HNHacker News
TopNewBestAskShowJobs

ViscountPenguin

573 karma · joined June 13, 2023

submissionscomments
ViscountPenguin··on Transformers Explained Visually
I think a better simplification would be to remove the position-encoding altogether, and just discuss transformers as set models.
ViscountPenguin··on Transformers Explained Visually
Yeah it's a bit odd, especially since RoPe is a lot more conceptually simple imo.
ViscountPenguin··on English: A vs. An
Ask vs acks is just a regional pronunciation variant, nothing inherently wrong with it. I have noticed that a lot of people from backgrounds with enforced language uniformity (french and Russian speakers for example) have trouble with the idea though.
ViscountPenguin··on If math is more than proof, we need to better celebrate the rest of it
It's all good until we have superhuman appreciators :)
ViscountPenguin··on AI researchers debate how close we are to recursive self-improvement
I doubt that any algorithms would be able to find a global maximum in such a large multidimensional non-convex space
ViscountPenguin··on I resigned from Anthropic today
Exactly correct, they are only capable of tasks that any school child could do; like solving millenium prize problems, hacking into tech companies, or tuning particle colliders. Nothing to see here.
ViscountPenguin··on AI Responsibility – OpenAI and Anthropic
"Unplug all possible target datacentres" has a pretty big blast radius, and people are unlikely to be coordinated or cooperative enough to actually do it.
ViscountPenguin··on Harnessing the Universal Geometry of Embeddings
The point of the paper isn't that embeddings contain information, it's that even if you don't know what model generated a set of embeddings you can still recover information from the geometry of the point cloud itself.

The fact that this is possible also adds some pretty strong restriction to the set of possible maps you could use to remove that information. No linear map will work since all embedding spaces are ~an orthonormal matrix apart, so some form of encryption is necessary. This wasn't known until very recently.

ViscountPenguin··on Sonic Pi
At least for me, the Flatpack produces no sound.
ViscountPenguin··on P99 0 ms* autocomplete for 240M domain names
Fair enough, if people visit frequently you could probably save their previous searches as a cookie too.
ViscountPenguin··on P99 0 ms* autocomplete for 240M domain names
Unfortunately this approach doesn't feel that great down here in Australia, definitely a function of latency.

I think you could get a lot closer by framing this as an optimization problem, where you use the full alphabet dictionary, but add a residual prediction which aims to cover as much of the remaining domain name tree as possible weighted by popularity. This tree could then be pre-baked and stored with the same system. This would probably get you p99 0ms even in Australia.

ViscountPenguin··on Continuous Diffusion Language Models (CDLM's)
A quick look at the continuous diffusion models linked in the post shows lots of transformer models still
ViscountPenguin··on Curvature Beziers: Improving on a timeless recipe
This comes from the binomial expansion of (X+y)^n, where X^k y^(n-k) has coefficient (n choose k). This is since you "choose" X in K of the (X+y)'s (in the Bezier case it's kind of writing (t+(1-t))^n, the abcd are to not make it equal to 1).

The choice function is also exactly the same as the pascal triangle!

ViscountPenguin··on More than half of adults in U.S. say they lack basic statistical understanding
One of the really awkward points of stats is that not only does sample size matter, but also model specification. Very small misspecifications can easily lead to infinitesimal P values over large sample sizes.

Similar things are true of Bayesian stats, leading to things like predictively oriented posteriors being studied nowadays.

ViscountPenguin··on More than half of adults in U.S. say they lack basic statistical understanding
In my experience, a not insignificant chunk of students actively in a stats class think that a p value is the probability of the null hypothesis. My money is on more like 3% of educated professionals.
ViscountPenguin··on Ask HN: What are you working on? (August 2026)
A fun game would be to quickly guess the religion of a temple? Could familiarize people with the common iconography of the various dharmic religions.
ViscountPenguin··on Prime Agent: A self-improving RLM agent
I'd assume the best harness for a model will tend to be the one that it's RLVF'd on.
ViscountPenguin··on There Will Come Soft Rains (1950) [pdf]
Cooking is pretty different to lawn mowing here, you still can't replicate the output of a cooked meal with any easily available combination of home appliances and robotics (let alone just microwaves). I think a large chunk of people would probably stop cooking if that was possible.
ViscountPenguin··on Games at the press of a button: The Rip-O-Bot (1989)
I dunno, gamedev prototyping has always been fast, but I think AI tools make it way way faster; at least based on my experience writing visualizations in Godot (and to a lesser extent, my experience scaffolding datascience viz work in python).
ViscountPenguin··on Does Speaking to Agents Like Cavemen Save 65% of Tokens? We Test
What would matter more is the token length, depends how well your tokenizer was trained on french I guess.
ViscountPenguin··on The lost civic life of movie rental stores
I've always found the concept of mailing a letter from your house really interesting. My understanding is that it's common in the US (which is why they have the little red thing on their mailboxes I think?), but I've never lived in a place where it's a thing. For such a capitalistic country, the way the US historically treated the USPS as a common civic good is really cool.
ViscountPenguin··on Gpiozero Flow
Blender shaders maybe?
ViscountPenguin··on 7.1 Earthquake in Japan
Some of the webcams are fake fwiw, I noticed a while ago when there were ads emblassened with old dates on them.

Edit: to be clear, this was random ones on YouTube, not sure about that site.

ViscountPenguin··on Show HN: Physically accurate black hole you can put in your room
It's Sea World in Queensland, but confusingly I think the company is unrelated to the American Sea World, iirc they bought a license to the name back when it was probably a lot more positive than it is now.

EDIT: Apparently not licensed, just completely unrelated.

ViscountPenguin··on Turn And Face The Strange
All of the contents of the article are basically correct (minus the agi bit), but god the tone is insufferable. I doubt it's done anything but dig sceptics further in tbh.
ViscountPenguin··on IRGC claims it destroyed Amazon's Bahrain data center
There's a pretty clear back and forth between Israel committing warcrimes, and Israels enemies commiting warcrimes back. While I can't say that it's unexpected for Israel to commit genocide, I can absolutely blame them, and certainly blame Netanyahu for their constantly escalatory strategies.
ViscountPenguin··on Fractl.art – a fractal generator with 8 octillion different patterns
I've not taken psychedelics, but presumebly a large part of the experience comes from the fact that your brain is creating the imagery, which would make it qualitatively different from imagery that your brain is having supplied to it. It might not be meaningfully possible to replicate psychedlic visuals.
ViscountPenguin··on Agent swarms and the new model economics
Seems very reasonable to me tbh, the VCS was built to facilitate human work structures, rebuilding it to facilitate autonomous work seems appropriate.
ViscountPenguin··on Claude Fable produced a counterexample to the Jacobian Conjecture
I have at least some level of experience with advanced maths, but the thing that always struck me was how amenable to automation that a lot of it seemed. While it might be creative for a human to come up with the Schreier refinement theorem, for example, I don't think it fundamentally requires creativity.

On some level, I think the most straightforwardly creative work in maths comes from building up useful generalizations; I can't see current LLMs inventing even something as straight forward as a manifold from first principles; but I also don't think that's fundamentally required to prove useful things, I think it's more of a shortcut for humans.

ViscountPenguin··on Claude Fable produced a counterexample to the Jacobian Conjecture
I dunno, this screams of goal post moving. Even if LLMs lack whatever nebulous definition of "creativity" that someone favours, there's no inherent reason for "creativity" to be required to solve any problems at all, "creativity" could just be a human method for solving problems that evolved because of it's broad applicability but is suboptimal at any given task.
← PreviousPage 2 of 8Next →