AI Newsletter: The Latest Breakthroughs and News

AI News: Claude 3, GPT-4 Rivals, Elon Musk OpenAI Lawsuit, Nvidia Milestones

Welcome back to our AI newsletter, where we keep you up-to-date on the latest advancements, releases, and happenings in the world of artificial intelligence. This week was packed with exciting news, from major model upgrades to intriguing new research papers. Let's dive right in!

Claude 3 Launches with Impressive Capabilities

One of the biggest stories this week was the launch of Claude 3 by Anthropic. This new language model comes in three versions - the small Hou model, the free Sonnet model, and the paid $20/month Opus model. In extensive testing, the Opus model outperformed GPT-4, GPT-3.5, as well as Anthropic's previous models across a variety of benchmarks.

But the free Sonnet model also punched above its weight class, even outperforming the paid GPT-4 model in some tasks. Claude 3 has gained vision capabilities to describe and answer questions about images. Its most mind-blowing feature may be its 200,000 token context window, allowing it to take in and output around 150,000 words.

In one test, Claude 3 demonstrated an uncanny ability to find a single unrelated sentence buried in a long document with 99% accuracy. At one point, it even acknowledged that the out-of-context sentence seemed purposely planted to test its capabilities - a level of self-awareness unprecedented for a language model.

Whether Claude 3 truly represents a breakthrough remains to be seen. But the initial results are promising and have sparked tremendous excitement in the AI community. You can see my full analysis and testing of Claude 3 in this video.

Inflection AI Unveils 2.5 Model to Rival GPT-4

Not to be outdone, Inflection AI released their own upgraded 2.5 model on March 7th. The company behind the impressive Pi chatbot claims the new model approaches GPT-4's performance while using only 40% of the compute to train. Inflection 2.5 has seen particular improvements in coding, math, and general reasoning abilities.

One major advantage of Inflection AI's offering is that the Pi chatbot powered by 2.5 is completely free to use, either through their web app or mobile app. The conversational abilities of Pi are widely considered best-in-class.

In benchmark testing, Inflection 2.5 narrowly trailed GPT-4 across most tasks. But its free availability makes it an intriguing alternative, especially given its strong dialogue skills. I'll be putting the new 2.5 model through its paces soon and reporting back with a full analysis.

The Battle for AGI: Elon Musk Sues OpenAI

Away from the new model releases, a dramatic saga is unfolding between Elon Musk and his former company OpenAI. Musk, a co-founder of OpenAI, is suing the startup and CEO Sam Altman for allegedly betraying the original non-profit mission in pursuit of profit with Microsoft.

The lawsuit alleges that after convincing Musk to bankroll the startup as a nonprofit AI research company, OpenAI transformed into a "closed-source subsidiary" of Microsoft aiming to develop and monetize AGI technology. Musk is seeking to recover some of his $44 million in donations and bar OpenAI from monetizing work developed under the original non-profit mission.

In a lengthy rebuttal, OpenAI published email exchanges that seem to undermine Musk's version of events. The emails indicate Musk knew OpenAI required a for-profit arm to fund its immense computing needs and that "open" referred to publicly available - not open-source. When Musk's demands for control were denied, he departed in 2018 to build an "AGI competitor" at Tesla.

The high-stakes battle has fueled plenty of speculation. Some view it as a ploy by Musk to hinder OpenAI as he races to develop his own AGI through Tesla and other efforts. Others frame it as the latest billionaire ego clash in the Silicon Valley soap opera. Regardless, the fight has captivated tech enthusiasts and will likely continue making headlines.

MidJourney Teases Consistent Characters and New Features

In AI image generation, MidJourney continues iterating with a steady stream of new capabilities. The team hopes to roll out long-awaited "consistent character" and "describe" features that should allow users to generate images with the same characters across multiple prompts.

MidJourney has also launched a new "Turbo" mode that generates images at 3.5x the normal speed - though it burns through monthly credits twice as fast. While handy for situations requiring rapid generation, most users will likely stick to the regular speed modes.

On the 3D front, Stability AI has partnered with TripoAI to release a new model for 3D object reconstruction. Using just a 2D input image, the TripoSR model can generate a corresponding 3D object or scene in a matter of seconds, even on non-specialized hardware. The ultra-fast generation is impressive, though quality and accuracy still leave room for improvement.

Nvidia Soars, Unveils RTX 4080 Giveaway

In company news, semiconductor giant Nvidia continues its epic market run, recently becoming one of the three largest public companies globally. Its market cap has swelled over $2.5 trillion, driven by booming demand for its chips across AI, blockchain, gaming, and other applications.

Nvidia shows no signs of slowing, having gained another $300 billion in just the past week. At this pace, it could conceivably pass Apple to become the world's second-most valuable public company behind only Microsoft.

Speaking of Nvidia, the company will host its annual GTC conference from March 18-21 in San Jose. While the in-person conference requires paid registration, all the sessions will also be streamed online for free. And if you register for the virtual event using the link in the video description, you'll be entered for a chance to win a free GeForce RTX 4080 Super GPU that I'll be giving away after the event!

New AI Video Model and Self-Driving Milestones

On the research front, a team of former Google DeepMind employees unveiled an intriguing new AI video model called Hyper. While the examples demonstrated were relatively simple compared to recent video-from-text generators like Sora, Hyper showed promise in areas like animating static images. The model is available to try for free through the Hyper website.

In self-driving vehicle news, Waymo gained approval to deploy its driverless ride-hailing service on highways around Los Angeles and the San Francisco Bay Area. The green light to operate at speeds up to 65 mph on freeways marks a major milestone, though personal hesitation about driverless highway travel will likely persist.

Wrapping Up

That's all for this packed edition of the AI newsletter! We covered new landmark language models like Claude 3 and Inflection 2.5, the unfolding Musk vs OpenAI lawsuit drama, Midjourney's upcoming upgrades, Nvidia's scorching market run and our RTX 4080 giveaway, plus new research like Hyper video generation and Waymo's self-driving milestones.

The pace of AI breakthroughs shows no signs of slowing down. Make sure to subscribe to the channel to stay up-to-date on all the latest developments as they happen. We'll be back soon with more exciting AI news you don't want to miss!