Why offtopic is going to hurt us all...

[eTobey]

Just a little reminder that is worth thinking about:
I guess many of you have already experienced, that google has become a b***. You will find anything, except what you were actually looking for. When you now add irrelevant text to topics, it will get even worse, even with just the forums search engine (which is as "useful" as google unfortunately)!

Not to mention the hassle already reading through a lot of irelevant text, that is either offtopic, or does not help to answer questions at all.

[ebastler]

Try that search without the + sign, just with the quotes to force the appearance of "spreadsheet". I don't think the plus sign is still supported in Google searches, and it seems to invalidate the markup with quotation marks.

Edit: Double-posting is also going to hurt us all... Why does a failed Google search need two threads?

[eTobey]

Edit: Double-posting is also going to hurt us all... Why does a failed Google search need two threads?

I removed any mentions about google in the other thread...   

[brucehoult]

The baton for such simple things has been passed on...

[schmitt trigger]

Indeed, Google as the search-engine-to-go option is rapidly diminishing.

Which means that revenue from searches will decrease.

However Google as a corporation, has diversified significantly, I don’t see them becoming irrelevant as Altavista.

[magic]

Great, an offtopic discussion in an offtopic thread about the harms of offtopic.

The only thing still missing at this point is somebody complaining about it

[Stray Electron]

Great, an offtopic discussion in an offtopic thread about the harms of offtopic.

The only thing still missing at this point is somebody complaining about it

   You just did, didn't you?   

[magic]

I had to, because the next obligatory step is somebody complaining about the complaining

[tom66]

For technical problems I'm finding myself using ChatGPT a lot more now.  Maybe around 25-30% of the time.  I would previously dive deep into the annals of Google, Stack Overflow etc to get an answer but I don't need to any more.  It's great.  Of course, you have to be wary of the LLM bullshitting you... but it's not like the internet isn't full of bullshit either.

[brucehoult]

For technical problems I'm finding myself using ChatGPT a lot more now.  Maybe around 25-30% of the time.  I would previously dive deep into the annals of Google, Stack Overflow etc to get an answer but I don't need to any more.  It's great.  Of course, you have to be wary of the LLM bullshitting you... but it's not like the internet isn't full of bullshit either.

Never try to get ChatGPT to do something original. It simply can't.

But looking up existing information, finding something that some human already wrote, somewhere (online), it's very good.

[K5_489]

I had to, because the next obligatory step is somebody complaining about the complaining

I gotchu, bro! 

Seriously???  No one cares about your whining, and we don't need to see it 

 

[T3sl4co1l]

Think about clustering of terms.  "Sheet" is too generic.  Get specific, "Excel" for example.  Most spreadsheets use common syntax so the formulas will also work.

Or if you just want some copypasta, spreadsheet data can be generated trivially in TSV format. Do it in your favorite language, or type it out by hand.

But you didn't even ask for formulas, and I guess you don't want formulas, you want an answer even more complete than that; direct generated data perhaps.  But what format would you even want, what does "pwm sweep" mean?  Is this sampled (time-series) data, perhaps 0 and 1 in cells?  GPT for its part, seems to be aware of your question's incompleteness.  But you are not--?

I wonder if there's some error in expectation, misaligned goal, here?

Is your perspective of, trying to ask Google itself, for such a formula, or output?  But Google doesn't know anything (except the specific things programmed onto it: certain keywords bring up special results, like dictionary/definitions, or calculator).  You're asking Google to find webpages that have such an answer; and such a specific answer has little purpose to exist on the internet (mind, that's not zero, and there are some very tedious/niche webpages out there, as can be seen from searching random numbers for example, and maybe there's pages out there offering such answers already), but even if that were the case, why aren't you searching for the data they contain?  Perhaps loading the query with "0 0 0 0 0 0 1 1 1"... isn't going to be very successful anyway, but it seems like you don't know what you're looking for in the first place, or don't know how to use a traditional search engine?

Tim

[RoGeorge]

Think about clustering of terms.  "Sheet" is too generic.  Get specific, "Excel" for example.  Most spreadsheets use common syntax so the formulas will also work.
...

Wrong topic, right?  Not to be confused with offtopic.

