I, ChatGPT

Gribnit

@Applied-Mediocrity said in I, ChatGPT:

companies have an aversion to selling things they don't understand and cannot test rigorously

Lies!

Gribnit

@remi said in I, ChatGPT:

hot by cold, though sometimes that was more "hot by hot" but it didn't work either

Well yeah. You need Water, Rock, or Ground.

Gribnit

@HardwareGeek said in I, ChatGPT:

@dkf said in I, ChatGPT:

@remi said in I, ChatGPT:

chemistry never managed to turn lead into gold

But particle physics is now able to explain exquisitely why that's a dumb thing to try to do (it's energetically unfavourable; leave that stuff to supernovae).

Dumb, but possible. I vaguely recall a physics experiment in college in which we exposed lead (?) to neutrons in a small reactor and produced (a few — very few — atoms of) gold. Or something like that; it's been a long time, but we definitely transmuted something, and gold was involved.

Lead-foil bombardment with alpha particles, iirc. You get short-lived gold isotopes. Everyone knows this.

Applied Mediocrity

@Gribnit said in I, ChatGPT:

@Applied-Mediocrity said in I, ChatGPT:

companies have an aversion to selling things they don't understand and cannot test rigorously

Lies!

Bor Gullet!

boomzilla

GOG

@boomzilla Honestly, the whole thing would work as a routine.

Bulb

@Zerosquare said in I, ChatGPT:

https://www.cell.com/fulltext/S1535-6108(02)00133-2

Chinese saying that it is difficult to find a black cat in a dark room, especially if there is no cat

LaoC

@HardwareGeek said in I, ChatGPT:

@LaoC said in I, ChatGPT:

CRT electrons have only about 30keV depending on the size of the tube.

IIRC (which I may not), TV CRTs are limited to less than ~25keV, because above that you get too many X-rays.

I know I took apart some that used 27 and they weren't the biggest on the market, hence my estimate. I suppose the bigger ones needed so much thicker glass for stability that they still stayed within the norms for x-rays.

Gern_Blaanston

Possibly fake, but retarded enough that it might be real:
.

loopback0

@Gern_Blaanston Bard is based on Excel?

Gern_Blaanston

@loopback0 said in I, ChatGPT:

@Gern_Blaanston Bard is based on Excel?

Gern_Blaanston

@Applied-Mediocrity said in I, ChatGPT:

@Zerosquare said in I, ChatGPT:
.... companies have an aversion to selling things they don't understand and cannot test rigorously

That has never been true, at least not in my lifetime

Carnage

@Gern_Blaanston said in I, ChatGPT:

@Applied-Mediocrity said in I, ChatGPT:

@Zerosquare said in I, ChatGPT:
.... companies have an aversion to selling things they don't understand and cannot test rigorously

That has never been true, at least not in my lifetime

Considering the proclivity of sales and marketing to sell things that don't exist, I don't see how not understanding should be a problem.

Zerosquare

@Carnage said in I, ChatGPT:

the proclivity of sales and marketing to sell things that don't exist

: They don't exist yet. By the way, the deadline is next Friday.

BernieTheBernie

@Zerosquare said in I, ChatGPT:

@Carnage said in I, ChatGPT:

the proclivity of sales and marketing to sell things that don't exist

: They don't exist yet. By the way, the deadline ~~is next~~ was last Friday.

boomzilla

Horace He on Twitter

It's a thread. Basically, it can't solve anything it hasn't already seen, but once it's seen it, it can reproduce it.

DogsB

@boomzilla still makes it better then most of the industry to be honest. I'm in the middle of a project that is about to make the exact same mistakes as last time. I'm still on the fence about moving on or watching the same car crash again.

Mar 25, 2023

How smartphone AI ‘enhancement’ is ruining our photos

Samsung's pin-sharp shots of the Moon risk harming trust in imagery

Telegraph is about a decade or three behind here. You haven't been able to reliably gauge if an image is even a facsimile of reality since the mid-nineties. The technology can just do it in real time now.

Carnage

@boomzilla said in I, ChatGPT:

It's a thread. Basically, it can't solve anything it hasn't already seen, but once it's seen it, it can reproduce it.

