A week in Generative AI: Claude, Nvidia & Video
A week in Generative AI: Claude, Nvidia & Video
News for the week ending 23rd June 2024
After a relatively quiet week last week, there’s lots of great GenAI news to cover this week. We’ve had the release of Anthropic’s Claude-3.5 Sonnet, which is now the most capable GenAI model, Nvidia has become the world’s most valuable company and there area couple of interesting GenAI video tools that have been release, following up on the release of Dream Machine last week.
There’s also been lots of interesting things in AI Ethics News too, with Ilya Sutskever launching a new company focused on superalignment, Apple and Meta delaying the launch of new AI features in the EU, and Call Centers Introducing ‘Emotion Canceling’.
Enjoy!
Anthropic introduces Claude-3.5 Sonnet
This is a really impressive released from Anthropic, which will also be updating their smaller and medium size models to version 3.5 later this year. There are some great videos demonstrating Claude-3.5 Sonnet’s vision capabilities, coding, and creativity that are well worth a look.
Claude-3.5 Sonnet looks to be the most capable model available now with Anthropic’s own tests showing it outperforming all other models at various benchmarks. I’ve also done my own testing, and the new model from Anthropic definitely has the highest marketing knowledge from my benchmarks (project still a WIP!)
Nvidia becomes world’s most valuable company amid AI boom
This week, Nvidia became the world’s most valuable company, overtaking Microsoft and Apple, as its chips continues to be the go-to for the majority of companies working in AI. Nvidia is now worth $3.3tr, with its stock price increasing by 180% this year alone. It took Nvidia 24 years to get to a valuation of $1trn but only another 9 months to get to $2tr and another 3 months to get to $3tr.
Nvidia will probably continue to dominate the chip market for AI for at least another couple of years as we wait for new AI-specific chips to be developed that are designed specifically for running large language models. Even then their lead might be unassailable. They’ll probably be the first company to hit a valuation of $4tr by the end of this year.
AI the Product vs. AI the Feature
Great short video from Marques Brownlee on whether AI is a product or a feature, inspired by a WIRED podcast. I mostly agree, but with a twist…
I think right now AI is just a feature. BUT I think in the (near) future AI will enable new types of products to flourish in ways we haven’t seen yet. I wrote about this extensively back in September and I still stand by my thoughts then - we’re starting to see a lot of GenAI features but I do think that we will see GenAI-specific platforms (products) in the future. We just can’t imagine what they might be yet.
Apple delays European launch of AI features because of EU rules
This is disappointing for us Europeans, but probably to be expected. Many new AI features have been delayed in Europe from ChatGPT’s memory feature to Meta delaying the launch of the Meta AI assistant.
I worry that we’re seeing the start of AI features seriously lagging behind for consumers in the Europe. The ChatGPT memory features have been available in the US for months with no signs of being released in the EU and now Apple will delay the launch Apple Intelligence in the region. Apple Intelligence is likely to be the first time many consumers are exposed to GenAI technology on a daily basis, so it’s a big shame that Europeans won’t get to experience it for a while.
Runway introduces Gen-3 Alpha
There’s been a lot going on in the GenAI video space, spurred on by OpenAI introducing Sora earlier this year. We had the release of Luma Labs’ Dream Machine last week and this week we have Runway introducing Gen-3.
Runway have always talked about their video models being ‘world models’ as to create high quality, photo-realistic video, GenAI models have to be able to understand physics and how objects interact in the real world. It’s a really interesting way of thinking about video and why it’s such an exciting area for GenAI as it will take models’ understanding out into the real world and have big implications for robotics.
Gen-3 can create photorealistic humans and has what Runway calls ‘fine-grained temporal control’ that allows it to create videos that accurately flow from one frame to the next. From my perspective, Gen-3 looks better than Dream Machine and is closer to where OpenAI is aiming for with Sora.
Hedra Labs launches Character-1 that animates speaking characters from images
On the theme of video…. a company called Hydra Labs has just launched Character-1 that creates animated characters. You just drop in a still image, write a script, select a voice, describe the character, and that’s it. Character-1 will generate the voice, animate the image and sync the lip movements to the audio. There are some crazy examples doing the rounds on Twitter such as talking apples, a female Elon Musk, and a rapping Thanos and my favourite - potato head!
It’s great to see how much creativity these tools are unlocking, but we can’t be naive to the challenges this could create if its’ so simple to create convincing videos of people just from a still image.
AI Ethics News
Ilya Sutskever launches new company for safe superintelligent AI
‘What’s in it for us?’ journalists ask as publications sign content deals with AI firms
OpenAI expands lobbying team in battle to influence regulation
Anthropic's "Beta Steering API" offers developers a sneak peek at the future of controllable LLMs
Call Centers Introduce ‘Emotion Canceling’ AI as a ‘Mental Shield’ for Workers
Meta pauses plans to train AI using European users’ data, bowing to regulatory pressure
Not all ‘open source’ AI models are actually open: here’s a ranking
TikTok ads will now include AI avatars of creators and stock actors
Long Reads
Benedict Evans - Apple intelligence and AI maximalism
One Useful Thing - Latent Expertise: Everyone is in R&D
Aeon - What is Intelligent Life?
MIT Technology Review - What happened when 20 comedians got AI to write their routines
Bloomberg - Google DeepMind Shifts From Research Lab to AI Product Factory
“The future is already here, it’s just not evenly distributed.“
William Gibson