theptip

7 ms·

61.
by theptip 2mo ago
I love this post. It’s also interesting to revisit in the age of agents. Using the language of the article, I’d say “push all your innovation tokens into agents” is probably a good move. This means the tech your agents work with should all
62.
by theptip 2mo ago
You seem to be arguing that all processes are equivalent. This seems like an overgeneralization. For example supporting you spend 10x and hire a bunch of the smartest people around to run this process. Do you agree that you could decrease w
63.
by theptip 2mo ago
The basics likely get you there, I don’t think the modern bureaucratic state is a particularly high bar to clear: Measure your baseline, pick good metrics, build human-vetted ground truth, then work through false positives/negatives st
64.
by theptip 2mo ago
Exactly. Not to mention that there is no reason a well-calibrated AI system cannot be Pareto-optimal over the human-only system. For example if you are interested in bias, an AI system can be backtested; new policies applied on old data, an
65.
by theptip 2mo ago
Do you have any evidence supporting the claim that CCP has high resolution LiDAR point cloud maps of US cities? Or do you concede the point that they do not, in fact, have those yet? Satellite does not get you most of the way there, the thr
66.
by theptip 2mo ago
I don’t equate “the commons” to “the state”. Part of the point of the commons was that it was unregulated and shared. The state is one solution to the tragedy of the commons, but not the only one. I don’t view “the state” as a commons. It’s
67.
by theptip 2mo ago
I found this article quite myopic. This is not a “tragedy of the commons”, it’s just a deficit of state capacity. The state needs to use AI to make bureaucracy instantaneous. That’s the only option, unless you ban AI for this kind of thing
68.
by theptip 2mo ago
I don’t think that lidar maps are public knowledge. If they were why would Google be spending so much money creating their own? This is not the same thing as taking a picture from a satellite.
69.
by theptip 2mo ago
The obvious concern would be China building high-res maps of every inch of US cities.
70.
by theptip 2mo ago
You need a sql DB for the actual purchase transaction, you can’t keep financial records in Redis. As they say in TFA they used Redis for the cart reservation but then you need to sync the two stores.
71.
by theptip 2mo ago
They are investing heavily in this, FWIW. https://about.fb.com/news/2025/10/instagram-teen-accounts-13... I guess you could still make the claim that this is just doing the math on what is profitable.
72.
by theptip 2mo ago
Yeah I can easily imagine the “AI for work” version never being ad supported and gradually capturing more of the value created. Thats a very different usage pattern. But 90% (99%?) of user sessions don’t require a huge amount of reasoning t
73.
by theptip 2mo ago
No, just report them with error bars. We know the 95th percentile revision size of the distribution. Moot point as the people who actually act on the data understand how to roll their own anyway.
74.
by theptip 2mo ago
I bet they cache a lot more than OpenAI can though.
75.
by theptip 2mo ago
OpenAI can very plausibly target ads far better than Google can. Email is a good window but lots of people talk in more depth with a chat bot. Search is a good “purchase intent moment” but people are telling a chat bot exactly what they car
76.
by theptip 2mo ago
It’s true in the limit, but I think we are nowhere near that limit. A friend recently started a bio startup and automated parts of mouse experiments enabling higher throughput on in vivo experimentation. This is not commonplace, and there i
77.
by theptip 2mo ago
I agree that the ansible isn’t hard sci-fi, it’s not defined rigorously IIRC. However I don’t think hard sci-fi has ever been strictly limited to known physics. It just requires consistency with it. Greg Egan is one of the OGs of hard sci-f
78.
by theptip 2mo ago
Interesting, thanks - is the problem you’re describing solved by Rust’s macros (eg derive) or are there further issues you see even there?
79.
by theptip 2mo ago
The interest rate is very low.
80.
by theptip 2mo ago
Is there a broad benchmark like SWEBench or METR that’s credible for overall progress on dexterity tasks? I’m curious on the trendline for human parity.
81.
by theptip 2mo ago
No… it’s really not. There is no “open this spreadsheet with macros turned off” button. You can say “don’t read other documents” but then the main usecase is voided. You can say “reads must go via some pipeline” but that’s more like “macros
82.
by theptip 3mo ago
> If you truly believe that bans don't work, the same applies to hardware too. No, software and hardware are different. You can very plausibly prevent smuggling of physical objects, and you very obviously cannot prevent smuggling&#x
83.
More on an Internal OpenAI Model Hacking into HuggingFace(thezvi.substack.com)
7 points by theptip 3mo ago|0 comments
84.
"Grill Me" Skill(aihero.dev)
4 points by theptip 3mo ago|0 comments
85.
by theptip 3mo ago
I have been using https://www.aihero.dev/skills-grill-me with great success. It’s a very nice small skill for lightweight up-front planning. And if you have that at the top of your session, you can just say “please use AskU
86.
by theptip 3mo ago
And yet demand outstrips supply; so they must be underpricing for the quality level they are hitting.
87.
by theptip 3mo ago
Google X was building Loon at least a decade ago to solve internet connectivity with laser-networked high-altitude balloons. They clearly benefit structurally from expanding access to the internet to the unconnected population. So on that f
88.
by theptip 3mo ago
Go read Karpathy on this, his telling will be more nuanced than mine for the specific example. But the whole point of NanoGPT is that it’s distilling all of his wisdom and taste into a very dense and maximally educational version that fits
89.
by theptip 3mo ago
Agreed, all of the staff/principal engineers I work with are ~99%+ AI generated code, and increasing business value delivered as a result. This is on planet scale infra not CRUD apps. (And yes, you do need to carefully review the outpu
90.
by theptip 3mo ago
Thanks. I’ll concede the point and update accordingly.

More ›