I didn't know if I was training OpenAI's GPT models or not. Turns out I was in fact training Google's Gemini model.

I'm not sure what the relationship between the company I'm working with and Google are (if Google owns the company or if Google simply outsourced training the AI to them), but I'm at least indirectly working for Google.


And side-note: I can't edit my profile on here. This site is still buggy huh.

Eventually, there will be some lawsuits that seek compensation for all this unpaid "work".

This will:
- make some lawyers very rich.
- get everyone else a $1 coupon to spend at McDonald's.


You might be able to edit your profile via the legacy site.
https://legacy.cplusplus.com/user/

I can't edit my profile on here.

Click your user name on the left side menu setup. That should let you change user account settings. It works with the new or legacy site.

There aren't a lot of settings you can change. User name, email, password setting and user profile.

At least it appears to work, I actually haven't tried to change any setting except user name and that was a while back.

You might be able to edit your profile via the legacy site

Click your user name on the left side menu setup. That should let you change user account settings

Nothing works unfortunately:(

Eventually, there will be some lawsuits that seek compensation for all this unpaid "work"

Well actually it does pay, quite well actually..

That sort of subcontracting happens all the time. Companies take on projects that are too big for them and rather than hiring or contracting people they subcontract some other company to work on some specific module or functionality. One time I was involved in a project related to a provincial lottery, and we were something like fourth in the subcontract chain. We ourselves subcontracted another company to work on the front-end code. Based on how much work we did, I'm pretty sure all the ones that were above us were just skimming off the top and didn't allocate a single person on the project. I wouldn't be surprised if the whole thing was 60% inefficient. Well, at least it was for the province of Santa Fe, so I didn't pay for it (but I would imagine that sort of thing happens all over the place).

I'm pretty sure all the ones that were above us were just skimming off the top and didn't allocate a single person on the project

One expensive finders fee..

The interesting part is they won't tell us which AI model we're actually training (either for secrecy, to keep us from becoming bias, or maybe both). However, it became increasing obvious that it was Gemini.

I need to catch up on the AI front. I am not even sure if the 'new' crap is just the same old crap with more hardware thrown at it or something radically better.

All I ask for is that they continue to suck when they call you for whatever scam of the day. If they get smarter, it will be harder to confuse them and cause problems.

Nothing works unfortunately:(

I was able to change my user name, I did it via the legacy site.

I access the entire site, not just the fora, via the legacy feature.

Sure, it looks retro, but it is easier for my aged eyes to read stuff. One of the drawbacks of being 65+, and having had lens replacement surgery done on both eyes to get rid of cataracts.

I do believe the Public Profile feature is currently disabled/inactive/not working, same as the new user creation feature.

Honestly I prefer to remain any-mouse for the most part. Even to myself. If I don't know who I am, the better.

All I ask for is that they continue to suck when they call you for whatever scam of the day. If they get smarter, it will be harder to confuse them and cause problems.

AI is pretty amazing. They imitate voices to a fairly well degree, generate videos/pictures that are realistic, etc..

I do believe the Public Profile feature is currently disabled/inactive/not working, same as the new user creation feature.

Yea that's what I'm trying to change :( Unfortunate

Yeah, computers are finally capable enough to show us how Turing would look like if he was a furry thinking about picking us up.

https://imgur.com/a/S6zxkdz

😳

AI is pretty amazing. They imitate voices to a fairly well degree, generate videos/pictures that are realistic, etc..

Yea. But they are not yet handling nonsense well. If you say like "pink sausage in a penguin" as a response to "do you want to buy medicare from some random guy who can't speak english" it really can't handle it yet. It tends to just repeat the question or tell you to repeat your answer.

Well, to be fair, the average scammer probably wouldn't handle that very well either.

I remember trying to figure out how to calculate the torque needed in a particular situation. I didn't know what formulas to apply or anything, was way outside my scope of physics knowledge. I went onto two different forums (including physics/engineering exchange), each giving me a kinda-of answer and nothing solid.

This was before AI, so I decided to ask a few AIs. They both figured it out, gave me the formula, the answer, everything. A week of asking real people that went poorly could have been replaced with a single query that would have taken a minute.

On the other hand, humans will usually tell you their level of confidence in their answers. AIs will make shit up and pretend they're not. You can't use anything they say without checking it by other means.

helios wrote:
You can't use anything they say without checking it by other means.

Yeah, on another topic, I encountered problems getting ChatGPT to write C++ code. It seems fine if there are no gotchas in the methodology, but really one needs to know what they are doing in the first place, in order to check the result.

An example of a gotcha is in the Sine Rule, if the angle one is trying to work out is greater than 90 degrees, it returns a value less than 90. For example, what should be 120 degrees returns 60 degrees.

Yeah, the other day I wanted to test its capabilities and gave it a problem I was working on some weeks earlier: write a function that takes two 64-bit integers and returns their product encoded into two 64-bit integers.
I tried it three times. The first one was subtly incorrect. The other two were correct (AFAICT), but it took me like twenty minutes to figure out the first one because the carry logic was much different from what I had done. If I have to carefully review the code, I may as well just write it myself. The only way I could see this being helpful to someone is if they have absolutely no idea how to do it, so it's either rely on the AI or nothing.

An example of a gotcha is in the Sine Rule, if the angle one is trying to work out is greater than 90 degrees, it returns a value less than 90. For example, what should be 120 degrees returns 60 degrees.

Hmmm. IF you are using the Sine Rule to calculate an angle then you need the length of the opposite side to that required and one other angle and the length of the side opposite it (or the diameter of the circumscribed circle). You end up with:



However this is not a function as this can have more than one value. eg Any value of x >= 0 and x < 1 will give 2 values in the range 0 to 180. The sine of 120 and the sine of 60 are both (root 3) / 2. So the inverse sine of (root 3) / 2 will have two correct results. For the sin^-1 function which can only return 1 value this will be the principal value. In this case this is 60 which is what you obtained. If you need the non-principal value then you need to determine all possible valid values (the general solution) and then using other data determine the required value.

If I have to carefully review the code, I may as well just write it myself. The only way I could see this being helpful to someone is if they have absolutely no idea how to do it, so it's either rely on the AI or nothing.

Also useful if you know what the output should be - which you can actually use the AI for too. You can ask it to write the code and some unit tests to check normal and edge cases. Then you'd just be able to run it and check that it passed the tests.

That step is almost required for any code that isn't easy to check by glossing over.


AI also just has made it so easy to jump into new languages. I don't have to look up documentation that's full of jargon, I can just ask it and have the answer with a code example that relates to my situation.

There's plenty of uses for AI - I'm just waiting for it to be able to interact with the OS. Then you can ask it stuff like, "rename every file in this folder..", "generate a text file with 50000 characters", "use program X to do Y", etc.. That would be infinitely useful.

@seeplus

I was aware of all those things, but thing is the difficulty in getting ChatGPT to write the code.


@zapshe

I like the idea of getting it to do unit tests.

zapshe wrote:
.... - I'm just waiting for it to be able to interact with the OS.

Well one can ask it to write bash scripts, not sure if it can do PowerShell.

For some things AI works rather well, I asked it to produce C++ code that implements a 7 parameter Helmert Transformation, and it worked just fine because there are no gotchas. I guess some other things that might work well would be look up tables in the code.

I wonder if asking the AI to find an existing library might be a better strategy than getting it to write the code. Or maybe: "Use this library to implement the code:" ?