gdiamos

17 ms·

241.
by gdiamos 2y ago
Even then, why is Intel better grout than AMD today?
242.
by gdiamos 2y ago
We did some user studies and found that people found it less intuitive.
243.
by gdiamos 2y ago
Yeah! - outlines, guidance, jsonformer were inspiring for this line of work
244.
by gdiamos 2y ago
We've had this for over 1 year in Lamini - https://lamini-ai.github.io/inference/json_output/ Works with any open LLM, including Llama 3.1
245.
by gdiamos 2y ago
What? That's incredible. They played in the same league as male players? Clearly I'm missing a history lesson here.
246.
by gdiamos 2y ago
You totally ignored the Lamini JSON output mode - which is full speed and supports enums for classifiers https://lamini-ai.github.io/inference/json_output/
247.
by gdiamos 2y ago
That’s fair - I’ll try to go through the weekend and write out some of the equations for the kernel that loads the weights out of the index and does the adaptor ops. It’s inspired by cross attention in retro but there are some differences f
248.
by gdiamos 2y ago
That is a bachelor’s thesis - someone should hire that student Clear writing and ideas
249.
by gdiamos 2y ago
It’s closed source in one company unless every other company does something about it.
250.
by gdiamos 2y ago
Founders/CEOs don't lose track of equity or contracts around it. Every 1/10 of a percent is tracked and debated with investors.
251.
by gdiamos 2y ago
Why is AI so dramatic? I just watched mean girls and this is worse.
252.
by gdiamos 2y ago
It looks like Anthropic is now leading the charge on safety
253.
by gdiamos 2y ago
Agree that the current generation cannot do this. We have some better ones...
254.
by gdiamos 2y ago
That's not exactly the same thing as the LLM photographic memory that we launched at Lamini a month ago - https://www.lamini.ai/blog/lamini-llm-photographic-memory-ev... We focused on LLMs that can reason while st
255.
by gdiamos 2y ago
Does Meta plan to open source these models?
256.
by gdiamos 2y ago
Does it count as a public disclosure if you disclose an invention to a slack AI? The Wikipedia definition says: “a communication to one or more members of the public”. https://en.m.wikipedia.org/wiki/Public_disclosure
257.
by gdiamos 2y ago
What kind of programs would you write in bend? Would you use it instead of CUDA kernels or openmp pragmas? What about weird stuff in python like objects, duck typing, side effects, io/print statements, etc
258.
by gdiamos 2y ago
I don’t think people give Dario enough credit
259.
by gdiamos 2y ago
Ironically built by Ilya
260.
by gdiamos 2y ago
that sounds really sad
261.
by gdiamos 2y ago
I'm surprised no one has done this for a GPU cluster yet - we used to do this for RNNs on GPUs & FPGAs at Baidu: https://proceedings.mlr.press/v48/diamos16.pdf Or better yet - on Cerebras Kudos to groq for wri
262.
by gdiamos 2y ago
I feel like we screwed up the AI chip segmentation with separate “inference” and “training” chips Every machine learning application ever involves both training and inference. What is the point of a model that can’t learn? What is the point
263.
by gdiamos 3y ago
CUDA was always harder to program - even if you could get better perf It took a long time to find something that really took advantage of it, but we did eventually. CUDA enabled deep learning which enabled LLMs . That's history. What s
264.
by gdiamos 3y ago
As someone who worked on CUDA 15 years ago - it’s amazing to me that someone on the internet posted this statement. Did GPUs win?
265.
by gdiamos 3y ago
rare words are out of vocab errors in vectors Especially if they aren’t in the token vocab
266.
by gdiamos 3y ago
We do this for the null hypothesis - is uses an LLM to bootstrap a binary classifier - which handles null easily https://github.com/lamini-ai/llm-classifier
267.
by gdiamos 3y ago
Here’s a pipeline for earnings calls we built https://github.com/lamini-ai/lamini-earnings-calls We ran it on 1 million LLM queries and posted the results here. https://huggingface.co/datasets/lami
268.
by gdiamos 3y ago
I wasn't happy when one of my LLMs messed up an equity calculation. Startups highly value equity and a missing 1/10th of a percent would be worth fighting over. Once we started seeing issues like this in several enterprise apps, w
269.
LLMs hallucinate in critical enterprise scenarios(lamini.ai)
6 points by gdiamos 3y ago|1 comments
270.
by gdiamos 3y ago
Show the proof? Does it include IFT?

More ›