And it’s not just them. To me this trend screams “valuations are too high”, and maybe hints at “progress might start to stagnate soon”.
https://www.anthropic.com/news/the-long-term-benefit-trust
https://time.com/6983420/anthropic-structure-openai-incentiv...
It's not the best choice, it's spacer's choice!
Then the people who funded / trained this "justice" out of their good heart, would actually have leverage, in terms of concrete power.
It's a much more subtle way to capture power, if you can replace the judges with your software.
Brave new world, indeed...
The whole thing about no ethical consumption under capitalism is a just a way to enjoy the conveniences of capitalism on a moral high ground. It's totally doable, you just might not enjoy it haha.
The camel's gotta get its nose in the tent somehow.
It wouldn't specifically brag about doing it, while leaving out that they were specifically dealing with Palantir, because they know what they're doing is unethical: https://www.anthropic.com/news/expanding-access-to-claude-fo...
Being available for use by militaries is incredibly irresponsible, regardless of what scope is specifically claimed, because of the inherent gravity of the situation when a military is wrong. The US military maintains a good deal of infrastructure in the US; putting into their hands an unreliable, incompetent calculator puts lives at risk.
It would be structured as a non-profit (there are no teeth to a PBC; the structure is entirely to avoid liability, and if you have no trust in the executive body of an organization, it has zero meaningful signal).
It would have a different leadership team.
It would have a leader who could steelman his own position competently. Machines of Loving Grace was less redeeming than Lenat's old stump speeches for his position, despite Amodei starting up in an industry significantly more geared for what he had to say, and Lenat having an incredibly flexible sense of morality. Its leader would not have a history working for Chinese companies and jingoistically begin advocating for export controls.
It would have different employees than the people I know who are working there, who have a history of picking the most unethical employers they can find, in a fashion not dissimilar to how Illumination Entertainment's "Minions" select employers.
There are sane investors that prefer investing in companies that adopt these corporate structures. Based on data, those investors see public benefit corporations as more profitable and resilient. They are able to attract employees and customers that would otherwise not be interested or might be less interested.
What is "the agency problem"?
In modern management compensation theory (https://saylordotorg.github.io/text_introduction-to-economic... ) this is key to why executive compensation has increased much faster than workers in the last 50 years.
Stock based compensation mix evolved from this thesis, and quite common in the valley and why almost all OpenAI staff wanted Sam Altman back even though the non profit board did not.
Aligning key talent's compensation to enterprise value is only viable in unrestricted for profit entities any other structure with limits (capped profit, public benefit corporation, non profit, trust, 501c's etc) does not work as well.
Talent will then leave to a for-profit entity who can offer better compensation than a restricted entity can because they share a % of their enterprise value which restricted ones either cannot or not have same liquidity/value [1] etc.
---
[1]This is why public companies are more valuable for RSU/options than private companies, and why cash flow positive companies like Stripe still raise private money to just give liquidity to employees .
Semi-relevant sidenote: ChatGPT, spent $8m on a super bowl commercial yesterday just to show cool visualizations instead of any emotional product use case to an ultra majority audience has never had a direct experience with the product.
These companies would be best served building a marketing arm away from the main campus in a place like LA or NY to separate the gen pop story from that of the technology.
I think AI in its current iteration is going to settle into being like a slightly worse version of Wikipedia morphed with a slightly better version of stackoverflow.
At the base of LLM reasoning and knowledge is a whole corpus of reasoning and knowledge. I am not quite convinced that LLMs will breach the confines of that corpus and the logical implications of the data there. No “eureka” discovery, just applying what we already have laying around.
Well over 90% of work out there is not novel. It just needs someone to do it.
Plenty of value is already added just by converting unstructured data to structured data. If that is all LLMs did they would be still be a revolution in programming and human development. So much manual entry and development work has essentially evaporated overnight.
If there was never a chat based LLM "agent" LLMs just converting arbitrary text to structured JSON schema would be the biggest advancement in comp sci since the internet. There is nothing equivalent that existed before except for manual extraction or rule based hard coding.
Judging LLMs based on some criteria of creativity or intuition from a chat is missing the forest for the trees.
Just to try it out, I uploaded the paper to DeepSeek-R1 and wrote a paragraph on the desired algorithm, that it should code it in Python and that the code should be as simple as possible while still working in exactly the way as described in the paper. About ten minutes later (quite a long reasoning time, but inspecting the chain of thought, it did almost no overthinking, but only reasoned about ideas I had or should have considered) it generated a perfect implementation that worked for every single test case. I uploaded my own attempt, and it correctly found two errors in my code that were actually attributable to naming inconsistencies in the original paper that the model was able to spot and fix on the fly. (The model did not output this, this I had to figure out myself.) I would have never expected AI to do that in my lifetime just two years ago.
I don't know whether that counts as "novel" to you, but before DeepSeek, I also thought that Copilot-like AI would not be able to really disrupt programming. But this one experience completely changed my view. It might be the case the model was trained on similar examples, but I find it unlikely just because the concrete algorithm cannot be found online except for the paper.
But this is not the majority of what software developers are doing and working on today. Most have a set of features or goals to implement using code satisfying certain constraints, which is what current reasoning AI models seem to be able to do very well. Of course, this test was not rigorous in any meaningful way, but it really changed my mind on the pace of this technology.
Combined with the old “nothing new under the Sun” maxim, in that most ideas are re-hashes or new combinations of existing ideas, and you’ve got a changed landscape.
And if the flywheel is that AI begets AI exponentially in an infinite loop then those share certificates you own probably won't be worth much. The AI won.
Coincidentally, Anthropic's mission is AI safety.
That said, this doesn't seem like completely superfluous "fat" like what Mozilla does. It seems very much targeted at generating interesting bits of content marketing and headlines, which should contribute to increasing Anthropic's household brand name recognition vs. other players OpenAI, as well as making them seem like a serious, trustworthy institution, rather than a rapacious startup that has no interest in playing nice with the rest of society. That is: it's a good marketing tool.
My guess is that they developed it internally for market research, and realized that the results would make them look good if published. Expect it to be "sunset" if another AI winter approaches.