GPT is not a reasoning engine, it cannot reason or solve problems. It can statistically analyse input and pull a response that seems linked to that analasys. It must have seen the solution to provide it.
One way to think about it is as a lossy database that use natural language as input for queries.

boomzilla

@Carnage apparently. The pure volume of stuff it has and its ability to respond with natural language probably masks that.

topspin

@Carnage said in I, ChatGPT:

@boomzilla said in I, ChatGPT:

It's a thread. Basically, it can't solve anything it hasn't already seen, but once it's seen it, it can reproduce it.

GPT is not a reasoning engine, it cannot reason or solve problems. It can statistically analyse input and pull a response that seems linked to that analasys. It must have seen the solution to provide it.
One way to think about it is as a lossy database that use natural language as input for queries.

Yes, but on the flip side, as mentioned here and , the stuff it sometimes can solve is still very impressive for what is "only a language model".

LaoC

@boomzilla said in I, ChatGPT:

Horace He on Twitter

It's a thread. Basically, it can't solve anything it hasn't already seen, but once it's seen it, it can reproduce it.

dkf

@Bulb said in I, ChatGPT:

@sockpuppet7 said in I, ChatGPT:

@HardwareGeek @dkf would it be possible to convert gpt-4 model weights to physical logic gates and make something like a gpt-4 chip?

Application-specific chips for calculating neural networks do exist. One interesting approach is basically a flash memory where the cells are charged proportionally to the weights and then you just apply voltage according to the input and read the output from the orthogonal lines. Very fast, not very accurate.

"Very fast, not very accurate." is a wonderful summary of just about every analog approach to neural networks. Even your brain is actually basically digital (formally, it is event based) since the activation of synapses on the axon is essentially an all or nothing thing. Weights are probably just approximations to the number of neurotransmitter receptors or ion channels in the synapse area.

LaoC

@dkf said in I, ChatGPT:

@Bulb said in I, ChatGPT:

@sockpuppet7 said in I, ChatGPT:

@HardwareGeek @dkf would it be possible to convert gpt-4 model weights to physical logic gates and make something like a gpt-4 chip?

Application-specific chips for calculating neural networks do exist. One interesting approach is basically a flash memory where the cells are charged proportionally to the weights and then you just apply voltage according to the input and read the output from the orthogonal lines. Very fast, not very accurate.

"Very fast, not very accurate." is a wonderful summary of just about every analog approach to neural networks. Even your brain is actually basically digital (formally, it is event based) since the activation of synapses on the axon is essentially an all or nothing thing. Weights are probably just approximations to the number of neurotransmitter receptors or ion channels in the synapse area.

Guess we'(d!better use tomethinf elsf for textual applicationsthen.

GuyWhoKilledBear

@dkf said in I, ChatGPT:

@Bulb said in I, ChatGPT:

@sockpuppet7 said in I, ChatGPT:

@HardwareGeek @dkf would it be possible to convert gpt-4 model weights to physical logic gates and make something like a gpt-4 chip?

Application-specific chips for calculating neural networks do exist. One interesting approach is basically a flash memory where the cells are charged proportionally to the weights and then you just apply voltage according to the input and read the output from the orthogonal lines. Very fast, not very accurate.

"Very fast, not very accurate." is a wonderful summary of just about every analog approach to neural networks.

Also, the Things That Remind You of WTDWTFers Thread is .

Gern_Blaanston

Microsoft accidentally revealed why people don't trust tech companies

This really was an excellent way to put it.

The company recently launched something called Copilot.

Hark the words of Microsoft VP of Modern Work and Business Applications Jared Spataro: "Sometimes, Copilot will get it right. Other times, it will be usefully wrong"

Bulb

@dkf said in I, ChatGPT:

Even your brain is actually basically digital (formally, it is event based) since the activation of synapses on the axon is essentially an all or nothing thing. Weights are probably just approximations to the number of neurotransmitter receptors or ion channels in the synapse area.

Activation of each synapse is all or nothing, and the artificial neural networks follow that model. Whether executed in analog or digital hardware.

The weights come into play when determining whether the given combination of inputs—because each neuron, natural or artificial, has many input synapses—causes the output to be active or not. And the natural system does not do that digitally either.

Also while the synapsis is either active, and releases all the neurotransmitter, or not, and releases nothing, a lot of things in the body depend on frequency of the activation. Because unlike the artificial neural networks, where the signals are synchronized, in the natural networks they are not and the neurons will combine the signals also depending on how close they are in time.

dkf

@Bulb said in I, ChatGPT:

And the natural system does not do that digitally either.

