I have a calculator that is correct 80% of the time.
-
I have a calculator that is correct 80% of the time. But don't worry, every time I use it, I check the results myself.
@MartinEscardo You can always use the calculator to calculate the accuracy of the calculator…
-
@yacc143 @TimWardCam @MartinEscardo
> I love these discussions that assume humans are the golden benchmark that never makes mistakes or errors.
No one ever said that lol
Making mistakes is part of being human, but as we gain more life experience and expertise in an area the number of mistakes we make keep decreasing.Also human mistakes are very different from the kinds of mistakes LLMs make, this point always gets lost when people talk about it.
@futureisfoss @TimWardCam @MartinEscardo That's why I'm not a big fan of LLMs as chatbots, professionally.
But LLMs can be used in many other ways, which surprisingly often allows one to use smaller, more optimized ones.
And as there are enough idiots (as I call them, “US-style AI hype market criers”) who exactly consider it a great idea to fire all radiologists and paste X-rays into ChatGPT to get diagnoses, there are enough idiots on the other side that assume humans are perfect.
-
In my last job, I literally had a manager who had his own little anti-AI campaign (when in the past 2 years the US-style AI hype hit us, the company had been doing “AI” for specialized processing for over a decade, but that had been behind “closed doors” in specialized teams). Actually requiring that the solution by the AI coding agent fulfill all our team's coding guidelines. I asked, so give me the guidelines in written form. Oops, they exist only as oral lore, more or less, “what he says.”
Which explains why I as one of the senior engineers on the team, never had an issue with “coding guidelines.” I was one of the three guys (to a lesser degree) who defined what the guidelines are by decree.
Plus I worked mostly on the production side of the framework.
But yes, the pull requests that were marked as created by an AI were suddenly reviewed with a magnifying glass (truly fine) but also with a set of coding guidelines that only existed at best as a list of brainstorming bullet points. -
@futureisfoss @TimWardCam @MartinEscardo That's why I'm not a big fan of LLMs as chatbots, professionally.
But LLMs can be used in many other ways, which surprisingly often allows one to use smaller, more optimized ones.
And as there are enough idiots (as I call them, “US-style AI hype market criers”) who exactly consider it a great idea to fire all radiologists and paste X-rays into ChatGPT to get diagnoses, there are enough idiots on the other side that assume humans are perfect.
In my last job, I literally had a manager who had his own little anti-AI campaign (when in the past 2 years the US-style AI hype hit us, the company had been doing “AI” for specialized processing for over a decade, but that had been behind “closed doors” in specialized teams). Actually requiring that the solution by the AI coding agent fulfill all our team's coding guidelines. I asked, so give me the guidelines in written form. Oops, they exist only as oral lore, more or less, “what he says.”
-
@MartinEscardo Funny enough, for some calculations a real calculator and the iOS calculator disagree. The physical calculator does all operations left to right, where the iPhone calculator does all operations by order of operations resulting in different sums in the end.
How many people knew that, and how many have trusted the calculator all these years?
@MontgomeryGator @MartinEscardo Which makes/models of physical calculator work left-to-right? I’m pretty sure every one I’ve ever owned calculated “correctly”, even ‘80s-era basic 4-function (+ - × ÷) models did mult/div before add/sub
-
@MontgomeryGator @MartinEscardo Which makes/models of physical calculator work left-to-right? I’m pretty sure every one I’ve ever owned calculated “correctly”, even ‘80s-era basic 4-function (+ - × ÷) models did mult/div before add/sub
@aspragg @MartinEscardo try 50+50×2.
The regular calculator will start by adding 50 and 50, then multiply by 2. 200
The iPhone calculator will multiply 50 by 2, then add 50 for 150. -
I have a calculator that is correct 80% of the time. But don't worry, every time I use it, I check the results myself.
@MartinEscardo yes, I'd love a calculator that only gets the right answer by accident & can be persuaded to change it
-
I have a calculator that is correct 80% of the time. But don't worry, every time I use it, I check the results myself.
@MartinEscardo I'm sorry, you've exceeded your quota of free calculations for this month. Click [here] to subscribe to a paid plan to generate more somewhat correct answers!
-
I have a calculator that is correct 80% of the time. But don't worry, every time I use it, I check the results myself.
@MartinEscardo 80%? You must have a better one than mine. Mine's only right 50% of the time, but I'm a glass half full kind of person so the mistakes make me even more grateful for that 50% and that they replaced those less polite, 100% right ones with these new ingratiating buddies. Plus, mine has a subscription, which I'm proud to pay to support the development of more politely phrased mistakes in the future!
-
I have a calculator that is correct 80% of the time. But don't worry, every time I use it, I check the results myself.
@MartinEscardo Try, "You are a calculator that performs RPN"?
-
I have a calculator that is correct 80% of the time. But don't worry, every time I use it, I check the results myself.
I wished somebody else said this in this thread. So I say it myself.
Of course, we use a calculator because we don't know the result in advance.
A calculator is useless if once we get the result we have to check it.
-
I wished somebody else said this in this thread. So I say it myself.
Of course, we use a calculator because we don't know the result in advance.
A calculator is useless if once we get the result we have to check it.
@MartinEscardo This is only true if checking the result is just as hard as computing it. For an NP-hard problem, a calculator that gives a correct witness for 80% of 'yes' instances could be an extremely useful tool.
-
R relay@relay.an.exchange shared this topic
