⛰️ Amazon’s massive LLM
Plus details of the AI Pin leak
Login
Amazon is investing heavily in an LLM that could one day take on OpenAI’s ChatGPT and Google’s **Bard, according to **Reuters sources.
The LLM, which goes by the codename Olympus, reportedly has 2 trillion parameters, which would make it one of the largest around. For comparison, GPT-4 is said to have approximately 1.7 trillion parameters, while Bard sits at roughly 1.6 trillion.
Reuters says the project is being overseen by Rohit Prasad, who was previously in charge of Alexa.
**Why it matters: **It’s of note that Amazon has already trained its Titan LLM (reportedly 45 billion parameters), and has committed a cool $4 billion to its partnership with Anthropic. But if the reporting about Olympus is true, it could cast a long shadow over those endeavors, and perhaps even the wider AI industry.
The hype about Humane’s AI Pin (sometimes styled Ai Pin) has been frothing for a while, but it seems some major details have leaked on the eve of its official launch.
According to documents seen by The Verge, the screen-less device will cost $699 and require a $24-a-month cellular plan that runs on a Humane-branded MVNO.
In terms of functionality, AI Pin users will interact with the device using their voice, with AI smarts provided by Microsoft and OpenAI. The report notes users will have to physically touch the AI Pin in order to activate it.
Specs include a built-in speaker, a small projector (for displaying content on a hand), and a battery pack that is also a magnet, allowing the device to be connected to a user’s clothing.
Humane is set to fully introduce the AI Pin today, with details of availability also expected.
**Why it matters: **Some have spoken about the AI Pin as a smartphone alternative, while others have been more skeptical. Now the wait is over, we get to see whether there’s a real appetite for a standalone AI-powered device.
**YouTube **is testing a conversational AI feature that lets users ask questions about the video they are viewing.
Dubbed YouChat, the video platform has made the tool available to select users, allowing them to “dive deeper into the content they are watching.” YouTube notes it will soon roll the feature out to Premium customers using Android devices.
YouTube is also piloting an AI tool that can summarize comment sections. As the Google-owned** **company explains, the tool could help creators “learn what their subscribers are talking about at a glance.”
It’s unclear if or when the AI tools will be accessible to more users.
**Why it matters: **For YouTube creators and viewers, it could be interesting to see if these AI features ever make it to prime-time; an AI chatbot to assist with how-to tutorials makes a lot of sense and video-makers will surely appreciate getting the pulse on viewer feedback.
With AI everywhere, it seems everyone (including this newsletter) is talking about artificial general intelligence (AGI). But researchers at Google are suggesting we don’t have to worry about machines taking over any time soon.
According to a pre-publication research paper, a team of Google AI experts say transformers (the T in GPT) are not good at generalizing. The authors claim transformers struggle to perform “simple extrapolation tasks” when asked to take on something that they’ve not been previously trained on. In short, they can be experts on what they have learned, but essentially useless when it comes to anything else.
Of course, this is all counter to what others are saying, with some even claiming AGI has already arrived.
Why it matters: Google’s commentary on the effectiveness of transformers carries a lot of weight - particularly as they came up with the technology. But with so many concerned about the AGI and its potential for disruption, expect the debate to rumble on.