Yes, but the vast wealth of complexity of an individual synapse is essentially of very little interest because it is highly non-linear; once things fire, you effectively get a burst of current sent down the dendrite; then nothing much happens until the refractory period is complete (which is mostly the time for the ion pumps to reestablish resting potentials). Weights correspond to the strength of the signal sent down the dendrite, and probably correlate to the number of receptors for neurotransmitters. Once a big enough signal reaches the main body of the neuron, the axon fires.

There's also the delay, which is comprised from things like the time for signals to move down axon and dendrites and for neurotransmitters to cross the inter-neuron gap. It's very much not constant! The same two neurons can be connected together multiple times with different delays, and this is important for detecting temporal patterns. And there can be inhibition in the tree of dendrites; this allows more complex neurons like cortical pyramidal cells to do non-trivial computation in their dendritic trees (signal XOR processing has been observed as a direct operation, IIRC). I wouldn't be at all surprised if this stuff is necessary for our reasoning capabilities, and damn few simulators are anywhere near taking it on. (Much of the brain is simpler than that, and does stuff like basic signal processing, pattern detection and homeostatic maintenance, all stuff that should be within the capabilities of at least some neural network systems.)

DogsB

Mar 27, 2023

Google’s ‘code red’ crisis grows as ChatGPT races ahead

Google Bard insiders are growing nervous as Microsoft's chatbot races ahead

It still amuses me that they need so many guardrails to prevent Tay. It's like we can’t develop anything that won't nuke us at the first opportunity.

Also interesting that quite a few of the normies in the comments are finding basic mistakes and have observed that google’s search has been going downhill for a few years.

cvi

We better get on with the times, We're now in the age of AI (content warning: slashdot).

Some predictions from Gates:

"Eventually your main way of controlling a computer will no longer be pointing and clicking or tapping on menus and dialogue boxes. Instead, you'll be able to write a request in plain English...."

"Advances in AI will enable the creation of a personal agent... It will see your latest emails, know about the meetings you attend, read what you read, and read the things you don't want to bother with."

"I think in the next five to 10 years, AI-driven software will finally deliver on the promise of revolutionizing the way people teach and learn. It will know your interests and your learning style so it can tailor content that will keep you engaged. It will measure your understanding, notice when you're losing interest, and understand what kind of motivation you respond to. It will give immediate feedback."

"AIs will dramatically accelerate the rate of medical breakthroughs. The amount of data in biology is very large, and it's hard for humans to keep track of all the ways that complex biological systems work. There is already software that can look at this data, infer what the pathways are, search for targets on pathogens, and design drugs accordingly. Some companies are working on cancer drugs that were developed this way."

AI will "help health-care workers make the most of their time by taking care of certain tasks for them — things like filing insurance claims, dealing with paperwork, and drafting notes from a doctor's visit. I expect that there will be a lot of innovation in this area.... AIs will even give patients the ability to do basic triage, get advice about how to deal with health problems, and decide whether they need to seek treatment."

Better get a working mic, because Windows 12 might require you to talk to it... (I'm betting they'll replace -partially- the ~~Control Pannel~~Settings App with a voice-driven configuration prompt.)

Zecc

@cvi They won't like what most people have to say...

topspin

@cvi said in I, ChatGPT:

AIs will even give patients the ability to do basic triage, get advice about how to deal with health problems, and decide whether they need to seek treatment

cvi

@topspin Well, it'll be better than Bing. If you claim to be feeling alright, Bing will argue with you that you are in fact not alright, and that you are unreliable and a bad user. And, well, ill. Mentally (in addition to physically).

JBert

ChatGPT has some weird strings it recognizes but just doesn't know or no longer knows what do with:

https://www.youtube.com/watch?v=WO2X3oZEJOA

loopback0

HardwareGeek

@GuyWhoKilledBear said in I, ChatGPT:

@dkf said in I, ChatGPT:

@Bulb said in I, ChatGPT:

@sockpuppet7 said in I, ChatGPT:

@HardwareGeek @dkf would it be possible to convert gpt-4 model weights to physical logic gates and make something like a gpt-4 chip?

Application-specific chips for calculating neural networks do exist. One interesting approach is basically a flash memory where the cells are charged proportionally to the weights and then you just apply voltage according to the input and read the output from the orthogonal lines. Very fast, not very accurate.

"Very fast, not very accurate." is a wonderful summary of just about every analog approach to neural networks.

