jerpint

24 ms·

271.
by jerpint 3y ago
Gradio is popular for machine learning but is versatile enough for interactive web apps, really easy to tinker with
272.
by jerpint 3y ago
I would have loved to see videos on the blog post of completions
273.
by jerpint 3y ago
GATO was heavily biased towards tasks in a simple simulator, but didn’t exhibit emergent behaviours
274.
by jerpint 3y ago
Using memorization is something that can be a feature in some cases, especially in reducing hallucinations. Perhaps instead of embedding retrieval you’d condition a model to only repeat memorized relevant passages, something that can be tra
275.
by jerpint 3y ago
This is very likely common practice by many other governments
276.
by jerpint 3y ago
A real test would be fine tuning gpt4 and comparing that to gpt4 medprompt
277.
by jerpint 3y ago
gpt4 is capable of reasoning “in distribution”. Its reasoning drops when you go outside the goldilocks zone
278.
by jerpint 3y ago
It’s a lot cheaper to replicate than to innovate
279.
by jerpint 3y ago
Allow me to put my conspiracy hat on: Microsoft has an open “embrace, extend, extinguish” policy since forever. ChatGPT integration into Microsoft has been a huge win for them. Maybe they cleverly figured out a way to guarantee openAI would
280.
by jerpint 3y ago
I doubt this to be true openAI does not allow you to download models (with few exceptions)
281.
by jerpint 3y ago
Yes mostly python
282.
by jerpint 3y ago
Not sure if it’s related or not but I noticed the UX has changed on the web interface and when I ask for code, it now defaults to running it through an interpreter with a hidden view by default. This is very annoying to me as I want to most
283.
by jerpint 3y ago
I had to implement a relatively simple auth flow recently for an app (something I haven’t done in the past). Gpt3.5 struggled to give me anything useful other than high level ideas but GPT4 gave me the exact boilerplate I needed and unblock
284.
Show HN: RAGTheDocs, one-click deploy RAG for any readthedocs website(github.com)
1 points by jerpint 3y ago|0 comments
285.
by jerpint 3y ago
The weights are the inference and result of training. I can give you all the training details and you might not be able to reproduce what I did (google does this all the time). As a dev, I’d much rather an open model over an open recipe wit
286.
by jerpint 3y ago
It definitely is open source even if they don’t disclose all details behind the training
287.
by jerpint 3y ago
There’s something odd about their Track 1 results: Flan-base gets an alright score, but Flan-large gets 0.0. That seems very fishy as it sounds this is pretty much a multiple choice kind of task. I would not expect a score less than 1/
288.
by jerpint 3y ago
I made my own animations once upon a time using manim, not as shiny but might be helpful too https://www.jerpint.io/blog/cnn-cheatsheet/
289.
MNIST Clock(jerpint.io)
2 points by jerpint 3y ago|1 comments
290.
by jerpint 3y ago
Each clock digit is generated by a VAE trained on MNIST. Every transition between numbers traverses the latent space of the decoder, interpolating between the digits.
291.
by jerpint 3y ago
This was my reaction too. How do you “prove” my gradient is valid? Well perhaps one way is you could have another LLM take a look at the data you are submitting and have it predict p(useful|not useful) , and create an incentive for users to
292.
by jerpint 3y ago
I believe the term for this is enshitification
293.
by jerpint 3y ago
The symmetric keyboard seems “as obvious” as using tau instead of pi as a universal constant, but then again I don’t play piano
294.
by jerpint 3y ago
I’d be very interested in a similar comparison for RAG style tasks
295.
by jerpint 3y ago
How “secure” is this zero-trust? Are there cryptographic guarantees?
296.
by jerpint 3y ago
Also reduces your cognitive load when you just want to answer as quickly as possible without thinking about e.g. “am I being too rude/dry etc”?
297.
by jerpint 3y ago
If this holds true, this would support the idea that much smaller, human curated datasets will be of much higher value than synthetic datasets generated by LLMs
298.
by jerpint 3y ago
What do you do when inheriting from a base class with a defined __init__ ?
299.
by jerpint 3y ago
I added a keyboard tray to my standing desk, made a big difference for ergonomics on my wrists and shoulders
300.
by jerpint 3y ago
There have been many attempts to do multi modal pre training, the difficulty is finding the right combination of data for it to be “useful” and “scalable”. It’s not trivial to just train a transformer on video, text, audio, etc. mainly due

More ›