nestorD

13 ms·

121.
by nestorD 4y ago
Doing small projects that exercice skills I am interested in one small step at a time. These days it is a lot of LLM as it is a moving target with a large number of applications (they are the perfect toy to go quickly from "I wonder if
122.
by nestorD 4y ago
The main illustration is done in the Ligne Claire[0] comic book style made famous by Herge in Tintin (and common in French and Belgian comic book). [0]: https://en.wikipedia.org/wiki/Ligne_claire
123.
by nestorD 4y ago
I am very tempted to try writing my next paper (maybe a low stake arxiv publication) with ChatGPT. Having done the research and gotten the results, asking it to turn draft paragraphs into properly written text.
124.
by nestorD 4y ago
Yes, for example JAX is a strictly functional language, designed for high-performance computing and embedded in Python.
125.
by nestorD 4y ago
I have heard Neil Gaiman say something similar. Writing by hand is slower and you end up having to type it which pushes you toward cutting things down and streamlining the prose.
126.
by nestorD 4y ago
I do not know if t has been done but one could resuscitate the chatbot's minds by copying the chat history into a GPT based program. My own impersonator[0] is not designed for that (no persistent chat and a text based interface) but
127.
by nestorD 4y ago
I would add additional `system` messages with information fetched as a function of the user's request.
128.
by nestorD 4y ago
I want to insist on the fact that "get rich quick by picking a trendy topic and selling a bullshit book / story / video that you did not write" schemes have been sold to people for a long time. Previously those have be
129.
by nestorD 4y ago
I mean that results that would have previously been considered good enough for classical methods could now be considered improvable leading to renewed work into application of traditional methods to this subject.
130.
by nestorD 4y ago
It has but I am hoping that a paper accomplishing the task with machine learning will move the goalpost (it has happened in other areas of computer vision).
131.
by nestorD 4y ago
It does look like it would be doable without neural networks. My guess (having read their paper but not worked in the domain) is that it would be doable but requires a lot of iterations, tweaking parameters and dealing with corner cases. Th
132.
by nestorD 4y ago
"I woke up from a dream in which I was crawling in an impossible landscape. So I jumped onto my keyboard and, after a sleepless night, conjured the image I saw from the numeric ether. This is that image."
133.
by nestorD 4y ago
The other answer give you a few of the current solutions. In the long term I am hoping that JAX (/XLA) will get better support for the CPU backend of their compiler and in particular, use SIMD and multicore better than it currently doe
134.
by nestorD 4y ago
That is a very good point: having tight control on style and composition is significantly harder than getting a good-looking image with those techniques (it is doable but it takes practice to learn how to engineer prompt accordingly)
135.
by nestorD 4y ago
1 or -1 times the learning rate.
136.
by nestorD 4y ago
Take a look at langchain. It lets you interface with language models and has a number of interesting demos to get you started.
137.
by nestorD 4y ago
It is exactly what www.perplexity.ai does (and yes, it has replaced google for some of my usecases).
138.
by nestorD 4y ago
He got his wishes granted (but not the monopoly), there is an tool to call Wolfram Alpha in langchain's agents: https://colab.research.google.com/drive/1AAyEdTz-Z6ShKvewbt1...
139.
by nestorD 4y ago
Here is the aforementioned example from their documentation, answering "Who is Olivia Wilde's boyfriend? What is his current age raised to the 0.23 power?" accurately: https://langchain.readthedocs.io/en/
140.
by nestorD 4y ago
The fact checking with prompt chaining [0] repository has something along those lines (their readme has an example). It is not perfect and it is costly in number of calls to a language model but it does help a lot. [0]: https://
141.
by nestorD 4y ago
For people wanting to play with that: this is very close to langchain's agents system (their documentation has a very impressive demo using both a calculator and Google searches as tools available to a language model).
142.
by nestorD 4y ago
Retrieval-based models have a number of strengths that make them particularly important from my point of view: - as said they can be updated with fresh information on the fly, - they can give you sources for their results (meaning that you
143.
by nestorD 4y ago
> We were a bit surprised to hear that he essentially memorized a large part of a table of logarithms. But maybe he was just trolling us. I am not familiar with that particular demonstration but having studied similar things, they are al
144.
by nestorD 4y ago
You don't use them as money but they have consecutive serial numbers which has uses...
145.
by nestorD 4y ago
The scratchspace could be in natural language, preserving some debugeability and letting us know about the model mental process. This is doable but it introduce a sequential dependency which would make the training significantly slower.
146.
by nestorD 4y ago
It is a huge time saver for people doing applied math and needing quick answers to questions in order to iterate (I do have a soft spot for SymPy if you need to answer several related questions).
147.
by nestorD 4y ago
Sharing ideas led to other people playing with them and, sometimes, finding significant improvements you have not thought about. I have seen this being such a benefit that I now never keep an idea to myself. However, there is value in kee
148.
by nestorD 4y ago
I just did a quick test, it could do with slightly fuzzier searching ('Epub' as an extension got me no result while 'epub' did, a drop down menu with options might be better/simpler for some fields) but, otherwise,
149.
by nestorD 4y ago
This might be a good time to note that a large number a Sherlock Holmes stories have been written with the autorisation of the estate since the death of Conan Doyle. Of those, I particularly like Neil Gaiman's The Case of Death and Ho
150.
by nestorD 4y ago
The usual argument is that between having macro and focusing on a strong type system, there are very few legitimate usecase for reflection left in Rust.

More ›