Ask HN: Is Anthropic being public better or worse for your org?
Now that Anthropic has a "fiduciary duty to maximize shareholder value"*, have you or your org started thinking about what that means medium to long term? For example, if you are using their models in production or production-adjacent systems, how do price changes impact your business?
* I know they have a complicated governing system, but at the end of the day they have different pressures now than when they were private.
I fully expect when push comes to shove, they will start to prioritize profit above all else.
Realistically the current state is that everyone's claude usage is very likely being heavily subsidized in order to juice growth.
Especially since the management structure they setup has nothing to do with how it's Capitalism other than google's same "dont be evil".
So start your count down clock. It'll be just as bad, but probably exponentially worse given the whole "singularity" thing.
Byebye internet.
I suspect they'll miss earnings, panic, raise costs and that will be worse for our org cost wise, but likely better in the long term if it results in less reliance.
I think it is both as they claim that their AI will cure cancer
Well at some point they will try and start to ramp up prices, so long-term it might be detrimental. In general I think the reasoning of these companies is that a single developer costs 10,000-20,000 USD per month so if they can be 3-4x more productive then basically companies should be able to pay around 10,000-20,000 USD for tokens for one employee. Today that sounds insane though, so not sure if we will ever get there. And of course that only works if running equally powerful development models on your own hardware is constrained, foreign providers are banned and domestic providers collude on price. Expect them to work more agressively towards all three things happening.
Alternatively they will work on expanding the scope of use for these models, you can already see that with ChatGPT that e.g. markets to small shop owners, house wives etc. with big ad campaigns, which is interesting as these folks will not spend thousands on tokens per month, so I guess they are more focused on ads and other ways to monetize personal data with cheaper models. Too early to say but seems like OpenAI might acknowledge that they are losing the enterprise / large business market to Anthropic and therefore are making a grab for the consumer / small business market. At least it seems like that for me, from my experience Anthropic is building a pretty good moat in the B2B space and OpenAI is quickly losing its footing there. Not sure if it will work out for them, Google, Meta and others are going for the same market it seems.
But yeah if Anthropic owns the B2B AI market expect them to come up with ways to extract a significant portion of companies staff budget. Whether that will work, who knows, they need to make sure local or self-hosted AI stays out of reach for most companies but not sure if that is realistic, though it could still make sense to companies if Anthropics' stuff "just works" and produces good artefacts with minimal prompting, I think that will more be the differentiator, with all the prompts they get they can optimize and "steal" results from users that manually steer the model to correct outputs for a specific domain problem to offer the same solution to other users through the guise of "anonymous" training... For example I spent months guiding AI through a complex software use case, correcting assumptions and putting all my domain expertise into the system. That conversation is then worth a lot of money for Anthropic if they can train on it and make sure my exact use case will work without all that manual guiding work for the next user that attempts it. So they can essentially extract my expertise, learn from it, then automate it for everyone else, without me getting anything in return (I even pay for the privilege to use the model). They will basically be like McKinsey, a sanctioned way to steal business secrets from your competitors in a way that white washes copyright theft. Not saying this is a bad thing, before this was also possible but you had to poach employees, now the knowledge transfer might happen through retraining of models based on proprietary data. I therefore expect them to soon demand a hefty premium for keeping prompts and conversations in their ZDR tier, lots of companies will then simply give this up.
That's also why the individual plans are so "cheap", we are essentially training their models for enterprise customers with different use cases for free. In that sense we're not much different from all the off shored folks that annotated ML models, except these folks actually get paid for their work...