> run a ton of agents in parallel most of the time
What makes you think everyone (and government) should play along and align with your way of dependency on AI? Not even 1% of the people use AI the way you do. Fable model is not a basic need. Government represents the average Joe. You could also say "I make a ton of nuke weapons and this government has stopped the public sharing of how to make them!".
I think you are missing the bigger picture that is around the "bigger picture" you are seeing. AI proliferation is more dangerous than nukes proliferation, as any highly capable tech would enable destructive usecases as well. If nukes related material and knowledge was safeguarded, then AI requires it as well.
nobody ever raised money for nukes from public/private markets on the premise that nukes will bring the world into an age of abundance. AI companies have done that. This comparison of AI and nukes is so silly.
Raising money has nothing to do with the bad usecase for tech. Tech companies never said that their tech can't be used against the nation or against the good of people.
That's not correct. If human attention requires human effort, that forces human effort all the way down the chain, with no machine output being possible.
You can't say "you can't feed me machine output directly". Machine output is meant to reduce the cognitive load for human processing.
If your colleague is forwarding AI output directly to you, that means they think the AI has reduced the cognitive load for you, and also you are the best person to process that output, instead of them.
You just need to change your perception about the purpose of AI.
I got off from all anthropic stuff a while back. And I feel the fresh air again. No bloated reasoning or code. No vendor lock-in (due to complexity increase in code). Money saved too. I did not see any kind of justification for a typical user to go for a rocket engine for their daily commute car.
Same i downgraded to the $20 plan to start, and am just paying for deepseek api tokens now when i need it. Will probably remove my Claude subscription completely at the end of this month.
Back in the day, personalization / customization was all the rage, as it lets the user feel the control, power and freedom. Now it's the opposite. It's about not letting user to have any control at all. I can't delete some junk apps from my phone and mac, because they are "system" apps. As a non-geek, I can't deal with complexity of the browser and account settings to stop it from what is doing. We are at the mercy of the machines.
It hurts to have all this control stripped away. Once upon a time, you bought iLife (suite of iPhoto, iMovie, etc) on a CD or DVD and installed it. Today, you physically cannot delete the Photos app no matter what.
On my work computer, where I never manage any photos, have no iCloud account and never will, I have to keep this app installed and anytime I so much as AirDrop a png to my computer I am prompted to "Add to Photos" with it. No thank you.
The .app is actually only 41MB, so obviously they've moved the majority of it to some mystery-meat libraries or frameworks installed elsewhere anyway.
Highly recommend trying a linux computer with kde. I have been having similar annoyances with my macbook and switching to kde felt really good. Even though the hardware of the linux pc isn’t nearly as good.
Why would the agent send the results of the query "Show me my recent transactions" to LLM? This pretty deterministic results which involve no LLM interpretation or decision making.
I understand that people are no longer writing IF expression in their code, because they think it's too brittle, and so they delegate all "IF" branching logic to LLM, but it beats me why displaying of the results from a database query should involve LLM.
Because the question they're asking isn't "What is the best way to solve this problem" the question they're asking is "Where can I shove my AI into this product".
Because they want the user to be able to say things like “show me my transactions for business meals in the last month”. That requires an LLM to analyze the transaction descriptions.
Why would the agent send the results of the query "Show me my recent transactions" to LLM? This pretty deterministic results which involve no LLM interpretation or decision making.
The confusion is due to blurring of distinction between the roles: Tool, Tool user and Tool provider. Traditionally, a human-agent who can have an intent and trigger an action is held liable for the consequences of the actions, not the tool. That human agent can be the tool user or tool provider.
Tool user is liable in the case of misuse unintended purpose of the tool.
Tool provider is liable when the use of the tool, by design, causes unintended effects despite proper use of the tool.
A simple "AI may make mistakes" line under the box will not help while the box contains false information. The specific information (lines or words) should not be provided if that's a mistake of false.
The article shows a really good point: statements must be accountable on their own, and cannot rely on "further research". This makes complete sense: at what point can I libel someone, and then dodge consequences with "if you do your research, you can get the facts"?
Another point from the article: they are not just aggregating content, but generating it. If you generate falsehoods, that are not even stated by your sources, of course you're responsible.
This can have significant impact to AI in Germany and the rest of Europe, but it's good to question it and hold people accountable.
After people get tired of the "too dangerous to release" punchline, they might come up with "too big to fail". Oh, wait that's already invented in 2008.
reply