Claude magic

The hype surrounding Anthropic's new Claude models mirrors past overblown claims from OpenAI and other tech companies, where each new release is hailed as groundbreaking until it is quickly dismissed as obsolete. It suggests that Anthropic differentiates itself not through superior technology but by cultivating a quasi-religious narrative about creating sentient AI, which may prove unsustainable as the market demands financial returns over mystical promises. The author questions how long this "magic spell" of Anthropic's narrative can last in a tech landscape driven by feature releases and hype cycles.

Anthropic’s Claude models are having a moment. Their newest release supposedly “changes everything” and if you don’t use it every day you just cannot have any opinion on LLMs. Which is the same story we heard about previous OpenAI releases or previous Anthropic releases. Every new and more expensive model is groundbreaking and almost AGI until a few months later when you get yelled at for using that old piece of garbage. Remember when everyone was raving about Perplexity and how great their stochastic systems were? This of course is not fully new: Tech lives on closely following the release cadence of the big players in the market. You just need the newest version with all its features there is an interesting question in the distinction of feature and actual solution to a problem and why tech is sold based on the former and not the latter in order to even understand what’s going on. Aside from it being a convenient way for wealthy and well-connected tech evangelists and journalists sadly that Venn diagram is increasingly becoming a circle to lock out others from influencing public perception and opinion it allows tech to keep its momentum: Everything is always moving forward. Everything just gets more features all the time. Progress If you haven’t seen it I have spoken about this a bit in the past But there is a bit of a different vibe pun intended to how people talk about the Claude models and I find it interesting how non-technological narratives shape that seemingly rational crowd of “AI” users. Anthropic started out as a splinter sect you have to remember that all that “AI” stuff is a secular religion for dudes afraid to die splitting off from OpenAI because OpenAI was no longer just looking for god in whatever they could scrape from the Internet: OpenAI just released whatever they made quickly hoping to keep the hype going and pushing for integration into actual work pipelines without all the “oh we are so afraid if AI might be sad” and whatever other weird fan/slash fiction for their own models Anthropic keeps releasing. Claude’s recent iterations might be even better fine tuned to push out some random JS code but OpenAI or Microsoft will surely catch up. It’s all just a numbers game about who’s willing to burn even more money on energy and infrastructure. But OpenAI can no longer replicate the religious narrative as well as Anthropic can. And it’s gonna be interesting to see how long “sure all models are buckets of statistics but ours can be sad and will soon be god” can actually last in a market that’s increasingly looking for some financial returns. How long people are willing to accept the utter weirdness of Dario Amodei and his claims. “AI” is always playing with this notion of the great leap towards something magical. It’s never just about some form of automation or ways to massage data and build workflows. If you say “AI” you are pointing at humankind’s desire to create god or “new life”. Anthopic puts that front and center. That’s why they have their bunch of effective altruist philosophers tasked with writing “constitutions” for their stochastic models: It’s not about actual functionality but it is scaffolding that is supposed to hold up their actual narrative: Which is not that they built a good spicey autocomplete but that they have or will soon create life that will surpass us capacity wise. And I wonder when Claude’s magic spell will break. This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.