aquajet

I've been experimenting with a search engine/personal assistant to index the long list of material for me, and feed back snippets/articles as I want. I guess the analogy is having a person "read" the material for me and use it to answer general questions I have/point me to the article.

If you want to try it for yourself id be happy to give you beta access. I'm still experimenting for the best UX.

> Training a sequence-to-sequence language model (T5 derivative) on our custom dataset designed for factual generation yielded much better results with less hallucination.

Could you elaborate more on this or point to a paper/benchmark results?

How do you get it to work with sites that are javascript-heavy or have heavy bot restrictions? I've been trying to make something similar but spent weeks just fighting edge cases on HTML not caching correctly.

I've been using upvotes as a way to save things, and then using a script (https://github.com/anishthite/HN-Saved-Links-Export) to export them to json so I can search through them.

Why so much switching?

I've been working on a solution to this: https://diva.so mainly cause I had the same issue.

It's a third-party service unfortunately, but it can index the contents of your bookmarks + other sources to let you search them. I haven't got it to work locally yet since the search needs a decently large server to work (I want to use a LLM eventually) but I do encrypt all data. I don't know how it compares to similar systems linked here, but I'm down to try to help you out to set it up in a way to your liking.

Interesting, I'm curious, what phone/carrier was that? If you were able to get a decent sound on one phone, it sounds like then there's nothing intrinsically/insurmountably wrong with cell phones.

I grew up just as landlines were giving way to cell phones. How did high fidelity let us pick up on emotional state?

I go on https://lobste.rs pretty frequently in addition to reddit and twitter. Also do discord groups count as forums?

Hi! Co-Founder of Diva Labs

We made a took to perform semantic search across RSS feeds, without needing a Diva account

We've been working a lot with information management (see website for more) and made this as a demo for our algorithm as well as a way to test out potential backend models. We found that the tool actually helps a lot with searching sets of information especially since RSS feeds are already curated datasets.

We are just beginning to test out more algorithms and would love to hear feedback on the model, rss scraping, and the rest of our site!

0.0075 per message. I know it's not a lot but I'd rather not have to spend anything.

I've just started to index everything and then search it as I need it rather than spend time organizing. File storage is pretty cheap so it was the least expensive of all the options

I also made a chrome extension that would show similar saved items to whatever is currently open in my tab, which also ended up working well when I forgot what I had indexed.

It worked surprisingly well for me personal use, so now am trying to sell it as a product https://www.diva.so

Disclaimer: am a cofounder for Diva.

no paywall: https://archive.ph/DEZJ5

It's also a revered part of "the formula", mainly for aesthetics. I'm guessing they want the cars to look more like the other formula series.

Is there a reason fish isn't posix compliant? From my experience (which is rudimentary tbf) there are only detriments due to the lack of sh scripts not working out of the box.

I made a similar site to Tuchu a while ago that attaches the highlights to the pdf: https://anishthite.github.io/ailight/. It's a bit slow though, I've been trying to get it to run faster.

There's a web comic about something like this: https://www.powernapcomic.com/. The story is set in a world where the majority of people take a drug that eliminates the need for sleep, but the main character is allergic? to it so suffers a "handicap" for being what we would call normal.

Location: CT or Atlanta, GA (student)

Remote: Open to remote and in-person

Willing to relocate: Yes, anywhere

Technologies: Python, Rust, C, Bash, JS, Git, Linux, SQL, Flask, React, PyTorch, TensorFlow, Numpy, Pandas

Résumé/CV: http://anishthite.github.io/Anish_Thite_Resume.pdf

Email: anishthite@gmail.com

BS CS student graduating in Spring 2021, open to internships or full-time offers. I have worked on some ML and Full-stack projects, and am interested in applying NLP techniques to interesting problems (open to exploring other fields though). I am open to any offer though (I just want a job tbh). I'm interested in startups and small to mid sized companies. Ideally I want to have high impact within the company and high impact on the world.

The Best discussion: https://news.ycombinator.com/item?id=4755470

Thanks for the tip, but I meant more as in a general sense. I find it easier to remember things I learned from long-form text rather than comments (across many sites so it's not something only on HN). For example, I can remember things I learned from a book I read in 5th grade but I can't remember an interesting comment I read last week about irrational investors (at least not without looking it up) even though I think I gave the same amount of attention to both.

> especially not an hypothetical open source organization like discussed here

Hypothetical open source organization is not hypothetical: https://www.eleuther.ai/home

However, the compute constraints are correct. We plan on using TFRC's TPUs in order to replicate GPT-2 (with a few changes to support scaling to GPT-3 and beyond as well as the larger context length). With success on a GPT-2 model we hope to garner enough interest to get more compute to train a full-sized GPT-3 model. After that, it's 1T or bust.

Location: CT or Atlanta, GA (student)

Remote: Open to remote and in-person

Willing to relocate: Yes, anywhere

Technologies: Python, Rust, C, Bash, JS, Git, Linux, SQL, Flask, React, PyTorch, TensorFlow, Numpy, Pandas

Résumé/CV: http://anishthite.github.io/Anish_Thite_Resume.pdf

Email: anishthite@gmail.com

BS CS student graduating in Spring 2021, open to internships or full-time offers. I have worked on some ML and Full-stack projects, and am interested in applying NLP techniques to interesting problems (open to exploring other fields though). I'm interested in startups and small to mid sized companies. Ideally I want to have high impact within the company and high impact on the world.

Attention is all you need

Title should be renamed to "Language Models are Lossless Compressors"

Could you elaborate a bit more? What were the differences between transformer adversarial examples and cnn adversarial examples?

What exactly is the difference between pattern matching and reasoning?

Read the Elon Musk biography by Ashlee Vance. It describes the early days of spaceX. The short answer is incredibly hard effort.

How can new companies get involved in the space industry? It seems that the largest problem right now is rocketry but that has a high initial cost

I would guess so, but compressing it by 1/3rd it's size (ie. distilgpt) would still be quite large. To be fair, I don't know if distillation scales like that.

How much does that cost?