I've finally managed to break free of the basement so I'M BACK writing this week's news drop! (Tragically the thing I was working on didn't work)
Anyways, here's the news:

Starting out with Microsoft:

While this really isn't a development in AI, it's close enough that I'm still including it. Also, it's funny: In the latest version of Windows, the "copilot" feature that pops out a side panel to let you interact with basically ChatGPT was broken! Any attempts to use Copilot just resulted in an error message. In my opinion, this is actually a good change. Microsoft's approach to shoving AI everywhere has generally been met with quite a lot of pushback, and a lot of people agree that perhaps this should just be a regular feature...

Didn't expect Xbox to show up here: Xbox (so yeah, still Microsoft) has announced their own AI model. It's designed to watch over your shoulder and give "helpful gameplay advice." Due to the high latency of most current AI tools, the odds of this actually being helpful seem quite low, and imagining a future where you end up on a game over screen, only to have an AI tell you "you died!" really just makes me feel sad.

Nvidia's cooking with gas this week:

The tiny Blackwell (Nvidia's latest GPU architecture) supercomputers from Nvidia will get a larger desktop counterpart! This is huge as these are basically personal supercomputers, although they do come with a price to match. As of right now, these seem to be Apple's only competition in this space, as no one else is doing anything with nearly the same amount of performance or RAM.

Additionally, Nvidia dropped a new version of NeMO, called "Llama Nemotron..." and the models are:

• Nano (8B)
• Super (49B)
• Ultra (253B)

So this is not actually a new model, it's just a version of Meta's Llama that's been shrunk down while having its performance increased. Still definitely impressive stuff to be sure, but I'm beginning to lose it over these names. ...And yes, they all are chain-of-thought (thinking) models, as it seems most new models are.

Intel has a new CEO!

Lip-Bu Tan is replacing Pat Gelsinger as CEO. Intel's in some troubled times right now, and it'll be interesting to see if he can pull them back into relevancy as the AI race continues to heat up.

Google time:

Turns out Imagen3 with Gemini is really good at removing watermarks from images! It's exceptionally good at figuring out what should be behind the watermark, and is almost indistinguishable from the original image. For now the feature is only available to developers through AI studio, and has less of the guardrails that the "public" version has.

Gemini's decided that if everyone can steal the name of "deep research," then they're just gonna clone OpenAI's "canvas." Canvas is a new tool from Google that lets you work collaboratively with Gemini side by side on writing or coding projects. It makes it much easier to visualize as you go, and is very similar to OpenAI's implementation from a while back!

Audio Overviews from NotebookLM are now available in any Google Doc! Just hit the button and get a podcast that covers the topics of the document in detail. I've used NotebookLM a lot myself and the results are shockingly good, including pauses for breath, occasional coughs, and sometimes accidental microphone bumps...

Google's been releasing a TON of AI stuff considering they have a major event (Google I/O) coming up in a few months... I'm not sure what this means, but I expect big things from them in May!

That's it for this week!