CONVO OVER COFFEE ☕

Big topics to discuss with friends and colleagues

AI Dubbing is a space of new startups 🚩
Not only ElevenLabs unveils AI Dubbing, translating Video and Audio Content into 20 Languages, but also Captions introduces Lipdub - an AI Dubbing App oriented towards Gen Z.
My take: This is an interesting space, where clearly many new startups made a lot of waves. Spotify with their actions from last weeks heralded the way onwards - adapting the content generated in video/audio form into different languages. Here we see these models popping up in new spaces and being easily available for end users. The danger comes in form of manipulating messages addressing one population to another. Politicians could be the first victims. But effects for movie and podcast space? Groundbreaking.

Are developer AI tools entering a fight for survival phase? 🟠
We all know about magical properties of ChatGPT dropping code left and right. Although seemingly miraculous, ChatGPT is not the best coding buddy you can hire. Companies specialized in these spaces are enticing their users with AI Assistants promising help solely in writing good code. The issue? How to make money. GitHub, the king of storing code repositories, released Copilot a while back. It generated waves, with 1.5mln users taking advantage of this tool. But new report from Wall Street Journal suggests that GitHub looses money for running it, instead of bringing more revenue. Which may justify their recent blogposts claming how good Copilot is.
My take: Replit, a developer tooling vendor, deployed their own AI code companion to all users. Which only justifies this as being a trend going onwards. The issue remains with cost associated with running these models - while GitHub has a backing of Microsoft, Replit partners with Google for infrastructure access. Why are these organizations getting so closely linked with cloud giants? Because deploying models like code assistance at scale now requires ridiculously high computing cost and only those giants are able to deal with it now. Go around and take a look on the job market - plenty of roles at Amazon are Data Center related. You will see many similar positions at Microsoft and Google. Cloud giants know that the more tools like code assistance appear, the more their infrastructure will be needed. The question that we should all have is: will it save the innovation wave, or dwarf it? Ultimately, while companies can power on and develop less computational hungry models, they may found themselves too close to giants, like Icarus flying too closely to the sun.

‘Funny’ chatbots time 🚩🚩
After recent Meta’s announcement of creating avatars allowing people to chat with their favorite known individuals, character.ai jumps on the same bandwagon allowing group chats in their communications with avatars of famous and imaginary characters.
My take: while both seemingly compete in entertainment chat experiences, the deeper danger of these changes comes from threat to the privacy of public figures. These avatars not only frequently fails to acknowledge they are just avatars, but also can go into unexpected spaces talking about sensitive things without proper attention, potentially damaging stars and influencer’s reputations. I won’t even start on possibility of having some crazy fanatics popping up and claiming that they can assault a movie star in real life because they got ‘intimate on the chatbot’ - I hope you see the danger here as I am, with some people potentially mixing artificial world with realities to the dangerous extent.

AI Chips for everyone!🟠
AMD made waves by acquiring open-source AI software company - Nod.ai - to go deeper in adapting software for HPC (high performance computing). But that is just a tip of an iceberg. Microsoft reportedly works on its own AI Chip. OpenAI thinks of one as well. DeepMind has been using AI for a while now to optimize their chips for the best possible quality with training their models. Now, why is this happening?
My take: There are two reasons, in my opinion, that drives this push: you can not only optimize the hardware for better performance against your models (and deal with extravagant computational costs that way), but you can also deal with chip shortage that influence companies potential to develop and train new models. For now Nvidia is very happy with where things are going. But as much as noone expected Apple to develop their own M1 and M2 chips and break the partnership with Intel, now Nvidia may not be seeing the wave of ‘let me do it myself’ companies popping around. And there are many arriving. Furiosa.ai is one, charming example - with their processing unit called Warboy (witness me!). But if you would go around and check what giants are hiring for, you would see Google desperately looking for GPU PMs right now. The side effects? Potentially more space for operating systems and open-source projects.

In other mainstream news:

OpenAI goes after big bucks, with revenue at $1.3 Billion 🚩

Google can generate images in search, and search using images 🚩

Adobe goes deep into AI video manipulation to help you out in erasing what you don’t like from your video 🚩🚩