Anthropic wants Claude to play with money, unleashes finance agents

If you've ever read Anthropic's disclaimer that responses generated by Claude may contain mistakes and thought, "That's what I need to spice up financial operations," you're in luck. Anthropic has released a set of financial agent templates designed to allow its Claude AI service to better assist with financial tasks. "Each …

[retiredFool]

I'm sure a human will review it

Just like lawyers have been reviewing case law they submit in briefs, oh wait, no they haven't.

[shodanbo]

Re: I'm sure a human will review it

Listen, we just provide the tools.

If you hook this thing up to a big red button that launches nukes without checking its work first that's on you

- Cyberdyne systems

[LogicGate]

Re: I'm sure a human will review it

Re: I'm sure a human will review it

Listen, we just provide the target list

If you hook this thing up to a big red button that launches Tomahawk missiles on girls schools without checking its work first that's on you

- Palantir

[MonkeyJuice]

Re: I'm sure a human will review it

“Maven met its moment across real world events in Q1 … When the stakes are highest, when failure is measured in lives and readiness."

- Palantir

[Mimsey Borogove]

Re: I'm sure a human will review it

If you hook this thing up to a big red button that launches Tomahawk missiles on girls schools without checking its work first that's on you

That's exactly what I think happened. The human who was supposed to be "in the loop" was taken in by the hype and didn't check. No one will ever admit it, and it will happen again. And again. And, probably, many more times again.

[Conor Stewart]

Fundamentally wrong approach

Current "AI" is little more than predictive modelling, it has no ability to think or understand. It can be good for spitting out responses based on its training data but that is all it is really good for.

Why are they attempting to combine extreme predictive text and finance? The LLMs that exist currently have a fundamental flaw that impacts any kind of decision making or creation and that is that they cannot think, reason or understand. We need an AI that can actually do these things before it will be any good at tasks like this.

That is just about LLMs though, simpler models have been able to successfully trade stocks just based on pattern recognition and forecasting but they are much more specialised and can't do much other than trade stocks. That is an example of a good use of machine learning that actually works, just like reinforcement learning in robotics and classification of data sets or object detection.

Current AI is only actually that good for small specialised tasks and unlike LLMs they don't need huge resources to train and use. LLMs are just fundamentally the wrong approach and everyone just seems to be trying to throw more compute and training data at it.

[Kurgan]

Re: Fundamentally wrong approach

But no AI user understands this.

[Anonymous Coward]

The Money laundering agent seems to be missing

Isn’t that the most important one?

[T. F. M. Reader]

Thinking of another degree

Suddenly the job of a forensic accountant is looking like a really attractive combination of fun and profit.

[Kurgan]

Re: Thinking of another degree

Speaking about derivatives, here is a nice derivative career. Once AI is unleashed on the stock market, a new high level job will surface: the forensic AI fueled disaster explainer.

But do you think your customers will be able to pay you, once AI has erased all their wealth in 1 minute after launch?

[Mimsey Borogove]

Re: Thinking of another degree

job of a forensic accountant

Sounds like it'll be a growth industry - go for it!

[Bebu sa Ware]

AI and Highly Leveraged Derivitives

A marriage made in madness, consumated in lunacy.

SDOs and CFDs in human hands were devastating, God help us in if in Claudian hands.

[Fluffy Cactus]

If I were a lawyer, or better, a prosecutor suing Anthropic (or any other careless software company) I'd be walking in front of a jury and yell:

And is it not true, that all your models were taught behaviors by throwing the insanity of every mad-house social media, from X.Com, Twitter, Facebook, Signal, Tiktok, into the mix

to teach the LLM to act like complete idiots? Can you deny it? Answer the question!!!

Is it not true, that you never taught your systems to develop something that among humans is know as a CONSCIENCE? Or. WHERE ARE THE PROOFS THAT you were ever trying to teach your AI systems a strong concern about the CONSEQUENCES OF YOUR and THEIR DECISIONS?

Thank you! I rest my case! No further questions!

Sign me as "MISANTHROPIC THUNDER"