thesehands

This is very much what dspy aims to address. Learning the incantations necessary to prompt well can be replaced by an algorithmic loop and example labelled cases.

This paper is always useful to remember when someone tells you to just use the cosine similarity: https://arxiv.org/abs/2403.05440

Sure, it’s useful but make sure it’s appropriate for your embeddings and remember to run evals to check that things are ‘similar’ in the way that you want them to appear similar.

Eg is red similar to green just because they are colors

Created by Steve Ruiz, his timeline is full of little interesting thoughts played out in the development of this, a fascinating insight into the minutiae that users take for granted but that make the difference: https://twitter.com/steveruizok

My guess is something very similar to U2Net https://github.com/xuebinqin/U-2-Net

systems also need noisy labels..

Transformers suffer from a quadratic bottleneck when calculating attention. Much work has been done investigating where memory can be saved by being more explicit on which attentions to calculate. This repo implements transformers with noted improvements

ha! no way - I just cited your paper in the comments without noticing you already had.

for all common models (GloVe, fastText, word2vec) the means across word embeddings are tightly concentrated around zero (relative to their dimensions), thus making the widely used cosine similarity practically equivalent to Pearson correlation https://www.aclweb.org/anthology/N19-1100/

The author of FastAPI https://twitter.com/tiangolo is a Spacy employee

as a reference: https://twitter.com/AndrewM_Webb/status/1183150368945049605

You can train programmers because there is time and there aren’t often lives at stake. Nurses have to hit the ground running in specialist areas where there is (I’d assume) often little time to acquire all necessary skills in order to perform at the required level

Is this not a timeseries classification, which they do implement?

At this point, the UK is very much in the same state as the US.

This is what the mypersonality dataset collected on Facebook was. The 5 axes of OCEAN - Openness, Conscientiousness, Extroversion, Agreeableness and Neuroticism for all those who took the test.

The T-Sne view for papers is a killer feature. Loved it for ICLR, glad more will be able to use it. Thank You!

Working on a blood sugar tracker for my wife as an excuse to practice with flask/sqlalchemy

Wanted to see what Facebook Scuba was, and found this comment thread that really highlights the point of this blog. https://news.ycombinator.com/item?id=13463016

I use FastAPI, its awesome - thank you, just saying...

i think the ReMarkable might be of interest

Thanks for opps daily, i was a subscriber and enjoyed thinking through the problems posted but as with others here was too busy to commit pen to paper on any of them. good luck with topstonks.com

blog post with a better explanation of what's going on from Stephen Few - the designer of the bullet chart: https://www.perceptualedge.com/articles/misc/Bullet_Graph_De...

https://youtu.be/cBjpROfzoos?t=14m20s because I both love the fast show and a challenge

That is how ASOS started, As Seen On Screen. Now doing over 2 billion in turnover a year

There may be merit in tagging each of the words with their part of speech prior to fitting the model in a similar way to sense2vec. Using your example above you would then have 2 vectors, one for leaves|VERB and one for leaves|NOUN

Thanks for posting this. Super cool to see how they are suggesting subreddits

from his portfolio: https://pro2-bar-s3-cdn-cf1.myportfolio.com/483775244caf67b7...

There have been more of these fake video stories recently. Without wanting to get bogged down in politics, I have wondered if these stories are being ramped up to provide some plausible defence to possible 'tapes' mentioned in the Steele dossier? Not necessarily a legal defence, but enough to cast some doubt as to the legitimacy in the media

I use Monzo, I can see what I spent where, and eacd 'where' is part of a broader category so I can see what is being spent by category. It is the ui that banking has needed for a long time.

It must be relatively easy to only operate the liabilities side of banking, its the assets (loans) side I would expect to be the most difficult path to navigate for new banks.

Nothing on the Monzo roadmap re loans: https://trello.com/b/9tcaMB4w/monzo-transparent-product-road...

No mention of the time delay between accepting funds and withdrawing the funds. How would a merchant refund the customer in fiat or crypto?

Indeed, no mention of insurance on the offline storage page: https://www.coinbase.com/security?locale=en