jph00

9 ms·

121.
by jph00 2y ago
Highway networks add a skip connection, but LSTMs don't. Btw you might be interested in truncated backprop thru time, which we introduced in our ULMFiT paper.
122.
by jph00 2y ago
One important clarification I should have mentioned: buttons and form inputs only meet the definition of hypermedia controls when HTMX or something similar is used. Under normal HTML only links, images, complete forms, and iframes qualify,
123.
by jph00 2y ago
Since your attempt at improving the definition of hypermedia controls fails to actually mention hypermedia controls at all, I’m going to suggest that it is not in fact an improved definition.
124.
by jph00 2y ago
Academic papers are generally written to be read by peers in the field. If you're not a researcher in the hypermedia field, you shouldn't expect to immediately understand a paper without doing some work. The abstract is a paper di
125.
by jph00 2y ago
Makes me very happy to see FastHTML used to create something so great :D BTW, HN requires the proto for a clickable link, so here's the GPU puzzles link in that form: https://gpupuzzles.answer.ai
126.
by jph00 2y ago
Creating a CSV correctly has nothing at all to do with choosing a framework. In Python there's a CSV module in the stdlib that you can use -- but talking about that in a discussion of frameworks tells you nothing about the frameworks,
127.
by jph00 2y ago
The reason the FastHTML bare-bones from-scratch auth example is bare-bones is to show you the minimal pieces that need to be built if you want to do auth from scratch. That doesn't meaning doing it from scratch is a good idea -- it&#x2
128.
by jph00 2y ago
My bad - sorry!
129.
by jph00 2y ago
The API does it too for some of their models in some situations.
130.
by jph00 2y ago
(removed)
131.
by jph00 2y ago
Markdown is plain text. llms.txt is meant to be displayed in plain text format, not rendered to html.
132.
by jph00 2y ago
llms.txt is not for crawlers/scrapers, it's for creating context documents at inference time. You place it on your own site -- presumably if you create an llms.txt you're not looking to manipulate anyone, but to do your best
133.
by jph00 2y ago
From the post describing llms.txt ( https://www.answer.ai/posts/2024-09-03-llmstxt.html ): "The problem this solves is that today, constructing the right context for LLMs based on a website is ambiguous — do you: 1.
134.
by jph00 2y ago
Context windows for LLM inference are limited. You can't just throw everything into it -- it won't all fit, and larger amounts of context are slower and more expensive. So it's important to have a carefully curated set of wel
135.
by jph00 2y ago
This proposal isn’t designed to help training. It’s designed to help end-users.
136.
by jph00 2y ago
This proposal isn’t really for training. It’s for end users who want to know what information to include when they’re using models.
137.
by jph00 2y ago
Hi Jeremy here. Nice to see this on HN. To explain the reasoning for this proposal, by way of an example: I recently released FastHTML, a small library for creating hypermedia applications, and by far the most common concern I've recei
138.
/llms.txt–a proposal to provide information to help LLMs use websites(answer.ai)
3 points by jph00 2y ago|0 comments
139.
by jph00 2y ago
Dupe. https://news.ycombinator.com/item?id=41388750
140.
by jph00 2y ago
You seem to be implying that the top-down approach is a trade off that involves not breaking down the subject matter into as lower level details. I think the opposite is true - when you go top down you can keep teaching lower and lower laye
141.
by jph00 2y ago
fast.ai also does autograd from scratch - and goes further than Karpathy since it even does matrix multiplication from scratch. But it doesn’t start there. It uses top-down pedagogy, instead of bottom up.
142.
by jph00 2y ago
I haven't seen any research to support that contention -- rather, everything I've seen suggests you don't get coronaviruses like colds once off, so you don't "get it over with". Instead, they can weaken your im
143.
by jph00 2y ago
We used to think that "early and controlled exposure is preferable to a first case later in life" for chicken pox, but AFAICT that's no longer widely accepted in most western countries. Chicken pox in children can lead to s
144.
by jph00 2y ago
But the article isn’t implying that. In fact it explicitly says that helminths can be dangerous.
145.
by jph00 2y ago
Commensal bacteria are generally not airborne, whilst viruses that harm the immune system often are. So surface hygiene can be overdone, but air hygiene is a good idea.
146.
by jph00 2y ago
They would not disagree with you after, say, a measles infections, which thoroughly destroys the immune system at many levels. It’s not a good idea to confuse the impacts of vaccines and infections. Vaccines are carefully developed specific
147.
by jph00 2y ago
This is a common misunderstanding of the poorly-named “hygiene hypothesis”. For a fuller explanation of how immunologists think about this today, see this article: https://rachel.fast.ai/posts/2024-08-13-crowds-vs-frien
148.
by jph00 2y ago
Sonnet on Claude.ai can not execute code, although it often pretends otherwise.
149.
by jph00 2y ago
Yeah I feel the same way. The fact that Zed uses the system clipboard as the default register in vim mode, for instance, made it impossible to use in practice for me -- having every yank replace my clipboard was a bit of a nightmare!
150.
by jph00 2y ago
Your summary of the situation is not accurate. You can quibble about their choice of DMARC setting, but IMO it's the correct choice and best for customers. Your link says "Unable to determine STARTTLS status" -- which doesn&#

More ›