Relevant AI News - May 2025
Google's big comeback
Midjourney isn’t out of the picture yet, they’ve updated the V7 which allows you to use an image as reference into another image. However, this does seem to already be a part of the other model’s capabilities now. Black Forest Labs rolled out something similar, called Flux Kontext, please make these open weights…!
Apple is leaning into Vibe Coding and partnering with Anthropic to launch an Xcode enhancement. This will likely spur a bunch of new app store growth in new apps. This wasn’t without controversy though, as critics were concerned that Apple was behind on the AI train. By focusing too much on margins, they might have been too risk averse in a quickly changing industry. It’s also worth noting that for the first time in 22 years, searches on Safari declined, a nod to ChatGPT and other AI apps taking its place.
Anthropic released Claude 4 Opus and Sonnet, with Opus breaking new ground in the coding space but also being very expensive. They also added Claude Code to extensions in IDE, directly competing with Windsurf and Copilot. They also let employees cash out some shares with a $2M cap. Dario was in the news as well for cutting through the BS and saying that 50% of white collar jobs could be wiped out.
Deepseek released its new reasoning model DeepSeek-R1, early signals indicate it is very good as a general model, being comparable to Gemini 2.5 Pro but below o4. However, it is very slow compared to the other models but at least much cheaper in general to run. This might not level the playing field but it comes very close to a point where it can become a true alternative for a much cheaper price than the premiere models. After all, we’re already in an area where open source models are extremely valuable as they are, we are just used to choosing the best model for the task.
Windsurf released 3 new models, SWE-1, SWE-1-lite, and SWE-1-mini. These provide a cheap alternative to the premium models but SWE-1 in particular is on the level of Claude 3.5 Sonnet. By having free alternatives it gives its users optionality if they want to experiment a bit or if they run low on credits. Windsurf also released a whole suite of Enterprise features.
Google had a huge month. At Google I/O they announced an assortment of items. Some serious game changers include VEO 3, its new text to video model that also has audio. Not only does this look super realistic, but the audio component makes it so that you can immediately start posting content that is indistinguishable from real life in many cases. You can use the Flow app to build these videos. There’s also “AI Mode” which competes directly with Perplexity, though it is a bit buggy so far. There’s Project Astra which acts like Jarvis from Iron Man. There was also Google Beam which makes it look realistic to meet with people by having a 3D view of the other person virtually, but I figure this is probably more for Enterprise. There’s Jules which allows you to run a coding agent in the background. And finally there’s Android XR, which brings a heads up display while you wear glasses, it seems to monitor everything you do as well so you can recall things (as shown in the demo). While these new AI features are amazing, they do require a new Google AI Pro or Ultra subscription, which is running at $20 or $250 a month. The Ultra includes 30GB of storage and Youtube Premium but that’s a fraction of the $250 cost. A year ago, one would have thought Google was being disrupted, this month they showed they are not sleeping at the wheel.
OpenAI acquired io, maybe ironically on the heels of Google I/O, for $6.5B. This brings Johnny Ive, one of the most iconic designers (iPhone, iMac, iPad…), and some of the most talented Apple employees to create the next big AI hardware device. OpenAI also restructured to a Public Benefit Corporation (PBC, something Anthropic is already structured as), to allow itself a future IPO. OpenAI also released its background coding agent Codex as well.
Nvidia made a bet on quantum by investing into PsiQuantum. They are also partnering with Saudi Arabia for Humain, not to be confused with the pin. They’ll get capital to build infrastructure and also train multimodal Arabic language models. Nvidia also announced earnings that exceeded expectations, with an astonishing 69% jump in revenue at $44.1B for the quarter. Pretty insane given the scale it is growing at. Right now Nvidia and Microsoft are fighting for the top market cap spot, each at around $3.3T as of typing this.