Was the reply meant for https://www.eevblog.com/forum/chat/create-arbitrary-waveform-values-with-spreadsheet/ by any chance?  (complaining about wrong topic kind of offtopic in a topic about offtopic wouldn't be offtopic, innit?) 

[T3sl4co1l]

Oh posh, take it to the topic council   

Tim

[Xena E]

My problem with Gurggle is that every 'original' non suggested search I do is met with a friggin' captcha screen.

(Now, thats how we whine and winge in the shires.)

Is that OT enough for ya?

 

X

[langwadt]

For technical problems I'm finding myself using ChatGPT a lot more now.  Maybe around 25-30% of the time.  I would previously dive deep into the annals of Google, Stack Overflow etc to get an answer but I don't need to any more.  It's great.  Of course, you have to be wary of the LLM bullshitting you... but it's not like the internet isn't full of bullshit either.

Never try to get ChatGPT to do something original. It simply can't.

But looking up existing information, finding something that some human already wrote, somewhere (online), it's very good.

so the odds of it being confident but dead wrong is pretty high ...

[magic]

My problem with Gurggle is that every 'original' non suggested search I do is met with a friggin' captcha screen.

Really?
How are you doing it?
Never happened to me.

[tggzzz]

For technical problems I'm finding myself using ChatGPT a lot more now.  Maybe around 25-30% of the time.  I would previously dive deep into the annals of Google, Stack Overflow etc to get an answer but I don't need to any more.  It's great.  Of course, you have to be wary of the LLM bullshitting you... but it's not like the internet isn't full of bullshit either.

Never try to get ChatGPT to do something original. It simply can't.

But looking up existing information, finding something that some human already wrote, somewhere (online), it's very good.

LLMs assemble plausible sequences of words into grammatically correct sentences. "Plausible" is some vague form of "average" of the words that have previously been found in similar relationship to each other and the question.

That is not "finding something that a human already wrote". It is how "LLM hallucinations" are created, e.g.  non-existent legal case law presented as evidence to a New York judge. Oops; career limiting

There are people that operate in the same way as LLMs: they are called bullshitters, politicians and CEOs.

Next problem already worrying people: what happens when LLMs ingest other LLM output hallucinations?

[tom66]

For technical problems I'm finding myself using ChatGPT a lot more now.  Maybe around 25-30% of the time.  I would previously dive deep into the annals of Google, Stack Overflow etc to get an answer but I don't need to any more.  It's great.  Of course, you have to be wary of the LLM bullshitting you... but it's not like the internet isn't full of bullshit either.

Never try to get ChatGPT to do something original. It simply can't.

But looking up existing information, finding something that some human already wrote, somewhere (online), it's very good.

LLMs assemble plausible sequences of words into grammatically correct sentences. "Plausible" is some vague form of "average" of the words that have previously been found in similar relationship to each other and the question.

That is not "finding something that a human already wrote". It is how "LLM hallucinations" are created, e.g.  non-existent legal case law presented as evidence to a New York judge. Oops; career limiting

There are people that operate in the same way as LLMs: they are called bullshitters, politicians and CEOs.

Next problem already worrying people: what happens when LLMs ingest other LLM output hallucinations?

We've been over this before I'm sure,  you refuse to acknowledge LLM's cannot produce anything useful,  I disagree.

Just the other day I needed to use a program to zoom and pan around an image without any blended interpolation algorithm, ie. nearest neighbour only.  I did some Googling and did not find an exact result.  Perhaps there is some way to get a Linux image viewer to do that but the 3-4 I tried did not have any way to turn off interpolation.  But, ChatGPT in literally 30 seconds created me a tool that did that in Python using Pygame.  It worked perfectly.  Yes, I could have written the same tool myself, but it would have taken probably 15-30 minutes of my day. 

Those who refuse to acknowledge the capability of LLMs are just being silly at this point.  They are clearly not purely bullshit machines.

Now yes, it's funny to ask it about things that obviously don't exist and observe the bullshit because the models do have issues in places... but on the other hand I've had plenty of bullshit from Google results, and the amount of SEO tuning now done by websites mean that it's very easy to find something that's completely wrong.  As an example, just search for the text in a Windows BSOD error message.  A solid 90% of the results from Google will be absolutely useless:  download this scammy tool to fix vague 'driver' issues,  install this scammy antivirus,  some random post on the Microsoft forum which suggests that rebooting will cure all ills.

