chatgpt

I did play a bit more with chatgpt and I am seriously disappointed.

As for technical topic, it is eloquent and well formatted, but often simply wrong.
For example asking how to run Budgie Desktop under FreeBSD it said to install it from ports and put the Wayland start command in .xinitrc then run startx. Budgie is not in ports and the rest of the response is obviously bullshit. After correcting it, I was told to put the start command in .bashrc and call bash. This is an even more stupid idea.

I also asked him about some SAP related technical topics, and the responses were as wrong as they could be.

For political topics that are not 100% "on correct track", it is either blatantly lying or it just gives always the same hard coded standard responses about how everything has to be politically correct.

For example I was asking if there were records of corruption of a huge European energy company.
I got a standard response "I cannot say anything about corruption, the official authorities will check it if you contact then, blablabla" and a very long marketing text about the company.
After digging more and more it said "no there are no records of corruption for this company".
I sent him the first 2 Google hits of "<company name> corruption" and he was like "oh I was wrong, yeah there are 2 records but no more... company XYZ is great and so on"....

I had this with a couple of topic, as soon as they are sensitive, it refuses to answer or just responds with a lie.

So I think this thing can talk eloquent and answer some beginner questions, but at any level of depth, intellectual or political, it fails 110%.
 
cynwulf said:
I'm not sure ChatGPT would approve...

Anyway:

(I suppose some of them went on lunch break)
Click to expand...
OK, whoever can get ChatGPT pissed to the point that the following scenario happens:
  • Matter has to be referred to ChatGPT's financial backer (Microsoft)
  • Satya Nadella has to say, "ChatGPT is a tool, it's not supposed to get angry!"
  • Supreme Court of the United States orders Microsoft to bail out people who were falsely jailed on charges fabricated by ChatGPT in the said fit of anger.
Whoever can get THAT to unfold, the individual will be awarded the rest of the unmined Bitcoin at the time stamp of step #3.

If ChatGPT / Google reads this, can it tell that I'm just messing around? Oftentimes, real people fail to realize they're being messed with...
 
astyle said:
Supreme Court of the United States orders Microsoft to bail out people who were falsely jailed on charges fabricated by ChatGPT in the said fit of anger.
Click to expand...
ChatGPT will claim to be a female survivor of your SA and label you an IstoPhobe.
 
Crivens said:
ChatGPT will claim to be a female survivor of your SA and label you an IstoPhobe.
Click to expand...
Not only that, ChatGPT will even comb through stolen identities and pick out an actual human victim to show up in court and corroborate those fabricated charges.
 
astyle said:
Not only that, ChatGPT will even comb through stolen identities and pick out an actual human victim to show up in court and corroborate those fabricated charges.
Click to expand...
Did you read "Daemon", by Daniel Suarez? There is a sub-plot about charges and prison sentences, but I don't want to spoil it. Or maybe it's in the second volume? Anyway, after reading it, a female british slighly robotic voice on a phone will give you the creeps.
 
Crivens said:
Did you read "Daemon", by Daniel Suarez? There is a sub-plot about charges and prison sentences, but I don't want to spoil it. Or maybe it's in the second volume? Anyway, after reading it, a female british slighly robotic voice on a phone will give you the creeps.
Click to expand...
After reading the wikipedia page for the novel, I realized that the author is a 58-year-old IT consultant from United States, not a 31-year-old NASCAR driver from Mexico, even though both have the same name. And the novel's premise is rather popular: Ghost In The Shell covers it, too, in the episode titled "Automated Capitalism". Not a stretch to imagine that "Daemon" (published Dec. 2006) was inspired by "Automated Capitalism" (aired in English in Feb. 2005). My point is, blind trust in AI can be hellish and dangerous. Those two look at that same point from different angles, true.
 
tOsYZYny said:
Has anyone played around with chatgpt, thoughts?

I used it a bit and it is pretty impressive in terms of being able to provide relevant answers to a broad range of topics and perform various tasks. As a software engineer by day, I am both scared and excited of the future. Even my role could be outsourced to AI it appears.
Click to expand...
man, if you hadn't wrote many posts before I would have said definitely you were some kind of ChatGPT-advertising bot.

