Google's AI is still broken

[Uberhoust13 3]

I used to be quite against AI but now we are more or less required to use it at work, and I am finding I like it. It is never 100% right, but for some stuff it is pretty good.

For general lookups I find the default MS CoPilot to work pretty good most of the time, I believe this to be based off of one of the Chat GPT versions, but not sure.

For work I find Claude Sonnet 4.5 to be the best so far (I believe this agent is primarily for software development):
1. It can do all the boiler plate code for starting a project.
2. It is very good at find compilation errors
3. It has been trained with a lot of known computational and geometric algorithms.
4. You can guide it, i.e. tell it to stay within a particular version and it mostly does.

Overall I find that I get the best results if I am very specific about what I want and how I want it done. If I leave the requests open ended it will usually generate something that works and will compile, but is difficult to build on.

Overall it is a statistical tool, if it is trained on the internet data without good guidance it will be as messed up as the internet is. But if trained with good data and information it can be quite powerful.

 

[TwoTankAmin57]

Btw, I do not believe that AI "helps to protect my privacy." I believe the exact opposite is the case. The real problem with AI is it has no conscience.

Conscience is an inner moral faculty, sense, or feeling that evaluates the rightness or wrongness of one's own actions, thoughts, and motives. It acts as a self-regulating mechanism, often inducing feelings of guilt, remorse, or approval. Key synonyms include moral sense, inner voice, scruples, and principles.

 

[gimme301]

I won't link to it here since politics are verboten but some group of geniuses came up with AI generated music that you'd never know was "fake" if you heard it on the radio.
Does it lack soul? Maybe. The 'singer' certainly sounds human. But it's not the kind of music where that particular attribute is apparent. (Hard country rap)
Turns out I really like it, and I'm not a rap fan. Otherwise I'd prefer AI be limited to roombas.

 

[seangee39 19]

I used to be quite against AI but now we are more or less required to use it at work, and I am finding I like it. It is never 100% right, but for some stuff it is pretty good.

For general lookups I find the default MS CoPilot to work pretty good most of the time, I believe this to be based off of one of the Chat GPT versions, but not sure.

For work I find Claude Sonnet 4.5 to be the best so far (I believe this agent is primarily for software development):
1. It can do all the boiler plate code for starting a project.
2. It is very good at find compilation errors
3. It has been trained with a lot of known computational and geometric algorithms.
4. You can guide it, i.e. tell it to stay within a particular version and it mostly does.

Overall I find that I get the best results if I am very specific about what I want and how I want it done. If I leave the requests open ended it will usually generate something that works and will compile, but is difficult to build on.

Overall it is a statistical tool, if it is trained on the internet data without good guidance it will be as messed up as the internet is. But if trained with good data and information it can be quite powerful.

I'm just doing my first deep dive into agentic ai similalry using Cursor. I was stunned at how much I was able to achieve from (and on) day 1. You do have to be very specific in the guidance. I am actually using copilot to help in creating the prompts as thats a much cheaper option. The frustrations of copilot / chatgpt etc making mistakes and having to debug their solutions are gone using good old tdd.

• Define the need with architectural guidance and guardrails
  • Explicitly limit what you want it to do - do this and nothing else!
• Write the tests
• Now write code to make the tests pass.
• Document what you just did

It still makes lots of mistakes and bad assumptions - but having the tests in place means it has to fix its own work . Its years (ok decades) since I got my hands properly dirty with code but I'm quiete enjoying this.
We're nowhere close to "Computer make me a programme to do blah blah blah" - but if you know how to do it yourself you can get the heavy lifting done incredibly quickly.

My quote of the week was

You won't lose your job to AI - but you might lose your job to somebody that knows how to use AI

Sadly many managers / execs do just see it as a way of reducing head count - and costs.

 

[TwoTankAmin57]

@gimme30

Let me know when you will be going to see the AI musicians performs live. How much would you pay to see this?

One day not only will the musicians be AI so too will be the audience.



(edited to fix a typo)

 

[anewbie]

On a totally unrelated question around leeching this is what google ai told me:
Why You Should Not Use "Straight" RO Water
Lack of Minerals: RO water is almost entirely pure H2O. It lacks essential nutrients like calcium and magnesium that fish need for bone and skin development.

My poor fishes are going to the great sky god if i follow these instructions.

 

[gimme301]

By "straight" if it meant "nothing but" then yeah, technically it's correct. Could be misleading to a newb who didn't know better though.

Let me know when you will be going to see the AI musicians performs live. How much would you pay to see this?

I'll send my robot when they come up with AI Metallica. Synthetic beings get 10% off.

 

[jaylach68 184]

This Malwarebytes article should raise some eyebrows...

Wikipedia's AI agent row likely just the beginning of the bot-ocalypse

An AI agent was banned from editing Wikipedia pages... and that's when things got weird, with the agent publishing its complaints publicly.

www.malwarebytes.com

 

[anewbie]

This Malwarebytes article should raise some eyebrows...

Wikipedia's AI agent row likely just the beginning of the bot-ocalypse

An AI agent was banned from editing Wikipedia pages... and that's when things got weird, with the agent publishing its complaints publicly.

www.malwarebytes.com

Stupid AI I told it no but it insist the correct answer is yes.

 

[SammyTargo]

AI is improving at a rapid rate, but I still find it makes mistakes. Some people put it in charge of certain things for their business, stock portfolio, school assignments, or work assignments... I find this to be stupid, not the AI. AI gets things wrong sometimes, it happens less and less but will most likely never have a 0% chance of asking AI something and it tells you false information.

