[yor_on]

I don't think so, no more than expecting people with different ideologies, upbringings, ideals, values and beliefs to be able to convince each other that only one of them have the truth. It's easier with pure logic, but logic is just one small part of what those AI's will meet.

Although you can use logic to test ideas, and use references to crosscheck differing views with facts, well, as far as possible, even logic has limitations.

[hamdani yusuf]

I don't think so, no more than expecting people with different ideologies, upbringings, ideals, values and beliefs to be able to convince each other that only one of them have the truth. It's easier with pure logic, but logic is just one small part of what those AI's will meet.

I mentioned GAN to specifically answer your concern about inability to question itself. For it to work, it needs access to the ground truths. For games like chess and go, the AI agents need to know all allowed moves, and conditions required for a win. They may not be entirely available in real life problems.

[Zer0]

Logic is Not Everything!

If I'm playing Chess against a 6yr young on her Birthday...
I might be Undecided on which moves i shall play..
But certainly Not Indecisive on the End result of the Game.

Hopefully, it shall come under International Laws for an A.I. system to confirm it's identity as an AI.

ps - Not Everybody is into Bigger, Faster, Better etc.
Some are Nostalgic & prefer the old skool ways.

[Karen W.]

Here's a first hand information intended for general audience.

The Inside Story of ChatGPT’s Astonishing Potential | Greg Brockman | TED

In a talk from the cutting edge of technology, OpenAI cofounder Greg Brockman explores the underlying design principles of ChatGPT and demos some mind-blowing, unreleased plug-ins for the chatbot that sent shockwaves across the world. After the talk, head of TED Chris Anderson joins Brockman to dig into the timeline of ChatGPT's development and get Brockman's take on the risks, raised by many in the tech industry and beyond, of releasing such a powerful tool into the world.

Thank you... That's very Interesting and quite exciting really. My friend was just telling me about this and I find it fascinating!

[hamdani yusuf]

ChatGPT is getting better quickly.

OpenAI?s ChatGPT Nails 150+ Difficult Tasks!
The paper "Reflexion: Language Agents with Verbal Reinforcement Learning" is available here:
https://arxiv.org/abs/2303.11366

[hamdani yusuf]

Here's how it works from a more hardware oriented perspective. It's valuable to learn from a first hand information.

What runs ChatGPT? Inside Microsoft's AI supercomputer | Featuring Mark Russinovich

Get an inside look at the AI supercomputer infrastructure built to run ChatGPT and other large language models, and see how to leverage it for your workloads in Azure, at any scale.

Go behind the scenes:
-For how we collaborated with NVIDIA to deliver purpose-built AI infrastructure with NVIDIA GPUs
-How Project Forge checkpointing works to restore job states if a long training job fails or needs to be migrated
-How we used LoRA fine-tuning to update a fraction of the base model for more training throughput and smaller checkpoints
-How UK-based company, Wayve, is using Azure's AI supercomputer infrastructure for self-driving cars
-And how Confidential Computing works with Azure AI to combine datasets without sharing personally identifiable information for secure multiparty collaborations.

Mark Russinovich, Azure CTO, joins Jeremy Chapman to break it down.

► QUICK LINKS:
00:00 - Introduction
01:15 - AI innovation building specialized hardware and software
04:22 - Optimizing hardware
05:40 - Improved throughput
06:17 - Project Forge
08:01 - Project Forge checkpointing demo
10:02 - LoRA fine tuning
11:29 - Use AI supercomputer infrastructure for your workloads
12:34 - How Wayve is leveraging AI supercomputer infrastructure
​​13:47 - How Confidential Computing works with Azure AI
15:21 - Wrap up

► Link References:

Leverage Azure AI capabilities for yourself at https://aka.ms/AzureAIInfrastructure

[alancalverd]

If a robot rips off your arm instead of washing your face (like a human, it's quite capable of doing either) who do you sue?

It depends on the circumstances.
Am I the one who designed and was testing the robot?

no

Has the robot been tested and passed regulations?
Was it used according to designated functions?
Was it properly maintained?

and yes to the rest. Now please answer the question.

[hamdani yusuf]

If a robot rips off your arm instead of washing your face (like a human, it's quite capable of doing either) who do you sue?

It depends on the circumstances.
Am I the one who designed and was testing the robot?

no

Has the robot been tested and passed regulations?
Was it used according to designated functions?
Was it properly maintained?

and yes to the rest. Now please answer the question.

It means that the regulations aren't strong enough. I can sue the tester or regulator. Or blame myself for trusting them.
But don't forget that human workers also have non-zero probability to do crazy things. They can burn down entire city. Who would you sue in that case?

[alancalverd]

I can sue the tester or regulator.

You might sue the tester if you had reason to believe that the test results were falsified, but your case would fail if the incident was outside the scope of the test parameters, or if it was within the scope but the machine had performed correctly on test. Regulation isn't a guarantee of safety or performance - its function is to create a common market.

But don't forget that human workers also have non-zero probability to do crazy things. They can burn down entire city. Who would you sue in that case?

