--
agi, agi, agi, agi, agi, agi, agi
These are some of the words that end in agi. You can also use the word agi in a sentence. For example, "I am going to the grocery store to get some agi."
These are some of words that end in agi.
These are some words that end in agi.
maximize, maximize, maximize, maximize, maximize, maximize, maximize, maximize
These are some words that ends in agi
--
So I think this needs more work to get to "as good as ChatGPT". But having said that, congrats on the landing
I haven't heard anyone describe the phenomenon clearly, but I expect it is a challenge with reasoning over both intent of the prompt and specific token IDs.
Having said that, here are the words ChatGPT gave me for the same prompt:
Magi Nagi Sagi Yagi Adagi Galagi Tegagi Sigikagi Tagi Wagagi
It missed Unagi, surprisingly. But it is still leagues ahead of the response primordialsoup got from Lamini.
There are 7 instances of the letter 'e' in the sentence: "Try asking chatGPT to count how many e's are in a sentence."
another one:
The words with the letter 'e' from the sentence "Try asking chatGPT to count how many e's are in a sentence" are:
asking
sentence
Here are some English words containing three instances of the letter 'e':
Nevertheless
Extreme
Relevance
Precedence
Residence
Easement
Demeanor
What surprising here is that it's still capable of writing hundred lines of python code..
"what is pistacchio? explain the question, not the answer."
all these toy llm: "pistacchio is..."
gpt is the only one that consistently understand these instructions: "The question "what is pistachio?" is asking for an explanation or description of the food item..."
this makes these llm basically useless for obtaining anything but hallucinated data.
This is a bit like complaining that your compiler refuses to produce the right outputs for code you've already determined is incorrect.
---GPT-3.5---
Here are some words that end in "agi":
Strategy
Swarajya
Arthroplasty
Sialagogue
Podagric
Gynecology
Physiognomy
Ophthalmology
Esophagitis
Otalgia
--- GPT-4 ---
Here are some words that end in "agi":
Swaggy
Raggi
Magi
Gagi
Stagi
Please note that some of these words may not be commonly used or may be specific to certain dialects or regions.
I’m super excited to announce Lamini, the LLM engine that gives every developer the superpowers that took the world from GPT-3 to ChatGPT!
I’ve seen a lot of developers get stuck after prompt-tuning for a couple days or after fine-tuning an LLM and it just gets worse—there’s no good way to debug it. I have a PhD in AI from Stanford, and don’t think anyone should need one to build an LLM as good as ChatGPT. A world full of LLMs as different & diverse as people would be even more creative, productive, and inspiring.
That’s why I’m building Lamini, the LLM engine for developers to rapidly customize models from amazing foundation models from a ton of institutions: OpenAI, EleutherAI, Cerebras, Databricks, HuggingFace, Meta, and more.
Here’s our blog announcing us and a few special open-source features! https://lamini.ai/blog/introducing-lamini
Here’s what Lamini does for you: Your LLM outperforms general-purpose models on your specific use case You own the model, weights and all, not us (if foundation model allows it, of course!) Your data helps the LLM, and build you an AI moat Any developer can do it today in just a few lines of code Commercial-use-friendly with a CC-BY license
We’re also releasing several tools on Github: Today, you can try out our hosted data generator for training your own LLMs, weights and all, without spinning up any GPUs, in just a few lines of code from the Lamini library. https://github.com/lamini-ai/lamini/
You can play with an open-source LLM, trained on generated data using Lamini. https://huggingface.co/spaces/lamini/instruct-playground
Sign up for early access to the training module that took the generated data and trained it into this LLM, including enterprise features like virtual private cloud (VPC) deployments. https://lamini.ai/contact
So much click bait in the LLM space.
Introducing Lamini, the LLM Engine for Rapidly Customizing Models
Obviously it still takes a huge amount of work to customize a model to be as good as GPT4 or ChatGPT, that’s exactly why we are building Lamini.
To give developers tools to make it easier.
Hopefully it is clear that it will take more work than 1 day.
I don't really care to click on something I know is obviously lying to me.
Check out https://github.com/huggingface/peft -- they've packaged it up nicely- and read up on LoRA (https://arxiv.org/pdf/2106.09685.pdf) That should get you started.
What's the difference between this and other data pipelines like Alpaca?
This is a question you should instead ask in a user interview, or at a minimum qualify when asking here as one of the people involved in the project.
On the other hand, no amount of prompt engineering seems to make these LLM able to do question and answer over source documents which is the only realistic way by which factual information can be retrieved
You're welcome to bring examples of it tho if you're so confident.
You're welcome to demonstrate the same ability on other models tho.
It was never specified "exactly 3".
But my point was that ChatGPT, like any tokenized LLM, doesn't even have the concept of letters. The prompt "how many e's in this sentence" is rendered as the tokens [4919, 867, 304, 338, 287, 428, 6827]. There just isn't a pathway for it to consider the letters that make up those tokens.
I'm a little surprised it did that well on your prompt, which is rendered as [10919, 2456, 886, 287, 556, 72]. The interesting thing here is that 556 = " ag" (with leading space) and 72 = "i". So I'm not sure how got to those words. "Wagagi" is tokens [54, 363, 18013], so somehow it is seeing that token 18013 is what you get when you combine 556 and 72? That seems really weird.
I'd love clarification from someone deeper into LLMs and tokenization.
Intuitively I think there's an abstraction barrier there, but I'm not positive. It feels like asking us to list all of the words that trigger particular neurons.
"6769545085823578960 + 16027170449476717488"
ChatGPT said the answer is 22796715535300296448. It got the correct answer even though the problem wasn't in its training data.
This doesn't mean it's always correct, or can be trusted without verification.
https://www.google.com/search?q=stagi
There are a lot of genuine hits for stagi.
It belongs in the list just fine.
The strength of LLM is their ability to answer to unprecisely specified questions, being able to guess the speaker's intent, but in this particular case, it's failing the test.
And I mentioned ChatGPT because from context of your comments here it was unclear on first read-through what you meant. Maybe consider that it's possible your prompting is not geared for the models you've tried.
Not least, specifically given that if you expect a model to know how to follow instructions, when most of them have not been through RLHF you're using them wrong. A lot of them needs prompt shaped as a completion, not a conversation.