"How do you know about all this AI stuff?"
I just read tweets, buddy.
Page 1 of 1
I guess it's Prodigy but at some sort of scale. Or LabelStudio but every single plan demands you to contact them for pricing.
Except Hugging Face says it's "a free interface for validating and cleaning unstructured LLM outputs" so maybe it's just the hosted one that costs [a lot of] money. Could I explore it? Yes! Have I done it? No!
Source: https://argilla.io/
This is good in combination with Hugging Face's Synthetic data: save money, time and carbon with open source.
This post does a fantastic job breaking down how you use an expert labeler (teacher LLM) to annotate your data, then use it to fine-tune a student LLM. It's as good or better than crowd workers!
In this case they use Mixtral to prep data for RoBERTa-base, then get equal performance in the end. So much faster! So much cheaper!
OpenAI has continually claimed that the "model weights haven't changed" on their models over time, which many have accepted as "the outputs shouldn't be changing." Even if the former is true, something else is definitely happening behind the scenes:
For example, GPT-4's success rate on "is this number prime? think step by step" fell from 97.6% to 2.4% from March to June, while GPT-3.5 improved. Behavior on sensitive inputs also changed. Other tasks changed less, but there are definitely singificant changes in LLM behavior.
Is is feedback for alignment? Is it reducing costs through other architecture changes? It's a mystery!
Another fun pull quote, for code generation:
For GPT-4, the percentage of generations that are directly executable dropped from 52.0% in March to 10.0% in June. The drop was also large for GPT-3.5 (from 22.0% to 2.0%).
If you're building a product on top of a model you aren't running yourself, these sorts of (unreported) changes can wreak havoc on your operations. Even if your initial test runs worked great, two months down the line and you might have everything unexpectedly fall apart.
Meta has officially released LLaMA 2, a new model that's easily useable on our dear friend Hugging Face (here's a random space with it as a chatbot). The most important change compared to the first iteration is that commercial usage is explicitly allowed. Back when the original LLaMA was leaked, trying to use it to make sweet sweet dollars was a bit of a legal no-no.
In addition, this tweet from @younes gives you a script to fine-tune it using QLoRA, which apparently allows babies without infinite resources to wield these tools:
Leveraging 4bit, you can even fine-tune the largest model (70B) in a single A100 80GB GPU card!
Get at it, I guess?
Looks great!
Multilingual AI is a vey real issue, with literal lives on the line. Mostly because Facebook wants to use AI to moderate hate speech instead of using actual human beings (although that has problems, too). Ignoring content moderation on social media in non-English countries goes much worse than you'd imagine.
Lots of ways to contribute, from the Aya site:
Not that I know the details, but I have my doubts that BloombergGPT was even worth it. I think "maybe look at" is a little too gentle – if you think you need your own model, you don't.
Prompt engineering and even somewhat thoughtful engineering of a pipeline should take care of most of your use cases, with fine-tuning filling in any gaps. The only reason you'd train from scratch is if you're worried about the copyright/legal/ethical implications of the data LLMs were trained on – and if you're worried about that, I doubt you have enough data to build a model.
