You’re going to give your tax data - some of the most sensitive data to some constituents - to OpenAI / Google / some other startup?

That seems like a nightmare of a product as far as privacy is concerned.

▲

jordanb a day ago | parent | next [-]

Fwiw they have already bought all you financial info from Experian

https://theworknumber.com/solutions/products/income-employme...

	▲	vjvjvjvjghv a day ago \| parent \| next [-]
		I was flabbergasted when I heard of this. Basically you are totally transparent for anybody who wants to spend some money.
	▲	chneu a day ago \| parent \| prev [-]
		Being an American with so much freedom is so refreshing Oh shit, wait.

▲

simonw a day ago | parent | prev | next [-]

I think they meant that it should be a lot faster to develop software that implements the tax code with the assistance of AI coding tools.

	▲	latexr a day ago \| parent [-]
		You need legal documents to be accurate and deterministic, not for some LLM to make shit up and have you inadvertently and incompetently lie to the IRS.

▲

terminalshort a day ago | parent | prev | next [-]

The only reason I care about companies having my data is that it means the government can get to it. In this case I am required to give my data to the government anyway, so why would I care if OpenAI / Google has it?

▲

amluto a day ago | parent | prev [-]

ISTM one ought to be able to use AI to translate the official IRS forms to a machine readable format. No personal data needs to go anywhere near the AI.

Even if you do want to feed your personal data to an AI tax bot, this should be easily within the capabilities of a model that can run locally.

	▲	sublinear a day ago \| parent [-]
		> translate the official IRS forms to a machine readable format The instructions for each form published by the IRS every year are already written by professional technical writers to be unambiguous. Do you mean that someone ought to write a simplified english grammar transpiler? I think that would genuinely be interesting. What's missing are the guidelines the technical writers are using, but that can probably be derived.