March 8, 2024
Claude 3's Debut, Gemini 1.5, OpenAI's Visionary Text-to-Video, and Strategic Tech Alliances Reshaping Tomorrow
Anthropic Unveils Claude 3: Introducing Haiku, Sonnet, and Opus
Just a few days ago, Anthropic announced the launch of Claude 3, a new family of AI models.
Haiku, Sonnet, and Opus - 3 new AI models each offering different levels of intelligence, speed, and cost efficiency for various tasks.
With a context window of 200k tokens (and the ability to go up to 1 million tokens), Anthropic is shooting for the stars, rivaling the industry leaders.
In capabilities, the new Claude is doing very well. Some say it’s the best LLM yet, beating previous leaders like GPT-4 and Gemini on common benchmark tests.
Developers are quite pleased with its performance and pricing. It’s showing:
- better accuracy in responses;
- ability to cite sources for verification;
- exceptional memory and recall abilities.
Opus and Sonnet are currently available for use, while Haiku will be released shortly.
Claude on Cloud - Where can you get it
Claude is currently available on Anthropic, Amazon Bedrock, and Google Cloud.
Hyperscalers such as AWS provide co-funding for Generative AI projects. Should you be interested to know more about AI projects co-fundings, let us know at contact@blindspot.ai.
DeepMind Unveils Gemini 1.5 with Unprecedented Context
Google DeepMind introduced its newest AI model, Gemini 1.5. Tested against various benchmarks, Gemini 1.5 Pro outperforms its predecessor and maintains high performance even with a context window of up to 1 million tokens. This allows the model to process vast amounts of information, such as videos, audio, and codebases, with remarkable accuracy.
To illustrate, 1 million tokens can process up to 1 hour-long video, 11 hours of audio, 30k lines of code or 700k words.
There was a test where it was able to accurately answer questions about a 45-minute-long movie.
In comparison, the biggest context window currently available from Open AI is 128k tokens on GPT-4 Turbo, and 200k tokens on Anthropic’s Claude 3.
According to Demis Hassabis, CEO of Google DeepMind, they are already working on a model capable of 10 million tokens context length.
However, due to the extreme demand for computing power it needs for each run, it won't be available to the public anytime soon.
We’re curious to see how Gemini 1.5 will stand against Claude 3 and GPT-4 in terms of utility and business viability.
Sora: OpenAI's Groundbreaking Text-to-Video Technology Shaping the Future of Visual Storytelling
OpenAI has developed a new groundbreaking technology called Sora, which can transform a brief text description into a high-definition video clip lasting up to a minute.
This innovation marks a significant advancement in the field of text-to-video generation, offering more detailed and longer video outputs than previously possible.
Sora showcases its capability through examples like a detailed Tokyo street scene and a papercraft underwater world, indicating its ability to understand complex interactions in video form.
OpenAI has decided not to release Sora to the public just yet, citing the potential for misuse of such realistic video generation. Instead, the company is beginning to share the model with third-party safety testers and a select group of video creators and artists to gather feedback and ensure safe deployment.
Despite the potential for misuse, the development of Sora represents a major leap forward in video generation technology, offering new possibilities for storytelling while also highlighting the need for careful consideration of ethical implications.
What would releasing this model to the public cause is something we can’t even begin to imagine. But we can be sure it will have a massive impact on content creation and security. We will be keeping our eyes open for the newest developments in the field.
Another hit for Transparency? - Mistral AI launches a non-open-source model after partnering with Microsoft
Mistral AI, a European AI company originally recognized for its dedication to open-source AI models and positioning itself as a European alternative to more secretive US companies like OpenAI, has stirred controversy with its recent partnership with Microsoft.
This collaboration involves a stake for the tech giant and the introduction of a non-open-source product named Mistral Large.
Critics argue this move mirrors OpenAI's shift from open to closed models, raising questions about Mistral's dedication to transparency and open-source principles.
Despite the backlash, Mistral maintains a dual strategy of offering both open-source and commercial models.
The partnership with Microsoft is seen as a strategic move to monetize and scale Mistral's offerings, leveraging Microsoft's vast infrastructure for development and distribution.
This collaboration has stirred debate over its potential to strengthen Microsoft's dominance in AI and concerns about European AI sovereignty.
Nonetheless, some industry observers view the deal as a pragmatic step for Mistral to grow and compete globally, highlighting the challenges European AI firms face in balancing openness with commercial viability.
Stack Overflow and Google Join Forces: OverflowAPI to Boost Developer Productivity with AI-Driven Answers
Stack Overflow, a popular website where developers find answers to their coding questions, is teaming up with Google to make its vast collection of information more accessible to artificial intelligence (AI) systems.
This partnership involves a new tool called OverflowAPI, which lets AI, like Google's Gemini for Google Cloud, use Stack Overflow's data. The idea is to make it easier for developers using Google Cloud to get helpful, verified answers directly in their work environment.
This collaboration aims to blend human expertise with AI to ensure developers can trust the answers they receive, enhancing their efficiency without compromising on quality. Additionally, this integration into Google Cloud means that developers can find Stack Overflow's answers alongside Google's own responses, making their search for solutions more seamless.
This initiative highlights the importance of human expertise in maintaining the quality of information in an increasingly AI-driven world, ensuring developers have reliable resources at their fingertips.
That’s all for this issue. Hope you’re all caught up with the newest AI developments.
Until next time! 🙌
