Artificial Intelligence Updates
By Jeff Wang
At RocketFuel, we see tremendous opportunity where web3 meets AI. So we track AI closely. Here’s what we’re watching now (published here on April 1, 2025):
DeepSeek is back in the news again, it dropped a new model that performs on par with Claude. While this is still the V3 variant, we’ll likely see a reasoning model from this come out following their previous V3 -> R1 1 month timeline.
Sesame went viral for an insanely low latency and realistic voice demo, it responds almost even before I can stop speaking and seems to get a pretty good understanding of what you are trying to ask for. I was almost skeptical that I was talking to an AI and not a real person.
Manus also went viral for releasing an autonomous agent for your computer. It uses Claude Sonnet with a lot of different tools to access different files and systems to accomplish a number of complex tasks. They are also seeking funding at a $500M valuation. MIT wrote a poor review for it being very slow, one of the issues with Cognition’s Devin as well.
Xai released Grok Deeper Search, which I guess was trying to avoid using the name Deep Research again, to search a few quality sources to come up with its report. They also released Image editing capabilities which they have announced as its Aurora image model. They also acquired X in an all stock deal, which was at the $45B initial price point but also added $12B in debt.
Apple delayed their release of their Siri upgrade to 2026 and restructured their AI team. Apparently the accuracy ratings were far too low for general availability.
Nvidia announced the GB300 and renamed project digits to the DGX Spark, which is a personal supercomputer with a blackwell chip inside. If it runs gaming on a 4k monitor I might be game.
Google released Gemma 3 for local machine use and image manipulation on Gemini 2.0 Experimental. But it really made waves with Gemini 2.5, which is making waves as one of the top new coding models on the market.
Of course, OpenAI was heavily in the news this month, they were rumored to release agents that could cost from $2k to $20k a month. The rolled out GPT-4.5 which was a massive model but very expensive. They rolled out a new Voice API and product called OpenAI.fm. Softbank is close to finalizing the $40B round as well. Finally, OpenAI rolled out a new image model that uses auto-regressive techniques instead of diffusion that is fantastic with realism, editing existing images, and even text. This might be the new top model in image generation. They caught a major viral trend by generating Studio Ghibli versions of existing images, which I hope they do not ban because they show the delightful side of AI, oh and here’s me with as an anime character:
