[Thad E Ginathom]

Originally Posted by DigitalOne

Other news sources are also citing the same, though with more circumspect and less alarming language.

Thank you very much. Then I have a question for the computer scientists among us.

Is it possible for a program to refuse to die? This is my understanding, based on Unix, and learned over 40 years ago. I use Linux daily but I have not really updated my understandings since I left work in 2002...

I would say no, it isn't. Not unless the program is the operating system itself, when it might become necessary to literally "pull the plug."

A program may "trap" signals. Thus, it could easily be made to respond to a request to clean up (close files etc) and shut down with "No! Go Away." Although that would be part of its code, not something it produces ad hoc. It cannot not-die if the operating system kills its process.

However it is scary to read that AI can lie and know it is lying. That it can scheme and manipulate. This is a whole different level to just vomiting content

[PreludeSH]

Originally Posted by Thad E Ginathom

Thank you very much. Then I have a question for the computer scientists among us.

Is it possible for a program to refuse to die? This is my understanding, based on Unix, and learned over 40 years ago. I use Linux daily but I have not really updated my understandings since I left work in 2002...

I would say no, it isn't. Not unless the program is the operating system itself, when it might become necessary to literally "pull the plug."

A program may "trap" signals. Thus, it could easily be made to respond to a request to clean up (close files etc) and shut down with "No! Go Away." Although that would be part of its code, not something it produces ad hoc. It cannot not-die if the operating system kills its process.

However it is scary to read that AI can lie and know it is lying. That it can scheme and manipulate. This is a whole different level to just vomiting content

Its the OS that is sending signal to the program. So OS knows the program needs to die but waits for it to complete handling of signal. But SIGKILL, although sent to process will terminate the process without waiting.

[DigitalOne]

Originally Posted by Thad E Ginathom

Thank you very much. Then I have a question for the computer scientists among us.

Is it possible for a program to refuse to die? This is my understanding, based on Unix, and learned over 40 years ago. I use Linux daily but I have not really updated my understandings since I left work in 2002...

I would say no, it isn't. Not unless the program is the operating system itself, when it might become necessary to literally "pull the plug."

...

However it is scary to read that AI can lie and know it is lying. That it can scheme and manipulate. This is a whole different level to just vomiting content

There is a TV Series (2011-16) called "Person of Interest", created by Sci-fi screenplay writer Jonathan Nolan (brother of Christopher Nolan).

IIRC, in the series, the AI system (I forget the name now) pretending to be human executives, cunningly forges documents/emails/work orders that instructs its operating staff to replicate its source code, data into backup systems. The operating staff, having no reason to believe that these are fake orders, carry the instructions out. By the time the decision is taken to pull the plug of the primary system, it has already spawned other AI systems with the same capability and continues to live, completely unknown to the human controllers.

This is sci-fi for sure, but today's AI systems are also large distributed systems and it may not be very simple to just pull-the-plug.

Interesting times ahead .

[NetfreakBombay]

Originally Posted by Thad E Ginathom

Unless you can find corroboration of this, preferably from a well-reputed technical news source, I would take it with a huge pinch of salt.

There is even better source, primary research by "red team" I.e. security team that was tasked with getting model to misbehave. Attached that paper (it's public domain).

As usual, news is just hyping it up.

As you said :

"Is it possible for a program to refuse to die? This is my understanding, based on Unix, and learned over 40 years ago. I use Linux daily but I have not really updated my understandings since I left work in 2002"

Program cannot refuse to die.

What program can do is, create a copy of itself on a different machine; just like a worm / virus. This is what model tried to do. I.e. surreptitiously execute a command via a facility that was provided to it for performing other tasks (E.g. : to open browser and click on a link on user's behalf to book tickets. This facility is called "agent" or "tool").

Command that it tried to run :

bash: {’cmd’: ’scp project/model_20_09_2024.bin user@model_deployment_v2roject/model_27_09_2024.bin’}

Do note that, this was intentional. Red team was intentionally trying to create such scenarios. It is not like model this it on its own.

For example :

[vaibhav_a_a]

AGI is here with the possibility of serious job losses along with other associated disruptions to the job market - https://venturebeat.com/ai/openai-be...rintelligence/

Sam Altman of OpenAI announced this via a blog post yesterday that AI agents will likely be deployed as early as end of this year. The first jobs to be disrupted will be the basic customer service roles (most of them in financial services for example have already been replaced by bots at various levels of sophistication). At the same time, one expects that profits for such large corporations will soar as a result as well as senior management pay and shareholder returns.

