Welcome to unfolding ai

Welcome to our new weekly newsletter! We are excited to provide you with the latest insights and information on topics that matter to creatives, knowledge workers, and business leaders.

We will be exploring the latest developments in AI and why they matter, sharing tips and strategies you can use to get ahead and augment your productivity, we will lift the lid on the tools we are using, so you can be inspired to give them a try.

We hope you find this newsletter informative and valuable, and we look forward to hearing your feedback.

Best regards,

Paul, Co-founder (and newsletter editor)

p.s. As this is a brand new newsletter, we would really appreciate it if you spread the word and did a social share. The icons are at the top of the newsletter

What’s happening in regulation and copyright

The rapid emergence of powerful generative AI has outpaced regulation, leaving creators and developers in a legal grey area. Models like chatGPT and Stable Diffusion have been trained on, raising concerns.

With no clear framework, generative AI does not fit neatly into existing copyright and fair use laws. Calls for regulation and lawsuits demonstrate growing scrutiny, but legislative ambiguity persists. As generative AI advances, regulatory clarity becomes an urgent priority to balance innovation and creators' rights.

There are a number of lawsuits around content, including a new case raised by a number of authors. However the first case was given a heavy blow when it was ruled that on the majority of its complaints, there was not clear case or connection to any form of impact.

The conversation around extinction threat from killer robots and AI monsters, seems to be to some extent moving away from the click bait titles.

The four major AI Tech firms have all announced significant steps, plans and coalitions to try and get ahead (or side step) regulation. It will be interesting to see how this continues to evolve.

Is chatGPT getting ‘dumber’

There has been a lot of conjecture, including around the unfold:ai office if chatGPT is getting ‘dumber’. A study concluded that it has been, however the study was on its mathematical performance. Perhaps not the best test for a language model. Our observations are that in day to day use, we have not seen any degradation.

We have seen an increase in guardrails, where basically it steers away from doing a task. Very notable now on anything that looks like personal data, or website that could be personal data (like LinkedIn). Some of the guardrails we have seen make sense, some absolutely do not.

There is no evidence to support the conspiracy theories around model collapse, or that humans are making it dumber by using it. OpenAI have stated that the model has not changed, but reinforcement learning is continuing to shape the outputs.

viable competition on chat LLMs ?

Anthropic AI have launched Claude 2.0, which provides viable competition to chatGPT eco system. It has a 100,000 token context window, which is 10x chatGPT. That makes a tremendous difference in long conversations and more complex data assessment. In use at unfold, its not quite the go to, but its in use. Capability wise, better than chatGPT 3.5, almost as good as chatGPT4, but not quite.

Meta (yes Facebook et al) are one of the largest AI opensource contributors, they released Llama 2 into the domain with a commercial usage license (if you have less than 700 million accounts). Its not as well trained as chatGPT4, probably on par with 3.5. However it’s in the open source domain, and available as a model already on Microsoft azure and Hugging Faces. Fine tuning and training already released, this could be great for enterprise and app development.

Stable diffusion gets better again

Stable Diffusion, which is the technology behind most text to image, and image transfers (like the ones on TikTok) have been making progress on releasing the next model, SDXL. It produces images much more ‘MidJourney’, but is more capable (if less easy to use). Clipdrop is worth a play with, it has more than SDXL, have a wander around its toolset.

Quick things to check out

A great video (we would have preferred a more positive subject) by a digital artist, created using AI tooling. When his behind the scenes is released, we will cover it. Tools used included MidJourney & Runway ML 2 a real showcase of what artists are up to.

Microsoft had the Inspire event. A lot of announcements regarding AI tools and co-pilots in the enterprise suite.

A prompt to try in chatGPT

acting as a business analyst, can you help me identify the test cases needed for a software developer to implement for the following feature

Password prompt entry should have a strength indicator

ChatGPT also added a couple of new features, they moved the share button to the top of the prompt and added the settings to pre-define information to support the prompting. Might be useful in some cases.

(here are the results of the above prompt btw!)

Did you miss these articles on our blog?

Our Events

Subscribe to Premium to read the rest.