postalcoder

13 ms·

391.
by postalcoder 2y ago
and i was just planning to go to sleep…
392.
by postalcoder 2y ago
i’m at the point where i don’t trust any markdown formatted text. it’s actually become an anti signal which is very sad because i used to consider it a signal of partial technical literacy.
393.
by postalcoder 2y ago
Do I smell astroturfing here? Ton of comments, lots of fascination and compliments glazing a post that has a few hundred words saying nothing. On top of that, I don't see how using a 2D reduction of embeddings is any more rigorous than
394.
by postalcoder 2y ago
articles like this make me wish I could be reading it on a magazine
395.
by postalcoder 2y ago
This and the underlying tweet makes me feel seen. Why does developing with Google have to be so hard?
396.
by postalcoder 2y ago
LocalLlama is the place. Really high quality people and discussion there and surprisingly collaborative. Very different than most subreddits.
397.
by postalcoder 2y ago
Graph RAG is very cool and outstanding at filling some niches. IIRC, Perplexity's actual search is just BM25 (based a lex fridman interview of the founder).
398.
by postalcoder 2y ago
To be fair, that only works if you keep chunk windows static.
399.
by postalcoder 2y ago
Yup. Caching is very nice.. but the framing is weird. "Introducing" to me, connotes a product release, not a new tutorial.
400.
by postalcoder 2y ago
To add some context, this isn't that novel of an approach. A common approach to improve RAG results is to "expand" the underlying chunks using an llm, so as to increase the semantic surface area to match against. You can furt
401.
by postalcoder 2y ago
I wholly agree with you. I've been using every model extensively since early the Davincis and I strongly believe that gpt-4-0314 was the best model they've released to date. It's poor performance on benchmarks drives my skept
402.
by postalcoder 2y ago
without further explanation this is just a bottom of barrel type comment. the “science” of what convicted the man is as flimsy as it gets and yet you’d want to discredit the man’s case because his representation’s chosen post-nominal? Pleas
403.
by postalcoder 2y ago
I took out Glaive so as not to give them free publicity – all I did was mess up the formatting of my comment. And yes, you're correct. Glaive employee(s) contributed to the model uploaded on HF.
404.
by postalcoder 2y ago
An AI engagement farmer on twitter claimed to create a llama 3.1 fine tine, trained on "reflection" (ie internal thinking) prompting that outperformed the likes of Llama 405B and even the closed source models on benchmarks. The gu
405.
by postalcoder 2y ago
It's not. "Highest number of sales" can mean either "highest revenue" or "most units sold". The author ran with "Nikon Sold More Z9 Cameras than..." when that's 1. not what management said a
406.
by postalcoder 2y ago
Yes indeed. The arguments of long ago were truly unhinged. Every debate used to boil down to 100% crops of ISO 6400 images. Now, each camera system has meaningfully different properties. I've fallen in love with the simplicity of Fuji,
407.
by postalcoder 2y ago
AF is still Fuji's achilles heel, even on the newer models. That said, AF-S with multipoint focus is solid and face detection has gotten much better. AF-C is decent, but not great - the lens motor is the issue. Video AF is terrible for
408.
by postalcoder 2y ago
Agreed. I used to be a hardcore Nikonite (I'd argue with Canon people on Dpreview forumns) but the form factor of these cameras are just disgusting. I made the switch to Fuji's X-T* and X100* line a few years ago and I haven'
409.
by postalcoder 2y ago
The submission to comment ratio gives me the sense that this may be a bot used for paid vote farming. OP feel free to correct me, as I don't mean disrespect.
410.
by postalcoder 2y ago
They don't provide unit numbers, likely for a reason. Look at their earnings presentation. If you go purely on camera unit sales, this wasn't even their best selling year (trailing twelve months) for cameras on a unit basis in the
411.
by postalcoder 2y ago
I don't want to pour water on this announcement, but I do want to add context before we run to attach a narrative here. The most probable explanation for the extra-high sales figures is historic yen weakness. Management even point to i