Need to read this more closely. A few things stand out at first glance:
This is an addition to the geo-political stances that EU, US, China put forth w.r.t. everything digital and data. A comparison with the EU AI Regulation that is under negotiation is of interest.
It seems focused on generative AI solely. Are there other (planned) acts covering other AI applications and development. Why is generative AI singled out here, because it has a more direct population facing character?
It seems to mostly front-load the responsibilities towards the companies producing generative AI applications, i.e. towards the models used and pre-release. In comparison the EU regulations incorporates responsibilities for distributors, buyers, users and even users of output only and spans the full lifetime of any application.
It lists specific risks in several categories. How specific are those worded, might there be an impact on how future-proof the regulation is? Are there thresholds introduced for such risks?
Let’s see if I can put some AI to work to translate the original Chinese proposed text (PDF).
By emphasizing corpus safety, model security, and rigorous assessment, the regulation intends to ensure that the rise of [generative] AI in China is both innovative and secure — all while upholding its socialist principles.
Author Steven Johnson has been working with Google and developed a prototype for Tailwind. Tailwind, an ‘AI first notebook’, is intended to bring an LLM to your own source material, and then you can use it to ask questions of the sources you give it. You point it to a set of resources in your Google Drive and what Tailwind generates will be based just on those resources. It shows you the specific source of the things it generates as well. Johnson explicitly places it in the Tools for Thought category. You can join a waiting list if you’re in the USA, and a beta should be available in the summer. Is the USA limit intended to reduce the number of applicants I wonder, or a sign that they’re still figuring things like GDPR for this tool? Tailwind is prototyped on PaLM API though, which is now generally available.
This, from its description, gets to where it becomes much more interesting to use LLM and GPT tools. A localised (not local though, it lives in your Google footprint) tool, where the user defines the corpus of sources used, and traceable results. As the quote below suggests a personal research assistant. Not just for my entire corpus of notes as I describe in that linked blogpost, but also on a subset of notes for a single topic or project. I think there will be more tools like these coming in the next months, some of which likely will be truly local and personal.
On the Tailwind team we’ve been referring to our general approach as source-grounded AI. Tailwind allows you to define a set of documents as trusted sources …, shaping all of the model’s interactions with you. … other types of sources as well, such as your research materials for a book or blog post. The idea here is to craft a role for the LLM that is … something closer to an efficient research assistant, helping you explore the information that matters most to you.
LLMs usually require loads of training data, the bigger the better. This biases such training, as Maggie Appleton also pointed out, to western and English dominated resources. This paper describes creating a model for a group of 11 African languages that are underresourced online, and as a result don’t figure significantly in the large models going around (4 of the 11 have never been included in a LLM before). All the material is available on GitHub. They conclude that training a LLM with such lower resourced languages with the larger ones is less effective than taking a grouping of underresourced languages together. Less than 1GB of text can provide a competitive model! That sounds highly interesting for the stated reason: it allows models to be created for underresourced languages at relatively little effort. I think that is a fantastic purpose because it may assist in keeping a wide variety of languages more relevant and bucking the trend towards cultural centralisation (look at me writing here in English for a case in point). It also makes me wonder about a different group of use cases: where you have texts in a language that is well enough represented in the mainstream LLMs, but where the corpus you are specifically or only interested in is much smaller, below that 1GB threshold. For instance all your own written output over the course of your life, or for certain specific civic tech applications.
We show that it is possible to train competitive multilingual language models on less than 1 GB of text. .our model … is very competitive overall. … Results suggest that our “small data” approach based on similar languages may sometimes work better than joint training on large datasets with high-resource languages.
I think it is a bit of a ‘well-duh’ thing but worth underlining in general conversation still. The name Large Language Model is somewhat misleading and a misnomer as it does not contain a model of how (a) language (theoritically) works. It e.g. doesn’t generate texts by following grammar rules. How LLMs can generate code from natural language prompts because they have been trained with sofware code without the theoretical underpinnings of programming languages leads to this by extension. Veres suggests using the term of Large Corpus Models. I think getting people to write LCMs and not LLMs will be impossible. I can however for myself highlight the difference by reading ‘Large Language usage Model’ everytime I see LLM. As the Corpus is one of language(s) in actual use.
We argue that the term language model is misleading because deep learning models are not theoretical models of language and propose the adoption of corpus model instead, which better reflects the genesis and contents of the model.
It seems after years of trollbots and content farms, with generative algorithms we are more rapidly moving past the point where the basic assumption on the web still can be that an (anonymous) author is human until it becomes clear it’s otherwise. Improving our crap detection skills from now on means a different default:
As many others I am fascinated by what generative algorithms like ChatGPT for texts and Stable Diffusion for images can do. Particularly I find it fascinating to explore what it might do if embedded in my own workflows, or how it might change my workflows. So the link above showing an integration of ChatGPT in WordPress’ Gutenberg block editor drew my attention.
The accompanying video shows a mix of two features. First having ChatGPT generate some text, or actually a table with specific data, and having ChatGPT in ‘co-pilot’ style generate code for Gutenberg blocks. I think the latter might be actually useful, as I’ve seen generative AI put to good use in that area. The former, having ChatGPT write part of your posting is clearly not advisable. And the video shows it too, although the authors don’t point it out or haven’t reflected on the fact that ChatGPT is not a search engine but geared to coming up with plausible stuff without being aware of its actual information (the contrast with generating code is that code is much more highly structured in itself so probabilities collapse easier to the same outcome).
The blogpost in the video is made by generating a list of lunar missions, and then turning them into a table, adding their budgets and sorting them chronologically. This looks very cool in the vid, but some things jump out as not ok. Results jump around the table for instance: Apollo 13 moves from 1970 to 2013 and changes budget. See image below. None of the listed budgets for Apollo missions, nor their total, match up with the detailed costs overview of Apollo missions (GoogleDocs spreadsheet). The budget column being imaginary and the table rows jumping around makes the result entirely unfit for usage of course. It also isn’t a useful prompt: needing to fact check every table field is likely more effort and less motivating than researching the table yourself from actual online resources directly.
It looks incredibly cool ‘see me writing a blogpost by merely typing in my wishes, and the work being done instantly’, and there are definitely times I’d wish that to be possible. To translate a mere idea or thought into some output directly however means I’d skip confronting such an idea with reality, with counter arguments etc. Most of my ideas only look cool inside my head, and need serious change to be sensibly made manifest in the world outside my head. This video is a bit like that, an idea that looks cool in one’s head but is great rubbish in practice. ChatGPT is hallucinating factoids and can’t be trusted to create your output. Using it in the context of discovery (as opposed to the justification context of your output such as in this video) is possible and potentially useful. However this integration within the Gutenberg writing back-end of WordPress puts you in the output context directly so it leads you to believe the generated plausible rubbish is output and not just prompting fodder for your writing. ‘Human made’ is misleading you with this video, and I wouldn’t be surprised if they’re misleading themselves as well. A bit like staging the ‘saw someone in half and put them together again’ magician’s trick in an operating room and inviting surgeons to re-imagine their work.
Taking a native-first approach to integrating generative AI into WordPress, we’ve been experimenting with approaches to a “WordPress Copilot” that can “speak” Gutenberg / block-editor.
Copy-pasting paragraphs between ChatGPT and WordPress only goes so far, while having the tools directly embedded in the editor … open up a world of possibilities and productivity wins…
Joe Hoyle
An android robot is filling out a table listing Apollo missions on a whiteboard, generated image using Midjourney
