This week in the world of artificial intelligence brought some major developments, from the launch of Anthropic's powerful new Claude 3 models to Elon Musk's lawsuit against OpenAI and the unveiling of Google's impressive text-to-video game model called Genie. Let's dive into the key AI news stories of the week.
Anthropic Launches Claude 3 Models to Rival GPT-4
Anthropic made waves this week with the release of their new Claude 3 AI models, which come in three versions - Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. In benchmark tests, the premium Opus model outperformed GPT-4 and Anthropic's own Gemini 1.0 Ultra in areas like undergraduate level knowledge, graduate level reasoning, math, coding, and more.
Impressively, even the free Sonnet model beat out GPT-4 in several benchmarks. Claude 3 also gained sophisticated vision capabilities on par with other leading models. Another standout feature is Claude's long context window and near-perfect recall, allowing it to find a "needle in a haystack" of information with over 99% accuracy.
While the free Sonnet model has a daily message limit of around 20, the $20/month Opus offers 5x that usage. Overall, Claude 3 shapes up to be a very capable alternative to GPT-4 at a compelling price point.
Elon Musk Sues OpenAI Over Alleged Betrayal of Mission
In a surprising turn of events, Elon Musk, one of the original founders of OpenAI, filed a lawsuit against the company this week. Musk claims that OpenAI's leaders convinced him to help found and fund the startup in 2015 with promises that it would be a non-profit focused on countering Google's competitive threat in AI.
However, Musk alleges that OpenAI has since transformed into a closed-source, for-profit subsidiary of Microsoft, betraying its original mission. OpenAI fired back, sharing emails suggesting Musk knew parts of the company's science could be closed and that he even proposed merging OpenAI with Tesla at one point to stay competitive.
The lawsuit underscores the growing tension between Musk's AI ambitions with Tesla and OpenAI's close partnership with Microsoft. How this legal battle unfolds could have major implications for the competitive landscape in artificial intelligence.
Google Unveils Genie, A Text-to-Video Game Model
Not to be outdone by OpenAI's recent SORA text-to-video model demo, Google revealed their own generative AI called Genie that can create playable video games from a text prompt or hand-drawn sketch. Genie was trained on 200,000 hours of 2D platform game footage and contains 11 billion parameters.
While the games currently only run at 1 frame per second, the potential use cases are fascinating, from open-ended reinforcement learning environments to one day possibly training robots by observing real-world actions. Genie remains an in-house research project for now, but points to an exciting future for AI-generated interactive content.
Other Notable AI Stories This Week
- Mistral AI launched their new Mistral Large language model in partnership with Microsoft. It ranks as the 2nd most capable model behind GPT-4 available via API and supports function calling to convert natural language to executable code.
- Groq unveiled their innovative LPU (Liquid Processing Unit) chip that can generate an incredible 300 tokens per second, potentially transforming the AI hardware landscape as an alternative to Nvidia's GPUs.
- Meta announced plans to release LLaMA 3 in July, a more advanced open-source language model that can better handle complex and controversial questions compared to LLaMA 2.
- Salesforce debuted new AI tools to help reduce healthcare professionals' administrative burden, including Einstein co-pilot Health actions to automate appointment booking and referrals, and assessment generation to digitize health surveys without manual coding.
It was certainly an action-packed week in the world of AI! Be sure to subscribe to stay on top of all the exciting developments in this fast-moving field. We'll be back next week with another roundup of the latest and greatest in artificial intelligence news.