Also, the Things That Remind You of WTDWTFers Thread is .

I don't agree that's an appropriate thread. {{Some|Most} of us|We} are neither fast nor accurate.

dcon

@loopback0 said in I, ChatGPT:

Reminds me of the translation bot we used to have here...

Watson

@Gern_Blaanston said in I, ChatGPT:

Microsoft accidentally revealed why people don't trust tech companies

This really was an excellent way to put it.

The company recently launched something called Copilot.

Hark the words of Microsoft VP of Modern Work and Business Applications Jared Spataro: "Sometimes, Copilot will get it right. Other times, it will be usefully wrong"

Another quote from the article:

The company recently launched something called Copilot. This is a lump of AI that's apparently trained for the job of taking the weight off your mind.

It's there to help steer you to your destination. It's there to free you to focus on steering your life. And it's there to help you land on the perfect version of you, the one that does more in order to, I don't know, be more.

A "usefully wrong" Copilot. We've got a term for that around here.

LaoC

@loopback0 said in I, ChatGPT:

It's getting a bit more mainstream now but we've been there for large parts of the daily news output for years.

dkf

@loopback0 said in I, ChatGPT:

Any resemblance between input and output is optional. Used to play games like that at school...

PleegWat

This morning we got a long email describing all the things we cannot do with ChatGPT or other such bots.

cvi

@PleegWat Well, is there anything interesting that you're still allowed to do? Or is it largely a blanket ban on using it?

PleegWat

@cvi Play with it, as long as we don't input any corporate, customer, or personal data. Including our own personal data. And don't use the output for anything work-related either.

So basically a ban.

dkf

@PleegWat Can you ask ChatGPT to analyse that email to determine what you can do with ChatGPT?

PleegWat

@dkf No, because I do not want to give chatgpt my phone number.

sockpuppet7

@Rhywden said in How to find "common" types of exceptions in C#?:

So, I've got this method call here:
T? result = await HttpClient.GetFromJsonAsync<T>(string url);
and now I'm trying to figure out which exceptions can be commonly thrown by this - i.e. "no network connection" versus "server does not answer" versus "404" versus "JSON parse failed" and so on.

Is there some way to easily discover the exception types or do I have to manually create every error condition I can think of?

I'll put chatgpt4's answer on this thread, as there is already people annoyed with chatgpt being thrown everywhere, and I didn't check if it for hallucinations:

HttpClient.GetFromJsonAsync<T> is an extension method provided by the System.Net.Http.Json namespace, which combines the functionality of HttpClient.GetAsync with JSON deserialization using System.Text.Json.JsonSerializer. It can throw various exceptions based on the issues encountered during the operation.

Here's a list of common exceptions that might be thrown:

HttpRequestException: This exception can be thrown when there's an issue with the underlying HTTP request, such as a network error, server not responding, or an invalid URL.
TaskCanceledException: This exception can be thrown when the request times out, which is typically caused by the server not responding.
JsonException: This exception is thrown when there's an issue with JSON deserialization, such as invalid JSON or a mismatch between the JSON data and the expected data model (T).
InvalidOperationException: This exception might be thrown if the method is called with incorrect arguments, or if there's an issue with the configuration of the HttpClient or JsonSerializer.

To handle these exceptions, you can use a try-catch block, for example:

try
{
    T? result = await HttpClient.GetFromJsonAsync<T>(url);
    // Process result
}
catch (HttpRequestException ex)
{
    // Handle issues with the underlying HTTP request
}
catch (TaskCanceledException ex)
{
    // Handle request timeout
}
catch (JsonException ex)
{
    // Handle JSON deserialization issues
}
catch (InvalidOperationException ex)
{
    // Handle incorrect usage or configuration issues
}
catch (Exception ex)
{
    // Handle any other unexpected exceptions
}

This should cover most of the common error scenarios you mentioned. However, note that it's not possible to catch every possible exception that might be thrown by the method or its dependencies. You might need to examine the documentation or source code of the libraries you're using to understand other possible exceptions.

sockpuppet7

@sockpuppet7 and this is bing precise mode, I wanted to try all modes but bing got offline (none of the modes are answering it just now):

HttpClient.GetFromJsonAsync<T> can throw a HttpRequestException when non-success status codes are returned from the server¹. In .NET 5.0 and later, you can inspect the StatusCode property of the HttpRequestException to determine the specific HTTP status code returned by the server¹.

