software for the AI era
For comparison, 5x the cost of Opus 4.6, and 1.67x for Opus 4.1

I think this would be very heavily used if they released it, completely unlike GPT 4.5

conradkay··on System Card: Claude Mythos Preview [pdf]
Plausibly now. "As we wrote in the Project Glasswing announcement, we do not plan to make Mythos Preview generally available"
conradkay··on ChatGPT Won't Let You Type Until Cloudflare Reads Your React State
Not really, there's camoufox but the vast majority use modified chrome/chromium
conradkay··on Yann LeCun raises $1B to build AI that understands the physical world
He said that LLMs wouldn't have common sense about how the real world physically works, because it's so obvious to humans that we don't bother putting it into text. This seems pretty foolish honestly given the scale of internet data, and even at the time LLMs could handle the example he said they couldn't

I believe he didn't think that reasoning/CoT would work well or scale like it has

conradkay··on The changing goalposts of AGI and timelines
I'm not fan of Altman but the financial angle doesn't make much sense when he doesn't have equity in OpenAI

There's some indirect exposure and potential of being granted significant equity, but his actions don't read as being for his own wealth

As an example, he walks away with nothing in the very plausible timeline where he was fired but not then reinstated

conradkay··on GPT-5.4
I was basing it off my recollection of this:

https://www.anthropic.com/_next/image?url=https%3A%2F%2Fwww-...

basically 9/13 are very close

conradkay··on GPT-5.4
General consensus seems to be that it's still a better coding model, overall
conradkay··on GPT-5.4
Sonnet was pretty close to (or better than) Opus in a lot of benchmarks, I don't think it's a big deal
conradkay··on Fastest Front End Tooling for Humans and AI
It seems like all you have to do is paste 2-3 prompts
conradkay··on Lucidrains (Phil Wang) GitHub account has been suspended
Projects are here: https://codeberg.org/lucidrains

From his description: "Github suspended my account without warning or reason, so here for the duration"

conradkay··on Experts Have World Models. LLMs Have Word Models
Top LLMs are still very bad at poker, see this breakdown of a recent Kaggle experiment: <https://www.youtube.com/watch?v=jyv1bv7JKIQ>

What do you mean by sweep training here?

← PreviousPage 3 of 6Next →