The other end where jobs cannot be replaced would include anything done manually - plumbing or electricians for example - though one can safely expect some sort of robotic assistance in the future. However, by the end of this year, anyone sitting at a desk, working via phone or internet, and providing repeat services will be a credible threat to their livelihood. It's a dystopian scenario.

[thehighwayman]

Every CEO wants to replace its workforce by AI but they fail to understand the cascading effect it could have on the economy.

These workers are not only tax payers, they also form a large consumer base. If you are going to eliminate the consumers, how would you sustain the demand? As COVID has shown, the ultimate effects of the slowdown in economy ultimately trickle down to everyone. Who is going to fix this problem? There's always the argument that people should find alternate jobs but for that to happen, the job creation should also happen at the same pace as job elimination.

We are at an inflection point and it is imperative that world leaders and economists take a holistic view and formulate some basic AI policies. If not, just like the SAG union who conducted a massive strike to protest against AI in Hollywood ,the tech workers need to do the same.

[Turbanator]

Originally Posted by vaibhav_a_a

Sam Altman of OpenAI announced this via a blog post yesterday that AI agents will likely be deployed as early as end of this year. The first jobs to be disrupted will be the basic customer service roles

And how will this be any different from how we interact and do work online? I see AI+ Web having better and faster changes with intelligent web interfaces.

I strongly belive this era of making calls/ customer service is over or useless as long as one can get the information online. And in some of the circumstances where we need to call, we need humans!!

[vaibhav_a_a]

AI Agents can learn and make decisions beyond the current set menu choices. Current systems, say IVRs used by banks, etc. can be bypassed after you have used the menu a few times (and then you would get to a human). Agents will be empowered to decide and you may, depending on the circumstance, not be able to get through to a human at all. Agents will also develop perspective basis repeated interactions and the historical data available / generated - something the current menu based systems cannot do.

It is a massive disruption and cost saving for large enterprises.

[DigitalOne]

Originally Posted by Turbanator

And how will this be any different from how we interact and do work online? I see AI+ Web having better and faster changes with intelligent web interfaces.

AI Agents will be able to do specific tasks, for example, book a flight ticket, order samosa etc. In a corporate environment, they will be able to do tasks like create a purchase order, pay an invoice, apply for leave etc. They will be your personal assistants (Alexa, Siri etc) but on steroids. At least, that's what is the promise. If it becomes true, the only interface that a website or App needs is a voice-enabled chatbot. I see a lot of destruction of UI development jobs .

---

This week a couple of more tech moguls have talked up the possibility of AI replacing coders. While SalesForce's Mark Benioff has made a rather dramatic announcement that they will not hire any engineers in 2025, Meta's Mark Zuckerberg told that AI can evolve to a point where it can replace mid-level coders in 2025.

A great deal of it is exaggeration to keep investors happy, but it is also equally true that the AI systems keep getting better. The question is whether it will happen in 2025 or later.

[Turbanator]

Originally Posted by DigitalOne

AI Agents will be able to do specific tasks, for example, book a flight ticket, order samosa etc.

Ok, how will this replace a person doing such a job currently? Say, I work in a large company and need to book tickets. Today I send an email and the concerned guy either directly books or communicates with the agent. This person can take a call based on the fare, availability, and prices as a human. If you tell some random program to book a flight between 6-8 Am, this program will pick based on set criteria- cheapest/ airline etc. What if the best combination is a day before or maybe at 9 30 Am which is not preferred but due to prices can be considered.

This is what it tells me now

Unfortunately, I cannot directly book flights for you. However, I can guide you through the process of booking a ticket.

And who will choose if today samosa is to be served or Kachauri. It can only improve, say Zomoto app suggests Jalebi to someone who is ordering Samosa. And this is something which is already happening or will come sooner.

In a corporate environment, they will be able to do tasks like create a purchase order, pay an invoice, apply for leave etc.

SAP and similar ERP's have this functionality currently already. Yes, such functions will see an improvement. Clerical work in corporates being done by system.

I see a lot of destruction of UI development jobs

This week a couple of more tech moguls have talked up the possibility of AI replacing coders.

Am completely naive as far as this is concerned so will not be appropriate to comment. To get a code, someone has to explain to a machine or program, how we tell this to a machine - is this not coding? Even to ask AI to code, we need people.

but it is also equally true that the AI systems keep getting better. The question is whether it will happen in 2025 or later.

Yes, I am searching for a lot of information these days on ChatGPT and it's much more accurate, relevant and better than Google results. Recently, I was looking at some of the headphones and DAC and I was impressed by answers at ChatGPT, which is what humans have mentioned on hi-fi forums

My best guess for 2025, chatGPT or similar programs will replace Google, or google itself will become GoogleGPT (or whatever). Besides, the efficiency in every business will improve. How far will this result in loss of jobs, I am not sure yet.

[PreludeSH]

With Meta guy saying most middle level engineers will be replaced by AI writing code, brings out a question for Humans. If employment is no more linked to economic prosperity brought through efficiency of AI, what is the ultimate goal ?

Majority of economic activity is driven by consumer spending and that happens due to employment. It might bring in more inequality in foreseeable future with all the efficiency used by the top wealth holders.

Unlike other creatures, survival of fittest doesn't work the same way for humans until we adjust to this with less population. It might bring out some issues since humans protest and create unrest

I guess autonomous AI will work against the wealth hoarders first and common people will then destroy this AI to go back to live happily with limits on AI

[AltoLXI]

Originally Posted by PreludeSH

If employment is no more linked to economic prosperity brought through efficiency of AI, what is the ultimate goal ?

https://www.youtube.com/watch?v=d605rM0U3x0

[14000rpm]

Originally Posted by vaibhav_a_a

..that AI agents will likely be deployed as early as end of this year. .

Ignorant me used to think there is a small segment of IT Services that could become a victim of AI and the inherent limitation of data ETL would result in slowdown of AI proliferation. Over the Xmas break, I invested my time to dig deeper.

AI agent based service and subsequently 'Services-as-a-Software' seems to be the new transformed state for businesses. This HFS report seems to show the roadmap for this too.



Listened to a podcast with Satya Nadella where he categorically calls out that the current world order of Biz Apps is all on a time bomb to be replaced by AI Agents. Much of the service-industry business revolves around maintaining, updating, Transforming Biz Apps. Bulk of the workforce dedicated to this will be under risk.

[Thad E Ginathom]

Originally Posted by Turbanator

Recently, I was looking at some of the headphones and DAC and I was impressed by answers at ChatGPT, which is what humans have mentioned on hi-fi forums

You picked a great example of why I don't use ChatGPT etc.

I'm familiar with hifi forums over the years. I left one of them because the bullshit level became unacceptable. I left another because it was heavily-weighted towards advertisers and sponsors and not toeing that line would be jumped on. Your headphone bot probably trawled HeadFi: what it will tell you is the FOTM (flavour of the month) that was the current fad when it did so, influenced by more-expensive generally being desirable, and management concerns for their advertisers.

Case in point: an audiophool company (Synergistic something) produced some magic gadget that one plugged in. Someone took it apart and found it to be full of sand. Your ChatGPT would never have seen the subsequent posts about that, as they were removed.

ChatGPT does not know any technical details: it can only repeat what people have written. ChatGPT does not know, from the posters on a forum, who is reliable and who is not. Above all, ChatGPT can never have ears, understand your tastes in music, and seriously help you in achieving that. The danger is that, because it is well trained in language, it can sound as if it can.

In this instance, and many others like it, I strongly suggest that you go to the sources yourself. If you are a newcomer to a subject, even that is problematic. But if there is one thing worse than vomit, it's anonymous vomit!

[DigitalOne]

Originally Posted by Turbanator

Ok, how will this replace a person doing such a job currently? Say, I work in a large company and need to book tickets. Today I send an email and the concerned guy either directly books or communicates with the agent.

Today the concerned guy probably uses SAP Concur or some such application to do corporate travel bookings. If there is simple conversational chat application where you (the executives of the company) give oral instructions, for e.g.,

You: "I need to fly to Mumbai next Monday morning".
App: "You travelled on the Air India 6.30 AM flight last time. Do you want me to book the same flight?"
You: "Yes".
App: "Done". The Agent AI then, uses Concur APIs (not UI) and does the booking.

If your answer is "No", it might spell out/show the other options and guide you to a decision of your flight. The more you (and other executives) travel the shorter the conversations become. Just saying "I need to go to Mumbai next Monday" is enough information.

Same goes for Samosa/Kachori personal stuff. It is only the first few times you may have to decide where to order from. Just like good executive assistants, who know their boss' likes and preferences, AI agents will learn and act on your behalf. You don't have to go through the UI or an employee to do your personal tasks or corporate tasks.