msp26

11 ms·

121.
by msp26 2y ago
Have you seen how long the CoT was for the example. It's incredibly verbose.
122.
by msp26 2y ago
> THERE ARE THREE R’S IN STRAWBERRY Well played
123.
by msp26 2y ago
That's unfortunate. When an LLM makes a mistake it's very helpful to read the CoT and see what went wrong (input error/instruction error/random shit)
124.
by msp26 2y ago
> presumably if you create an llms.txt you're not looking to manipulate anyone I don't think the public sentiment around scrapers and LLMs is that friendly. I personally think scraping is very important and it allows for smalle
125.
by msp26 2y ago
Isn't this already their main use case for business? We use them primarily for extracting structured data from other forms.
126.
by msp26 2y ago
Would you ask a blind person to pick between two colours for you? LLMs do not see the individual letters in the word, they work with tokens.
127.
by msp26 2y ago
There's a world of difference between a song and an entire social media platform.
128.
by msp26 2y ago
Is the JSON actually being fed into the LLM's context or is it still being converted into typescript? The previous setup didn't allow for custom types, only objects/string/num/bool. Are the enums put into context or
129.
by msp26 2y ago
You can interact with the models on their project page: https://stable-fast-3d.github.io/
130.
by msp26 2y ago
Large 2 is significantly smaller at 123B so it being comparable to llama 3 405B would be crazy.
131.
by msp26 2y ago
Classification is just too damn convenient with LLMs.
132.
by msp26 2y ago
Pretraining on the Test Set Is All You Need https://arxiv.org/abs/2309.08632
133.
by msp26 2y ago
So the underlying main LLM they're using does answer the question but it's censored by another layer afterwards when deemed 'unsafe'.
134.
by msp26 2y ago
For tasks like data extraction, are people doing full finetunes or training a LoRA? Is it any different for classification?
135.
by msp26 2y ago
The section on deduplication was very useful thanks for posting
136.
by msp26 2y ago
I don't get it either. How is the LLM meant to know the details of how the perplexity headless browser works?
137.
by msp26 2y ago
Thanks for sharing, I've followed these authors for a while and they're great. Some notes from my own experience on LLMs for NLP problems: 1) The output schema is usually more impactful than the text part of a prompt. a) Field ord
138.
by msp26 2y ago
What jailbreak prompts are you using to extract this?
139.
by msp26 2y ago
> Have you tried to use LLMs to generate structured JSON output? Not only do all LLMs suck at relaibly following a schema, you need to use all kinds of "forcing" to make sure the output is actually JSON anyway. Yeah it's
140.
by msp26 2y ago
> Dont even get me started about how function calling in other LLMs costs me tokens. Something OpenAI provides out of the box. Not sure what you mean by this.
141.
by msp26 2y ago
Petty much yeah. Agreed on historical anime/manga generally being higher quality by default. The research required tends to filter hack authors. Have you read Magus of the Library?
142.
by msp26 2y ago
I hope this means that people actually start curating their training sets. The quality control is horrible on all of the datasets I've looked through. Especially image captions. Yes sheer quantity has a quality of its own but that won&
143.
by msp26 2y ago
Trim unwanted html elements + convert to markdown. Significantly reduces token counts while retaining structure.
144.
by msp26 2y ago
In practice it's actually very good for lower volume tasks with non-fixed sources. I haven't tried this library but I do use an LLM based scraper in addition to more traditional ones.
145.
by msp26 2y ago
Hi! Unless I'm missing something, they did add the eval scripts to that repo 4 days ago.
146.
by msp26 2y ago
Can someone test this with ruler please? https://github.com/hsiehjackson/RULER In practice all of these long contexts show degraded performance (there's a table on the repo). For my NLP work I find that GPT-4-turb
147.
by msp26 2y ago
Oh I'm interested to see how your batch prompt works. I've used the idea for a while and feel that it's very underrated.
148.
by msp26 2y ago
> Is there anything I didn't really notice that feels rewarding on higher levels? The thought that you put into strategy matters more when the game is harder. Personally, I don't find it fun past ascension 17 but if I play at
149.
by msp26 2y ago
Most people are vram constrained not compute constrained.
150.
by msp26 2y ago
Oh neat, I'm working on the same thing with python and playwright. I'm finding that the latency with web LLMs is a pain in the ass and hoping to switch to llama3 after I get that set up with function calling.

More ›