[magic]

There are people that operate in the same way as LLMs: they are called bullshitters, politicians and CEOs.

Next problem already worrying people: what happens when LLMs ingest other LLM output hallucinations?

Expanding the analogy, it becomes something akin to a political party
Nothing we haven't dealt with before, but uglier than one LLM alone.

[tggzzz]

For technical problems I'm finding myself using ChatGPT a lot more now.  Maybe around 25-30% of the time.  I would previously dive deep into the annals of Google, Stack Overflow etc to get an answer but I don't need to any more.  It's great.  Of course, you have to be wary of the LLM bullshitting you... but it's not like the internet isn't full of bullshit either.

Never try to get ChatGPT to do something original. It simply can't.

But looking up existing information, finding something that some human already wrote, somewhere (online), it's very good.

LLMs assemble plausible sequences of words into grammatically correct sentences. "Plausible" is some vague form of "average" of the words that have previously been found in similar relationship to each other and the question.

That is not "finding something that a human already wrote". It is how "LLM hallucinations" are created, e.g.  non-existent legal case law presented as evidence to a New York judge. Oops; career limiting

There are people that operate in the same way as LLMs: they are called bullshitters, politicians and CEOs.

Next problem already worrying people: what happens when LLMs ingest other LLM output hallucinations?

We've been over this before I'm sure,  you refuse to acknowledge LLM's cannot produce anything useful,  I disagree.

Strawman argument. Please point to where I have said that

Just the other day I needed to use a program to zoom and pan around an image without any blended interpolation algorithm, ie. nearest neighbour only.  I did some Googling and did not find an exact result.  Perhaps there is some way to get a Linux image viewer to do that but the 3-4 I tried did not have any way to turn off interpolation.  But, ChatGPT in literally 30 seconds created me a tool that did that in Python using Pygame.  It worked perfectly.  Yes, I could have written the same tool myself, but it would have taken probably 15-30 minutes of my day. 

Those who refuse to acknowledge the capability of LLMs are just being silly at this point.  They are clearly not purely bullshit machines.

Now yes, it's funny to ask it about things that obviously don't exist and observe the bullshit because the models do have issues in places... but on the other hand I've had plenty of bullshit from Google results, and the amount of SEO tuning now done by websites mean that it's very easy to find something that's completely wrong.  As an example, just search for the text in a Windows BSOD error message.  A solid 90% of the results from Google will be absolutely useless:  download this scammy tool to fix vague 'driver' issues,  install this scammy antivirus,  some random post on the Microsoft forum which suggests that rebooting will cure all ills.

I have no reason  to doubt that happened. There is no conflict with what I wrote.

My statements stand unchanged.

If you wish to correct my statements, please do so. But it doesn't make you look good if you "correct" statements I haven't made.

[coppice]

For technical problems I'm finding myself using ChatGPT a lot more now.  Maybe around 25-30% of the time.  I would previously dive deep into the annals of Google, Stack Overflow etc to get an answer but I don't need to any more.  It's great.  Of course, you have to be wary of the LLM bullshitting you... but it's not like the internet isn't full of bullshit either.

I think its more realistic to say Google have trashed their system to the point where even a flaky piece of garbage, like ChatGPT, can do better.

[tom66]

Never try to get ChatGPT to do something original. It simply can't.

(emphasis mine, though this post was by brucehoult above, to be clear.)

If you wish to correct my statements, please do so. But it doesn't make you look good if you "correct" statements I haven't made.

This is the statement I am correcting.  What was created was original as far as any human is concerned, because you will not find anything close to that method available on GitHub et al.  Was it based on some statistical model of data somewhere?  Sure.  But so is what most programmers write because we all learn off other software engineers.  There is very little genuinely original code to write these days, and only so many ways to solve some problems.  The real intelligence of a programmer comes in solving unexpected problems and debugging systems.  Writing code is the easy part .

The question does come when novel problems are given to advanced LLMs, that do not have a solution in the dataset at all, how well do they cope?  Surprisingly well.  Not quite as good as an average software engineer, but better than a junior engineer.  These models are not yet available to the wider public, primarily because the compute power required is monstrous (and likely to remain so for some time) but they show that a model based on statistical text and graphic processing is capable, at least when tested, of novel reasoning.  Getting very close to AGI.  But sure, it's just predicting the next token.  So it can't be intelligent, right?