You sue the person who did the deed. That's the law.

Boeing have offered compensation over the 737MAX disasters, because the entire system was designed in-house to overcome a "problem" resulting from their own design. But it was a simple, dumb system designed to consistently convert one input into one output. Who is liable for  the actions of a system developed by countless others with the intention of learning and reprogramming itself, years after anyone who might be considered the originator has died or gone bankrupt?

[hamdani yusuf]

You sue the person who did the deed. That's the law.

What if they commit suicide afterwards?

[hamdani yusuf]

You might sue the tester if you had reason to believe that the test results were falsified, but your case would fail if the incident was outside the scope of the test parameters, or if it was within the scope but the machine had performed correctly on test. Regulation isn't a guarantee of safety or performance - its function is to create a common market.

Then I can only blame myself.

[hamdani yusuf]

Who is liable for  the actions of a system developed by countless others with the intention of learning and reprogramming itself, years after anyone who might be considered the originator has died or gone bankrupt?

That would leave the users to be responsible for their decisions, just like normal investments. There's no guarantee that your investment will get to break even. I heard that 80% of start ups fail.
They must consider cost and benefits, as well as their probability to fail or succeed. If they decided to not use AI, they are in a risk of being out competedcompeted and lose their money.

[alancalverd]

So every "care robot" or selfdriving car must come with a clear warning "This machine might decide to kill you. The supplier accepts no responsibility." Like cigarettes, but with consciousness!

[alancalverd]

You sue the person who did the deed. That's the law.

What if they commit suicide afterwards?

In the first instance, you (or your executors) sue their estate. There have been more complex cases where the manufacturer of a single product has gone bankrupt and the courts have upheld a suit against the company that supplied the raw materials, even though those were entirely to specification. One result is that you now can't use any DuPont material in a medical device.

[hamdani yusuf]

Google Bard is not ChatGPT, but they use similar technology.
I asked it, what's the difference between design pressure and MAWP?
Here's the answer.

Design pressure is the highest pressure that a pressure vessel or other equipment is designed to withstand under normal operating conditions. It is calculated using a variety of factors, including the material of construction, the thickness of the vessel walls, and the safety factor.

MAWP, or Maximum Allowable Working Pressure, is the highest pressure that the equipment is allowed to operate at under any circumstances. It is typically set by the manufacturer or by a regulatory body, and it is always equal to or less than the design pressure.

The difference between design pressure and MAWP is a safety margin. This margin of safety is necessary to account for unexpected pressure surges, corrosion, and other factors that could weaken the equipment over time.

Here is a table summarizing the key differences between design pressure and MAWP:

Definition   
D: The highest pressure that the equipment is designed to withstand under normal operating conditions.   
M: The highest pressure that the equipment is allowed to operate at under any circumstances.

Calculation   
D: Calculated using a variety of factors, including the material of construction, the thickness of the vessel walls, and the safety factor.   
M: Typically set by the manufacturer or by a regulatory body.

Relationship   
D: Always equal to or less than MAWP.   
M: Always equal to or less than design pressure.

Purpose   
D: To ensure the safety of the equipment and its operators.   
M: To provide a safety margin in case of unexpected pressure surges, corrosion, and other factors.

Examples:

A pressure vessel is designed to operate at a pressure of 100 psi. The manufacturer sets the MAWP at 90 psi to provide a safety margin.
A pipeline is designed to withstand a pressure of 200 psi. The regulatory body sets the MAWP at 180 psi to provide a safety margin.
It is important to note that the MAWP should never be exceeded. Operating the equipment at a pressure above the MAWP could result in failure of the equipment and serious injury or death.

It doesn't seem to be aware of contradiction, as boldened statements above. This weakness needs to be fixed.

[alancalverd]

Which is why every decision must be traceable to a human or corporation that can be held liable for the consequences. Except, of course, politicians, priests, and economists.

[hamdani yusuf]

Which is why every decision must be traceable to a human or corporation that can be held liable for the consequences. Except, of course, politicians, priests, and economists.

Or when they are already dead.

[alancalverd]

A dead person or corporation can still be held liable for what they did when alive, and compensation can be sought from their estate or corporate successors When you buy a company, you buy its assets and liabilities.

I don't understand why Matt Hancock hasn't been sued over the deaths of those he mandated to be infected with COVID.

[hamdani yusuf]

A dead person or corporation can still be held liable for what they did when alive, and compensation can be sought from their estate or corporate successors When you buy a company, you buy its assets and liabilities.

I don't understand why Matt Hancock hasn't been sued over the deaths of those he mandated to be infected with COVID.

It's their heirs you're talking about.

[alancalverd]

No, their estate.The estate is the sum of monetisable assets that belonged to the deceased person. The heirs are the people to whom, after all prior claims have been settled, some or all of the assets have been bequeathed. Whether a claim can be retrospectively prioritised is, I think, dubious, but in any case the essence of all legal proceedings is to file your claim ASAP! In the case of a corporation purchased by another, the liabilities are transferred in the same way as the claims on a personal estate.