It’s a fair question. Two things:
1. It does happen. These conjectures can fall apart after seeming like a lock: https://en.m.wikipedia.org/wiki/Mertens_conjecture
2. Even if it is true, the process of proving can yield interesting insights.
6,600 karma · joined March 13, 2013
1. It does happen. These conjectures can fall apart after seeming like a lock: https://en.m.wikipedia.org/wiki/Mertens_conjecture
2. Even if it is true, the process of proving can yield interesting insights.
Gödel wait what? I thought of him as “only” the logician who killed Hiblert’s dreams and caused people to question the very foundations of mathematics.
Dude then took up physics as a hobby and trolled Einstein by discovering closed time-like curves?
It was multi-wave analysis not just visible light, IR spread can differentiate this.
It’s been missing since 2015. Probability of something being large enough to cover the star and stay on a path completely obscuring it for 10 years is shall we say, not likely.
It didn’t rage against the dying of the light, it just switched off.
Seems like combinatorics comes in second only to statistics.
This is clear because they referenced your quote about it being from the beginning.
No one was arguing that you typed in a question about unions.
For ex still cringe at people saying “anyways,” even though I know it’s a losing and pointless battle.
Not sure what the biggest driver is - being shamed while growing up, or just latent pedantry.
It feels more lightweight and consistent, and collisions aren’t super common once you adopt some conventions.
It’s a tradeoff for sure, but this preference comes from having lived in both worlds.
Sounds like there was more nuance to the story.
On one hand, cool. On the other hand wow have they been leaving a lot of performance on the table.
How does the performance compare to NVidia now?
Amazing to think there was a similar vibe in the 1920’s around trying to merge quantum mechanics and special relativity. They seemed incompatible at a basic level, yet this cool guy on the autism spectrum came along and said here ya go fixed that for you.
QFT doesn’t discard local fields and replace them with only nonlocal graph nodes.
Maybe this is coming from some speculative quantum gravity ideas.
Btw as far as throwing teraflops, the ability to scale with compute is a feature not a bug.
I realize it’s not a common business case, came across it testing how well LLMs can solve simple games. On a side note, if you bypass OCR and give models a text layout of a board standard LLMs cannot solve Scrabble boards but the thinking models usually can.
2. “Explore the Mistral AI APIs” (https://docs.mistral.ai) links to all apis except OCR.
3. The docs on the api params refer to document chunking and image chunking but no details on how their chunking works?
So much unnecessary friction smh.
In deciding it’s quite reasonable to stop, is the thinking that “mission creep” = bad, so it justifies stopping something regardless of the big picture benefits? Or do you not consider the benefits significant?
Given a business with many departments, say you notice one department has some mission creep but somehow it’s also increasing profitability for the company overall.
What do you focus on - only mentioning stopping the department mission creep, or first mentioning people should make damn sure the value is preserved as they consider any reorganization?
I like to give people the benefit of the doubt, can anyone speak to his credibility on this topic?
No it’s not. It’s that it’s oversold from a marketing perspective and comes with some big caveats.
But it does talk about big time savings for the right contexts.
Emphasis from the article:
“these things are useful”
I think many people would infer it to mean there’s a victim involved, someone’s personal credit card was stolen. If that’s the case it’s especially bizarre and not a customer you’d want to convert.
However I did find myself wondering how crucial really were the model changes?
Imagine trying to implement these features as a wrapper around frontier apis and no 70B bespoke model.
Starting with a user query, we could ask for relevant svg diagrams, fit instruction steps or guides into layout templates, or even filter photos as the demo shows.
How much less would this simple approach leave us with?
Didn’t the inverse square root trick rely on bit-level floating point and subtraction of a bias?
TSMC publishes industrial research, not pure science research like a university.
In this context the amount of detail they release is careful balance of PR benefit vs. competitive advantage vs. patent status vs. keeping their talent happy because good researches want their names in prestigious journals.