As I said, those who deny the capability of LLMs are just being silly because the actual results are stunning.  And the world had better figure out how to adapt.  In the next decade or so we will be overwhelmed with fake LLM generated prose (already we see spammers with obvious-GPT text, not too long before we see better versions of that), and many jobs will be replaced.  I think programmers are probably safe for now, but I'm keeping one eye on how things go, because there could be a lot of jobs shed when they get to the level of being able to replace junior/midlevel engineering roles.  On the other hand, new jobs are created by new technology all the time.  So, who knows. 

[tszaboo]

For technical problems I'm finding myself using ChatGPT a lot more now.  Maybe around 25-30% of the time.  I would previously dive deep into the annals of Google, Stack Overflow etc to get an answer but I don't need to any more.  It's great.  Of course, you have to be wary of the LLM bullshitting you... but it's not like the internet isn't full of bullshit either.

I think its more realistic to say Google have trashed their system to the point where even a flaky piece of garbage, like ChatGPT, can do better.

I don't think so. I asked that garbage LLM to give me some ICs with a specific feature. And it invented non-existing features for ICs with slightly similar functions. You ask it what's the second biggest capital in Europe, and it tells you that it's Istanbul. And you call it out on it's bullshit, it apologizes and it makes up some more bullshit.

[tggzzz]

Never try to get ChatGPT to do something original. It simply can't.

(emphasis mine)

Q.E.D.

I did not write that (emphasis mine). Bruce Holt wrote it in his post https://www.eevblog.com/forum/chat/why-offtopic-is-going-to-hurt-us-all/msg5667293/#msg5667293

If you wish to correct my statements, please do so. But it doesn't make you look good if you "correct" statements I haven't made.

This is the statement I am correcting.  What was created was original as far as any human is concerned, because you will not find anything close to that method available on GitHub et al.  Was it based on some statistical model of data somewhere?  Sure.  But so is what most programmers write because we all learn off other software engineers.  There is very little genuinely original code to write these days, and only so many ways to solve some problems.  The real intelligence of a programmer comes in solving unexpected problems and debugging systems.  Writing code is the easy part .

The question does come when novel problems are given to advanced LLMs, that do not have a solution in the dataset at all, how well do they cope?  Surprisingly well.  Not quite as good as an average software engineer, but better than a junior engineer.  These models are not yet available to the wider public, primarily because the compute power required is monstrous (and likely to remain so for some time) but they show that a model based on statistical text and graphic processing is capable, at least when tested, of novel reasoning.  Getting very close to AGI.  But sure, it's just predicting the next token.  So it can't be intelligent, right?

As I said, those who deny the capability of LLMs are just being silly because the actual results are stunning.  And the world had better figure out how to adapt.  In the next decade or so we will be overwhelmed with fake LLM generated prose (already we see spammers with obvious-GPT text, not too long before we see better versions of that), and many jobs will be replaced.  I think programmers are probably safe for now, but I'm keeping one eye on how things go, because there could be a lot of jobs shed when they get to the level of being able to replace junior/midlevel engineering roles.  On the other hand, new jobs are created by new technology all the time.  So, who knows.

To repeat: if you wish to correct my statements, please do so. But it doesn't make you look good if you "correct" statements I haven't made.

The points I made accurately describe how LLMs work.

Your lack of understanding completely ignores the phenomenon of LLM hallucinations, both their existence and cause.

Conventional searches do, of course, pull up lots of incorrect/unhelpful information that has been previously written, e.g. a search result could link to your post. Conventional searches wouldn't invent a post that you hadn't made; LLMs do do that kind of thing[1] - and nobody knows how to prevent that.

[1] infamous example: a NY lawyer used an LLM to write documents presented to a judge which included case law favourable to their client. The LLM apparently explicitly assured the lawyer that case laws were correct! The opposing party's lawyers could not find the cited cases because they did not exist; the LLM had invented/hallucinated them. The judge was most unimpressed, and the lawyer is in trouble. FFI, see comp.risks and elsewhere.