Anthropic for government
Source: press release (TLDR)
Claude launched "Gov-focused" models. This is likely clever repackaging for a lucrative market, but it speaks to the direction of the company and potential founder opportunities. Outside of ‘every market’, Anthropic has made it clear which market they want: enterprise. Just look at their website - the tagline reads “AI research and products that put safety at the frontier”. Targeting government contracts feels like a logical progression.
What opportunities might come from this? The models seem to be focused on strategic planning, intelligence analysis, and threat assessments. While clearly defense aligned, I’m hoping they expand to my bread and butter - financial-rooted analysis. Think sanctions avoidance (current hot topic) or international fraud. Could this disrupt the primary approach: gov hires lawyers → lawyers hire consultants → consultants use software? Are these models really going to provide any advantages over the thousands of custom-built solutions? Let me know I’ll build it.
Qwen’s impressive embedding series
Source: press release (TLDR)
Qwen launched an impressive new series of multilingual models for text embedding and reranking. The largest of these models sits at 8B parameters. That’s small enough to run locally on a high-quality laptop. You can run them yourself on Hugging Face.
I love this release, as it will inevitably directly benefit me (relax narcissist). I think this feeds into the narrative: don’t waste your time and money training and fine-tuning current models. Foundational development is just too fast. Yes, I’m speaking to you consultancies. And to you legacy-SaaS trying to ‘embrace AI’. Or rather, go for it. Allow me to take your place.
Other commentary
Enterprise solutions: develop and test locally, deploy in cloud.
More direct paths to premium markets?
The pace of iteration.