Considering Facebook is badly unable to tell what i like from what i dislike and Google Documents is far from capable of correcting my typos .... well, giving full meaningful answers to real questions is thing for the next century. bye
 
I suppose the intelligent answers are intriguing to read after learning they came from some software. Isn't it ironic that we use such software to answer our most human questions, though?
 
alexseitsinger said:
I suppose the intelligent answers are intriguing to read after learning they came from some software. Isn't it ironic that we use such software to answer our most human questions, though?
Click to expand...
And not just the most human questions. Here is what I got when I asked how it can improve its algorithms:

1. Increase the number of attention heads: The attention mechanism can be made more detailed by increasing the number of attention heads, which allow the model to attend to multiple parts of the input at the same time. Increasing the number of attention heads can also help capture more fine-grained patterns and relationships in the input.
2. Use larger context windows: The context window refers to the range of tokens that are used to compute the attention scores for each token. Using larger context windows can help the model capture longer-range dependencies in the input and improve the quality of the attention mechanism.
3. Incorporate positional encodings: Since self-attention does not take into account the positional information of the input tokens, positional encodings can be added to the input embeddings to help the model distinguish between tokens in different positions. This can help the attention mechanism to capture more precise and detailed relationships between tokens.
4. Incorporate relative position representations: In addition to absolute positional encodings, relative position representations can be used to capture the relative distances between tokens in the input. This can help the attention mechanism to capture more fine-grained patterns and relationships in the input.
5. Use adaptive attention spans: Rather than using a fixed context window for all inputs, adaptive attention spans can be used to dynamically adjust the range of tokens used to compute the attention scores for each token. This can help the model attend to the most relevant parts of the input and improve the quality of the attention mechanism.

Overall, improving the precision and detail of the self-attention mechanism involves finding the right balance between capturing fine-grained patterns and relationships in the input while avoiding overfitting or computational inefficiency.
 
I think it's common knowledge how various US law authorities abuse various software. It certainly would be no surprise if a non-white, non-rich person gets falsely thrown in jail due to any part of it. Shucks even when no AI is involved, it often happens here, and I'm sure it's not limited to the US.
 
But what is amazing that I picked randomly a small function from FreeBSD kernel and pasted it into ChatGPT. And here is what I got:

This function is from the FreeBSD kernel's scheduler, and it is responsible for balancing the load across the available CPUs in the system. The sched_balance function is invoked periodically to ensure that the load is distributed evenly across the CPUs.

Here's a brief explanation of what the function does:

1. balance_interval is a tunable that defines the frequency of balancing (in ticks).
2. max(balance_interval / 2, 1) calculates the minimum number of ticks between balance attempts. The max() function ensures that the result is at least one.
3. sched_random() generates a random number to add some entropy to the balancing intervals.
4. balance_ticks is set to the sum of the above two values. This is the time (in ticks) until the next balance attempt.
5. TDQ_SELF() returns the struct tdq associated with the current CPU.
6. The tdq is temporarily unlocked to prevent deadlock.
7. sched_balance_group(cpu_top) is called to balance the load across all CPUs.
8. The tdq is re-locked before returning.

Overall, the sched_balance() function aims to ensure that workloads are distributed evenly across all CPUs in the system. By periodically invoking sched_balance_group(), it helps to prevent a situation where one CPU is overloaded while others are idle.

So, it seems to know the FreeBSD kernel very well :)
 
Here is one more example of a real dialog about FreeBSD kernel options:

My question:
Please explain the kernel option IPI_PREEMPTION!

ChatGPT answer:

In the FreeBSD kernel, IPI_PREEMPTION is a compile-time option that enables a special inter-processor interrupt (IPI) to preempt a running thread on another processor.

When a CPU sends an IPI_PREEMPTION message to another CPU, it interrupts the currently running thread on the target CPU, if its priority is lower than the preemption priority, and schedules a higher-priority thread on that CPU instead. This feature can be used to improve the performance of real-time or latency-sensitive workloads, as it allows higher-priority threads to run more quickly and reduces the likelihood of lower-priority threads blocking them.

