ggnore7452

4 ms·

1.
by ggnore7452 5mo ago
Isn't that the memory files are loading by default?
2.
by ggnore7452 6mo ago
too bad that only the smaller on-device models support native audio input.
3.
by ggnore7452 8mo ago
imo, for health related stuff. or most of the general knowledge doesn't require latest info after 2023. the internal knowledge of LLM is so much better than the web search augmented one.
4.
by ggnore7452 9mo ago
I’m fine with hires without degrees. But if Google still filters people with LeetCode style coding questions, what’s the point of that in this day and age?
5.
by ggnore7452 1y ago
so immersive i actually hit ctrl+w and closed the whole tab.
6.
by ggnore7452 1y ago
I’ve done a similar PDF → Markdown workflow. For each page: - Extract text as usual. - Capture the whole page as an image (~200 DPI). - Optionally extract images/graphs within the page and include them in the same LLM call. - Optionall
7.
by ggnore7452 1y ago
I’ve been using LLMs for this kind of geo-guessing since Gemini 2.0. Even without access to internet search like o3, they perform surprisingly well.
8.
by ggnore7452 2y ago
appreciate the question on hparams for websearch! one of the main reasons i build these ai search tools from scratch is that i can fully control the depth and breadth (and also customize loader to whatever data/sites). and currently t
9.
by ggnore7452 2y ago
How’s this compare to likes of Fish audio? Wish they support voice clone using longer audio tho . Haven’t looked into this space for few months , but iirc, previously SOTA was like GPT VITS or something ?
10.
by ggnore7452 2y ago
anyone tried this? is this actually overall better than xgboost/catboost?
11.
by ggnore7452 2y ago
if anything i would consider embeddings bit overrated, or it is safer to underrate them. They're not the silver bullet many initially hoped for, they're not a complete replacement for simpler methods like BM25. They only have very
12.
by ggnore7452 2y ago
What’s more interesting to me here are the calibration graphs: • LLMs, at least GPT models, tend to overstate their confidence. • A frequency-based approach appears to achieve calibration closer to the ideal. This kinda passes my vibe test
13.
by ggnore7452 2y ago
Side note: I feel like ChatGPT's long-term memory isn't implemented properly. If you check the 'saved memories,' they are just bad.
14.
by ggnore7452 2y ago
question: any good on-device size image embedding models? tried https://github.com/unum-cloud/uform which i do like, especially they also support languages other than English. Any recommendations on other alternatives?
15.
by ggnore7452 3y ago
I've been working on a small personal project similar to this and agree that replicating the overall experience provided by Perplexity.ai, or even improving it for personal use, isn't that challenging. (The concerns of scale or c
16.
by ggnore7452 3y ago
more on the LPU and data center: https://wow.groq.com/lpu-inference-engine/ price and speed benchmark: https://wow.groq.com/
17.
by ggnore7452 3y ago
The Groq demo was indeed impressive. I work with LLM alot in work, and a generation speed of 500+ tokens/s would definitely change how we use these products. (Especially considering it's an early-stage product) But the "compl
18.
by ggnore7452 4y ago
it still works well with other LLM (like llama and more). various small, open sourced, and verticle LLMs vs one large GPT models would be quite interesting.
19.
by ggnore7452 4y ago
The place where I work was an early adopter of LLM, having started working on it a year ago. When I build stuff with GPT-3, especially in the earlier days, I get the strong impression that it's like we are doing machine learning withou
20.
by ggnore7452 4y ago
there is actually a paper by OpenAI themselves on summarizing long document. essentially, break a longer text into smaller chunks, and run a multi-stage sequential summarization. each chunk uses a trailing window of previous chunk as conte
21.
by ggnore7452 4y ago
I use large language model for work and use Notion daily. while I like the "AI" part (the large language model), think it would be more interesting and productive to use same backend for full text semantic search & question a
22.
by ggnore7452 5y ago
AlphaStar is really amazing achievement. but this is not like AlphaZero for Go, which completely made people rethink how Go should be played. AlphaStar didn't "solved" SC2. It can still fall to very simple early strategy, an
23.
by ggnore7452 6y ago
also, for some quick and simple Q&A system. Haystack https://github.com/deepset-ai/haystack (essentially dense vector similarity on Elastic Search) looks pretty promising and supports whole pipeline.
24.
by ggnore7452 6y ago
kinda interesting see people use "authoritarian" as if it's all the reasons China succeed (in term of infra, econ, and general well-being), and therefore nothing to see and learn. If authoritarian government is by default thi