HN user

levocardia

2,449 karma
Posts0
Comments346
View on HN
No posts found.

Indeed, the B2B / no-data-retention market is still going to provide plenty of business for American companies even if every hobbyist uses open-weight models.

(1) What does it score on the private test set? (2) Does this approach generalize to, e.g., Atari or NES games, or is it just hard-coding priors about the games into the model (as Chollet specifically warned was a chronic problem in benchmarks in the original Arc-AGI paper)

This is missing the most important thing which is why HRs are so damn useful. It's because (a) survival analysis is very statistically powerful, but (b) many survival curves do not follow a very well-described parametric function. The genius of David Cox was in realizing that, when proportional hazards hold, you can just cancel out the unknown survival function and get the multiplicative hazard ratio immediately, in a statistically powerful and statistically efficient manner. Extremely useful if you are, say, trialing a novel chemotherapy drug and want to end your trial ASAP to get everyone on the intervention arm if the drug actually works.

The places where proportional hazards gets squirrely (very long observation times, crossing curves) are a small fraction of the use cases of survival analysis, and dunking on them for "not being Bayesian" or whatever misses this broader context.

Grok 4.5 14 days ago

I think the "secret sauce" is not juicing the benchmarks. Claude models just feel like they are better than the benchmarks suggest, in terms of smarts and creativity, while models from every other company feel worse relative to what you'd think from the benchmarks. Only company to really internalize Goodhart's Law, IMO.

Also this argument is easily refuted by the US Postal Service, which physically delivers individual pieces of paper in a few days, for pennies.

Nano Banana 2 Lite 22 days ago

I sense a business opportunity: a web app that de-sloppifies real estate, airbnb, and vrbo photos! See what it really looks like, thanks to the power of AI!

Yes, you really need "dollars per amount of RAM you need for standard computing tasks." Windows 11 requires a bare minimum of 4 GB of RAM, Window 10 only needed 1 GB.

...which is why there are no emerging far-right-populist political movements in welfare-heavy societies like the UK, France, and Germany, right?

Conversely: the self-help nonfiction book existed because it was the only practical way to monetize "good advice" or "good ideas" at scale. Now you can do a podcast or a youtube series and try to make money from advertising/affiliates/etc, but for a very long time, "buy my book" was the only game in town

The intentional parallels are hard to miss:

- Pope Leo XIII wrote Rerum Novarum; current Pope Leo XIV chose his name as an explicit gesture to his nominative predecessor

- This encyclical is a return to the earlier tradition of latin names (Magnifica Humanitas) for encyclicals, as opposed to many of Pope Francis' which used Italian (Laudato si')

- The official date it was signed was 135 years to the day since Rerum Novarum

- The Pope is personally appearing and speaking at the presentation; usually these encyclicals are just released at a small press conference without the Pope himself being there

Rerum Novarum intentionally tracked a third path, rejecting both socialism and laissez faire capitalism at the end of the 19th century. Gesturing so overtly towards it suggests that this new encyclical will also try to establish a "third way," grounded (as the title suggests) in human dignity.

Leo XIV has not published any encyclicals yet; this will be his first, and an extremely ambitious one at that. I also am very eager to read it.

As it turned out, the online version of MIT is not a substitute for the actual thing.

More precisely, the people motivated enough to actually do the online MIT version were often already on a high-performance trajectory, and for the people who were not, few people took the online credential seriously, despite whatever skills they acquired.

Not even prestigious ones. The school needs to sound like it has strong penalties against cheating, so there are really strict-sounding policies ("zero in the course"). But also, so many students cheat that actually enforcing these policies uniformly would hurt your graduation stats, make unhappy customers (students + parents), and hurt your revenue if you actually expel them. So the equilibrium is that the burden of reporting cheating is foisted upon professors, and it is understood -- though never explicitly communicated -- that academic integrity proceedings will be a huge administrative pain for you, the professor, and it is in your interest not to initiate them.

The outcome is predictable: unless there is a scandal of massive proportions, the issues just..."go away" on their own. With some discretion for the professor to either just look the other way, or ding the student enough to feel vindicated, but not so much as to actually hurt the university's interests.

Probably the worry that the jump from "we defend against slaughterbots" to "we built a better slaughterbot" is just around the corner

The actual "sharing" was using the Meta pixel and TikTok's equivalent, presumably so the healthcare exchanges could do retargeting or similarity-based marketing to get people to sign up for health care coverage. Which, narrowly, seems like a reasonable thing to do. But of course using the pixel automatically "shares" the data with Meta/ByteDance/whoever, and they get to use it for whatever nefarious purpose they want.