Relevant AI News January 2025

A more consequential month than usual

As usual, let’s go over the relevant AI news last month in a bite sized article so that you don’t need to obsessively doom scroll Twitter like the rest of us, this time it’s a doozy

A massive AI infrastructure entity, Stargate, was unveiled at the White House, which brought the United States, Softbank, Nvidia, Oracle, OpenAI, MGX, ARM. According to The Information, OpenAI and Softbank would commit $19B and have 40% ownership, while Oracle and MGX would contribute $7B. The initial investment would be around $100B but could go up to $500B. In theory, it could be the world’s largest AI supercomputer and used to really push the frontier of where AI could have slowed otherwise. But…

DeepSeek proved that you might not need massive AI spend to push the limits of AI. After all, by imposing restrictions on GPUs for China, it would make sense for the smart folks over there to come up with more efficient ways to train and run models. We’ve written an entire article for this one as it has major implications for stock prices and for how companies will approach AI going forward. It’ll be a net benefit for the AI industry, but it may mean China has caught up to the United States. Despite not having pushed frontiers yet. Meanwhile, ByteDance announced Doubao-1.5, but basically no one talked about it because of the DeepSeek news. Alibaba also announced Qwen 2.5 Max, which is better than DeepSeek V3, but again, no one seemed to care…you’ll have to be R1 in order to grab the news cycle. If you haven’t already, check out our other article about DeepSeek

OpenAI announced “Operator”, which can run tasks on your web browser. I already have a ton of ideas for this so I’m eager to try it out. Speaking of tasks, OpenAI also announced “Tasks”, which allows ChatGPT to schedule reminders or jobs based on prompts. For example, “Everyday at 8AM, create a news brief for me and then email it to my inbox”. Combined with other agents, this could be powerful, for example, “Answer all my emails after 5PM if I haven’t read them yet”. Okay, maybe I can’t do that yet. It’s still early on both these features, but I’ll probably give them a try when they get more powerful. They also announced ChatGPT for the government. Finally, OpenAI is in talks to raise $40B at a $340B valuation, the round would be led by Softbank, who is rumored to invest as much as $25B into the round. This is going to pull up AI valuations all over the place, Softbank once again sets the tone for the VC world. Note, $340B would put OpenAI at the 29th most valuable company, ahead of Salesforce and SAP. OpenAI also released o3-mini on the last day of the month, priced competitively and likely in response to DeepSeek

Luma Labs unveiled Ray2, another advancement in the text to video arena.

Mark Zuckerberg’s quote on Joe Rogan went viral "Probably in 2025, we at Meta, as well as the other companies that are basically working on this, are going to have an AI that can effectively be a sort of mid level engineer that you have at your company that can write code."

Everyone is waiting on Grok 3, but we are getting glimpses that it will be the “uncensored” model. Grok mentioned it would probably have an “unhinged” mode. Meanwhile in the uncensored AI space, Venice.ai announced a massive crypto airdrop. Subscribers got $5-6k of VVV, the token used to access the API. Venice is led by ShapeShift billionaire Erik Voorhees. It uses Qwen and Flux behind the scenes which are uncensored models.

Nvidia unveiled the 5090 at CES, the next consumer GPU with 32GB and roughly 20-30% more power than the 4090, buyers are already camping out to buy it. It will cost $1,999, so my wife will probably not be happy if I bring the tent with me tonight. Nvidia also announced Cosmos, which are foundation models for the advances in robotics. They announced Digits, which is a blackwell chip in a box which will cost $3k. This looks to be trying to get back some of the hobbyist LLM projects that have been using Mac Minis. Also, they announced Blueprints which are templates for AI agents. Finally most important (to me…), they called out Codeium, Jensen Huang went on stage and said "Codeium...every software engineer in the world, this is going to be the next giant AI application, the next giant AI service period"