Here are three things I found interesting in the world of AI over the last week.
ElevenLabs launched a text to sound effect model - blog post, app
I bet most people who use it will never know the name ElevenLabs though. Just like HeyGen built their 'translate a video' product and farm out the voice cloning and translation to ElevenLas, providers like Adobe will bundle this into their editing software, and provide a bunch of other options as they become available.
The blog post emphasizes ethically sourced training data through a partnership with Shutterstock. Just like text to speech, it looks like text to sound doesn't need an incomprehensible amount of data to generate ok quality. OpenAI also have a deal with Shutterstock and I wouldn't be surprised to see them offer a similar feature this year. Either with a specific model, or a large multimodal model which can just make sound effects as a sideline.
LlamaIndex launched a knowledge graph integration - blog post, docs
Knowledge Graph's have been around for a while and are a way of storing knowledge in structured data - e.g. [JV (person) wrote this-newsletter (link)]. They are often mentioned in the same breath as the semantic web, or linked open data. My lazy mental shorthand has been "sounds like a good idea that hasn't seen much practical adoption"
But recently I've been seeing more noise from researchers and developers about how LLMs and knowledge graphs can play friendly together. One example is using a LLM to parse unstructured text and produce a knowledge graph in the desired schema. Another is about using a knowledge graph to augment queries to LLMs (similar to how you can use a vector data store).
LlamaIndex announced support
Property Graph Indexes last week (don't worry about the details if you aren't technical) and while I'm not ready to shout "next big thing", I am allocating more time to researching the tech.
Some interesting thoughts on the impact of AI for researchers - blog post
I'm not that close to the research world and don't have strong views about the claims in the post, apart from a strong dislike of the word Singularity - even with the explicit disclaimer in the preamble.
From what I've seen in the programming and education spaces, I do think that the impacts on specific fields will be different to each other and it's fascinating to read about current and potential impacts in research.
---
I'm currently running a pilot version of an 'AI for Impact' course to help non technical people in Enspiral use AI to do more good work in the world. I've no idea if I'll run it again, but if you're interested in something along those lines reply to this email and I can send you some more info.