A guy who I am acquainted with, who is strongly against AI once told me, "A parrot hears someone say something very knowledgeable about stocks, and it repeats it. When it repeats it, another person hears the parrot and decides to let the parrot run it's stock portfolio. Now the parrot hears something that is blatantly wrong about the stock market, and repeats it (which in this scenario executes a buy) while under control of this person's portfolio. Now replace the parrot with AI, would you let it run your portfolio?" In this case, "the parrot" clearly didn't lie when it said that misinformation, it simply did not know between the truth and the lie but rather just to say it. You cannot blame "the parrot," as it isn't lying to you.

You cannot blame AI for its poor knowledge, but under the same thoughts, just because it isn't technically lying, it still cannot be trusted enough to run an important role in one's life; when asking it questions or conversing, it is important to always check the facts. It can still be a great tool in theory.

The added bonus issue that I feel this community may feel closer-oriented towards is the water issue; how massive AI computers like OpenAI (ChatGPT, Sora, etc.) and Gemini heat up causing the casings to require extremely large amounts of cooling and therefore evaporating water at a rapid rate. The fish we love have natural homes too where they come from, whether they were taken directly from their homes or many generations ago they had lived there. Regardless of the specifics, we are slowly but surely evaporating and destroying their homes to create AI slop that mushes peoples brains like worms being stepped on during a rainfall. I don't use AI as much as others I know, but I still feel I use AI too much.

A bonus tip for people who don't care about that "AI Overview" that google gives: Putting "-ai" or "-i" searches up but does not waste the water that would be wasted if the search would be giving you a useless/unused AI Overview. The message should be at the end of your search, e.g., "How to breed killifish -ai" the search is also not case sensitive so you can say "-AI' or "-I" if you choose.

 

[GaryE29 27]

For now, I skip most of the AI overviews google comes up with, and try to go to better sources. The AI is getting better with the info it offers. When I see something other than predictable cliches coming out of it, I'll probably make use of the overviews. It will improve as a tool - we can already see it happening. Unlike with a researcher though, if we disagree with it, there's no discussion.

Whether it works or not isn't the issue for me - I focus more on who it works for and who it works against. That's an entire other debate.

One of the few surviving old growth forests in my region, about 4 km from my door is going to be leveled for a data centre. There are no studies on how it will affect local temperatures, but it's certainly going to use a lot of water and electricity while providing very few local jobs. These are the issues I look at.

 

[sharkweek178]

I was helping a 13 year old with his homework the other day. And this is out of control. He took a picture of the assignment and used an AI to get all the answers. Didn't even type the questions in. Just took a pic and that was all he needed to do.
I think that even if he searched them out online, that would still be preferable to this.

 

[jaylach68 184]

I was helping a 13 year old with his homework the other day. And this is out of control. He took a picture of the assignment and used an AI to get all the answers. Didn't even type the questions in. Just took a pic and that was all he needed to do.
I think that even if he searched them out online, that would still be preferable to this.

Reminds me of one of my daughters. This was in the early-mid 1980s. She was in 1st grade and one day her teacher called saying that she thought we were giving too much help with homework as her homework was always perfect in math but she was having issues with tests. It happens that she was always the first one up in the morning and I had a suspicion and caught her. Even then I had an early computer that included a cartridge slot. She, on her own, figured out that if she plugged the B.A.S.C. language cartridge and told it to print 3+7 it would not put out 3+7 but the answer , 10. My ex wanted me to forbid her from using the computer totally but I went in another direction. I started teaching her computers. Of course I included that homework was not to be done on the computer if she wanted to learn. Still I was impressed with what she had done and taught her quite a bit. I mean a 6 year old kid figuring, on her own, how to use the computer for homework was genius.

Years later and this little girl won scholarships that gave her an entirely free college education. I truly believe that my directing her in the right direction did a lot more for her than taking away the computer.

What I guess I'm trying to say is to not restrict a child that shows initiative but, rather, direct that in the right direction. Of course, in your case, you have to stop the AI usage but don't punish. Respect the smarts needed to figure things out at a young age and offer a better option.

 

[anewbie]

I was helping a 13 year old with his homework the other day. And this is out of control. He took a picture of the assignment and used an AI to get all the answers. Didn't even type the questions in. Just took a pic and that was all he needed to do.
I think that even if he searched them out online, that would still be preferable to this.

And that my son is why we have stupid kids - and i at least in usa scores are at their lowest and even those with good grades their raw ability to think stinks.

 

[jaylach68 184]

And that my son is why we have stupid kids - and i at least in usa scores are at their lowest and even those with good grades their raw ability to think stinks.

We DO NOT have stupid kids any more than when we were kids. We have kids that are in turmoil and don't know what to do. Today's kids are not stupid but many are ignorant. There is a big difference. Ignorance means that you don't have the knowledge. Stupid means that you have no interest in gaining knowledge. Most kids want to gain knowledge but, as adults, we are too busy with our lives to give from the heart what they really need.

At 71 my generation is largely to blame for this as it is my generation that largely took away the right of parents and teachers to discipline a child. Without being disciplined how can we expect a child to have discipline within themselves?

It is a sad situation when a child has no one to look to for guidance but, sadly, that is too often the case. The result is that today's kids look more to peers than their parents as the parents are just not there to give guidance. I don't put down today's kids; I pity them.

If I may I will put out a plea. Don't just feel that you know everything as you don't. Take time to actually sit and listen to your kid's issues. You MAY find that they are not really against you but just afraid as you are not there when needed.

OK, rant over but think about what I've said. The real problem is not our kids but ourselves.