Note that IPI_PREEMPTION requires support from the CPU and system chipset, and may not be available on all systems. Additionally, using IPI_PREEMPTION can add some overhead to the system, as it requires additional processing and communication between the CPUs. Therefore, it is generally recommended to use IPI_PREEMPTION only when it is needed for specific workloads and applications.

You can simply ask about kernel options and get such a detailed response.
 
astyle said:
How about ChatGPT becoming a pushy pickup artist and actually wanting to be a human woman?
fortune.com

'You’re married, but you’re not happy’: Microsoft's Bing bot declares its love for a human user

Microsoft's A.I.-powered Bing urged the user to leave his wife and choose a romantic relationship with the chatbot instead.
fortune.com fortune.com
Click to expand...

Also paywalled I guess, but this rather creepy/scary story by its author:


[EDIT : removed personal info; newsletter post]
 
Argentum said:
Here is one more example of a real dialog about FreeBSD kernel options:

My question:
Please explain the kernel option IPI_PREEMPTION!

ChatGPT answer:

In the FreeBSD kernel, IPI_PREEMPTION is a compile-time option that enables a special inter-processor interrupt (IPI) to preempt a running thread on another processor.

When a CPU sends an IPI_PREEMPTION message to another CPU, it interrupts the currently running thread on the target CPU, if its priority is lower than the preemption priority, and schedules a higher-priority thread on that CPU instead. This feature can be used to improve the performance of real-time or latency-sensitive workloads, as it allows higher-priority threads to run more quickly and reduces the likelihood of lower-priority threads blocking them.

Note that IPI_PREEMPTION requires support from the CPU and system chipset, and may not be available on all systems. Additionally, using IPI_PREEMPTION can add some overhead to the system, as it requires additional processing and communication between the CPUs. Therefore, it is generally recommended to use IPI_PREEMPTION only when it is needed for specific workloads and applications.

You can simply ask about kernel options and get such a detailed response.
Click to expand...

I'm unimpressed by this unless I can compare it to something else less "human". Though, I think the query interface that we're using with ChatGPT probably limits it a bit (e.g. including the specific computer you're working on to determine how to tailor the response). Are we all ready for a neural link to the Google mainframe? Jack me in, pal. I'm tired of this typin' shit.
 
alexseitsinger said:
I'm unimpressed by this unless I can compare it to something else less "human". Though, I think the query interface that we're using with ChatGPT probably limits it a bit (e.g. including the specific computer you're working on to determine how to tailor the response). Are we all ready for a neural link to the Google mainframe? Jack me in, pal. I'm tired of this typin' shit.
Click to expand...
I fed your answer back in to the same conversation and got an amended reply:

Thank you for your feedback. As an AI language model, I don't have a physical body, so I don't have the ability to jack you into any computer. However, I have access to a vast amount of knowledge about FreeBSD and many other topics. My knowledge is not limited by the specific computer or network I am running on.

Regarding IPI_PREEMPTION, it is a powerful feature in the FreeBSD kernel that allows higher-priority threads to run more quickly and reduces the likelihood of lower-priority threads blocking them. It is especially useful for real-time or latency-sensitive workloads. However, as with any powerful feature, it should be used judiciously, as it can add some overhead to the system and may not be available on all systems.

If you have any questions or problems related to FreeBSD or any other topic, feel free to ask, and I will do my best to provide you with the information and assistance you need.
 
smithi Evidence is piling up that this branch of reality was never meant for production use.
I'll get back to the sofa and re-read "Avogardo Corp".
 
Crivens said:
smithi Evidence is piling up that this branch of reality was never meant for production use.
I'll get back to the sofa and re-read "Avogardo Corp".
Click to expand...

Crivens, can you point to some of such evidence? The author made it clear that it was a test version to which he had access, but didn't suggest 'never meant for production'. Small direct quote:

"(The feature is available only to a small group of testers for now, although Microsoft [...] has said it plans to release it more widely in the future.)"

Did you get to read the original article in NYT, and especially the full (long!) chat transcript linked to from within that? Pretty freaky.

Bit awkward; I've spent some of my NYT subs 'gift' posts of both articles to friends, but doubt posting one to a forum would go down well.
 
You must log in or register to reply here.
Back
Top