bravura

27 ms·

361.
by bravura 3y ago
I mean, what you describe is an unfortunate and unavoidable issue in academia (and in the world in general). GPT4 doesn't work magic here, of course. You still have to: 1) Understand the work and the motivation. (GPT4 can help by playi
362.
by bravura 3y ago
Jeremy, if we want to roll up our sleeves on this, I'm pretty sure we can trace this even further. I'm pretty sure the earliest deep learning work from Bengio and Hinton (maybe even the unpublished circulating TRs from 2007) were
363.
NanoT5: T5 encoder/decoder training in under 24 hours on a single GPU(github.com)
1 points by bravura 3y ago|0 comments
364.
by bravura 3y ago
I've used it with great success.
365.
by bravura 3y ago
Go to arXiV and download the latex source. Plug it into ChatGPT and ask it to implement in pytorch. I've really enjoyed this approach.
366.
by bravura 3y ago
What don't you like about hubspot?
367.
by bravura 3y ago
Compression based ML approaches and hacks are fascinating. My colleague recently sent me this: https://arxiv.org/abs/2212.09410 "Less is More: Parameter-Free Text Classification with Gzip "...We propose a non
368.
by bravura 3y ago
Aside: I have an optimization algorithm and I'm curious if ELO ranking (or TrueSkill) would be a decent approximate solution. I have a sparse matrix of probabilities that I want to turn into a DAG. If x[m,n] = pr it means that m is a
369.
by bravura 3y ago
To be honest, there’s always been a tension in machine learning between the: “we won’t do it unless the theory is complete and sound” versus the “we don’t understand this but it works much better consistently so we do it” crowd. In the 90s
370.
by bravura 3y ago
This is more engineering than science. For a given task, try a bunch of approaches and keep the best one. Science would be replicable, ie demonstrate that this particular approach to prompt writing yields better prompts than some baseline p
371.
by bravura 3y ago
Isn’t that what this codebase is doing? I haven’t grokked it 100% yet. Recently I’ve been trying to engineer a prompt that I intend to run 1k times. Noticing GPT4 bug out on several responses, I’ve talked it through the problem more and ask
372.
by bravura 3y ago
You need a second phone number to do that correct? As I recall, there are very strict protections and verifications in place by big social networks to prevent bot and spam accounts. Is this actually allowed by the TOS? For individuals, not
373.
by bravura 3y ago
I guess no one you care about has a relentless abusive stalker ex. Or is a schoolteacher who wants to post their opinions freely on Facebook to their old friends. And those are “mainstream” folks, without even starting on people who are pos
374.
by bravura 3y ago
Is it possible to use threads without connecting it to my existing Instagram account? I know that Zuckerberg said: "You have one identity. The days of you having a different image for your work friends or co-workers and for the other p
375.
by bravura 3y ago
I've never used text-davinci-003 much. Why do you like it so much? What does it offer that the other models don't? What are funs things we can with it until it sunsets on January 4, 2024?
376.
by bravura 3y ago
Despite the other answers, I will tell you the grim truth: Your mileage might vary. It's an empirical question and depends upon the nature of your problem and data. You should try all three fp32, fp16, and bf16 as part our model select
377.
by bravura 3y ago
How about this? Taking a dense obtusely written ML algorithm with a ton of mathematical notation, plug it into GPT4, and get torch code? Cuz I do that often.
378.
by bravura 3y ago
Can I search algolia for comments on a particular keyword and summarize them?
379.
by bravura 3y ago
Life hack: If you are ever getting brusque / rushed / insincere medical care, make sure you have a small notepad and pen on you. Just take notes after you get possibly wrongly informed in front of the care provider... you would be
380.
by bravura 3y ago
Handwriting recognition on checks (LeCun et al 1989) wasn't really a small, purely academic experiment
381.
by bravura 3y ago
Mobile should be at the top of the roadmap, TBH. No mobile is a complete show-stopper. I think it's commendable to build a chat alternative that is snappy. But right now, it's basically going to be a sync-ed mirror with Slack for
382.
by bravura 3y ago
Since this is async, can it automatically handle provided rate limits and batch queries appropriately? Seems like everyone has to roll their own on this and it’s much nicer to have smooth tooling for it.
383.
by bravura 3y ago
LeCun et al (1989) had backprop working for digit recognition. LeCun, Bottou, et al (2002) in "Efficient Backprop" described techniques for improving backprop algorithms.
384.
by bravura 3y ago
bigvgan paper ( https://arxiv.org/pdf/2206.04658.pdf ) "We introduce a proper inductive bias of periodicity to the generator by applying a recently proposed periodic activation called Snake function (Liu et al., 202
385.
by bravura 3y ago
I'm in Europe with an iPhone and this didn't work, but I'd like to try it. [edit: telegram integration would be quite cool.]
386.
by bravura 3y ago
Nietzsche on German: "Das Verb ist der Herr und die anderen Wörter sind die Sklaven." ("The verb is the master and the other words are the slaves.") An personal observation I've noticed from living in Berlin: My Ger
387.
by bravura 3y ago
So mechanical turk, which was a robot that actually contained a human, is actually a robot that contains a human that contains a robot. We have achieved complete ouroboros.
388.
by bravura 3y ago
“Sufficiently advanced incompetence is indistinguishable from malice”
389.
by bravura 3y ago
What would be the easiest way to get his work on kindle with zero recurring effort on my part ? I love Money Stuff but hate reading for pleasure on an LCD display?
390.
by bravura 3y ago
Bias of a data set is when it doesn't reflect the true underlying distribution of nature. So a face corpus with only white faces doesn't reflect the diversity of faces one encounters in the world. With that said, unbiasing data is

More ›