Other exceptions that can be thrown include NotSupportedException when the content type is not supported and JsonException when there is an error deserializing the JSON response².

Is there anything else you would like to know?

Source: Conversation with Bing, 28/03/2023(1) c# - Using HttpClient.GetFromJsonAsync(), how to handle .... https://stackoverflow.com/questions/65383186/using-httpclient-getfromjsonasync-how-to-handle-httprequestexception-based-on Accessed 28/03/2023.
(2) Why HttpClient.GetFromJsonAsync doesn't throw an exception when the .... https://stackoverflow.com/questions/71269425/why-httpclient-getfromjsonasync-doesnt-throw-an-exception-when-the-response-is Accessed 28/03/2023.
(3) c# - What Exceptions can PostAsJsonAsync throw? - Stack Overflow. https://stackoverflow.com/questions/20802943/what-exceptions-can-postasjsonasync-throw Accessed 28/03/2023.
(4) HttpClient.GetFromJsonAsync<T> does not handle 204 returned by ASP.NET .... https://github.com/dotnet/aspnetcore/issues/43608 Accessed 28/03/2023.

sockpuppet7

bing seems to have implemented something to block you from repeating the same question (probably because people would insist when they are circumventing it's rules), but if I don't ask it right away, I got the answer in creative mode, and I think it changed enough to be worth posting here:

That's a good question. The HttpClient.GetFromJsonAsync<T> method is a convenient way to get a deserialized object from a JSON response, but it also has some limitations. One of them is that it does not expose the HttpResponseMessage object that contains the status code and other information about the response. Instead, it throws an HttpRequestException if the status code is not successful¹. However, this exception does not have a property that exposes the status code directly². You can either parse the exception message to get the status code, or use a different approach to get the HttpResponseMessage first and then read the JSON content from it¹. For example:

var response = await HttpClient.GetAsync(url);
if (response.IsSuccessStatusCode)
{
    var result = await response.Content.ReadFromJsonAsync<T>();
    // do something with result
}
else
{
    // handle different status codes
    switch (response.StatusCode)
    {
        case HttpStatusCode.NotFound:
            // do something
            break;
        case HttpStatusCode.BadRequest:
            // do something
            break;
        // etc.
    }
}

Another limitation of HttpClient.GetFromJsonAsync<T> is that it does not handle 204 No Content responses well. It will throw a JsonException if the response is empty³. You can either catch this exception and handle it separately, or use a different approach to check the content length before reading the JSON content³. For example:

var response = await HttpClient.GetAsync(url);
if (response.IsSuccessStatusCode)
{
    if (response.Content.Headers.ContentLength.GetValueOrDefault() > 0)
    {
        var result = await response.Content.ReadFromJsonAsync<T>();
        // do something with result
    }
    else
    {
        // handle empty response
    }
}
else
{
    // handle different status codes
}

If you want to learn more about HttpClient.GetFromJsonAsync<T> and its alternatives, you can check out these resources:

¹ c# - Using HttpClient.GetFromJsonAsync(), how to handle HttpRequestException based on HttpStatusCode without extra SendAsync calls? - Stack Overflow
² c# - What Exceptions can PostAsJsonAsync throw? - Stack Overflow
³ HttpClient.GetFromJsonAsync<T> does not handle 204 returned by ASP.NET Core Web API · Issue #43608 · dotnet/aspnetcore · GitHub

I hope this helps. 😊

Source: Conversation with Bing, 28/03/2023(1) c# - Using HttpClient.GetFromJsonAsync(), how to handle .... https://stackoverflow.com/questions/65383186/using-httpclient-getfromjsonasync-how-to-handle-httprequestexception-based-on Accessed 28/03/2023.
(2) Why HttpClient.GetFromJsonAsync doesn't throw an exception when the .... https://stackoverflow.com/questions/71269425/why-httpclient-getfromjsonasync-doesnt-throw-an-exception-when-the-response-is Accessed 28/03/2023.
(3) c# - What Exceptions can PostAsJsonAsync throw? - Stack Overflow. https://stackoverflow.com/questions/20802943/what-exceptions-can-postasjsonasync-throw Accessed 28/03/2023.
(4) HttpClient.GetFromJsonAsync<T> does not handle 204 returned by ASP.NET .... https://github.com/dotnet/aspnetcore/issues/43608 Accessed 28/03/2023.