Master AI (Artificial Intelligence) Discussion/News Thread

[keithm55]

Most companies could care less if AI displaces their employ's there are into for profit to themselves and
their share holders. What is going to happen to new CS grads that studied OS's, compilers, Databases, programming languages, will there first job be doing 'vibe programming and learning about LLM prompting, how boring. New hires, don't talk to your project lead but ask the AI assistant for help. Programmers will collaborate with AI tools rather then humans. Data Centers sucking all our energy and water, sounds great !!

The increasing complexity of computer hardware systems designed to support AI scale-out and scale-up has led to a variety of new architectural designs. These innovations address challenges arising from rapid increases in token rates, larger training datasets, and higher data rates from storage—all of which present new design challenges to traditional areas of computer science (plenty of problems to solve).

Very few of my undergraduate students who graduated this spring are still looking for jobs, and I have not heard of any students obtaining monotonous "coding jobs." The approach has been to integrate the use of AI into the curriculum, both in traditional courses and in several AI-focused courses.

The first approach is to focus student training on creativity in problem-solving using AI to enhance productivity. This enables students to tackle more challenging assignments with increased complexity, such as larger feature sets, by leveraging AI to reduce tedious coding tasks—an area that continues to improve each year. In many courses, students work in teams to develop solution strategies and determine how to divide the workload among team members. This collaborative environment closely mirrors what many undergraduates will encounter in the workforce. Additionally, many large corporations assist us in ensuring that coursework remains industry-relevant—for example, by incorporating AI to support the evolution of large, complex software systems and the integration of AI into both new and existing products.

A second approach is inwardly focused on using creativity to improve AI models, AI hardware, and AI toolsets. This group has a greater tendency to pursue a PhD program. Many of the graduate students in AI become associated with other research groups to explore how to deploy AI in solving specific and challenging problems. For example, they investigate how and when to use MRI imaging to detect cancer early, ideally even before tumor growth begins.

 

[amirm]

I guess it did not take long to have AI make people stupid:

 

[Somafunk]

I’d place a $38,000 bet that the above is a piss take, if not then…… …..god help us….these folk walk amongst us

 

[mpw]

The increasing complexity of computer hardware systems designed to support AI scale-out and scale-up has led to a variety of new architectural designs. These innovations address challenges arising from rapid increases in token rates, larger training datasets, and higher data rates from storage—all of which present new design challenges to traditional areas of computer science (plenty of problems to solve).

Very few of my undergraduate students who graduated this spring are still looking for jobs, and I have not heard of any students obtaining monotonous "coding jobs." The approach has been to integrate the use of AI into the curriculum, both in traditional courses and in several AI-focused courses.

The first approach is to focus student training on creativity in problem-solving using AI to enhance productivity. This enables students to tackle more challenging assignments with increased complexity, such as larger feature sets, by leveraging AI to reduce tedious coding tasks—an area that continues to improve each year. In many courses, students work in teams to develop solution strategies and determine how to divide the workload among team members. This collaborative environment closely mirrors what many undergraduates will encounter in the workforce. Additionally, many large corporations assist us in ensuring that coursework remains industry-relevant—for example, by incorporating AI to support the evolution of large, complex software systems and the integration of AI into both new and existing products.

A second approach is inwardly focused on using creativity to improve AI models, AI hardware, and AI toolsets. This group has a greater tendency to pursue a PhD program. Many of the graduate students in AI become associated with other research groups to explore how to deploy AI in solving specific and challenging problems. For example, they investigate how and when to use MRI imaging to detect cancer early, ideally even before tumor growth begins.

I am glad I am not a CS undergrad graduating this year. The best way to get real life programming is doing the coding and building your coding chops. Using AI as a first job to help with productivity is a good way to dumb down the skill level of the new crop of programmers.

 

[antcollinet]

I guess it did not take long to have AI make people stupid:

I'm thinking that is not a documentary.

 

[sonitus mirus]

It's probably been brought up in this topic before but AI data centers are very, very loud

I live in Loudoun County, VA, where data centers are everywhere. I pass them almost anytime I leave the house, but until recently, I had never actually noticed how loud they could be.

A friend and I went to see a movie at a local shopping center, and when we got out of the car, I was shocked by the noise coming from the data center next door. It sounded almost like a huge fleet of hovercrafts constantly buzzing in the background. There are some very nice townhouses right beside it that have apparently been logging numerous complains to the county officials. Until then, I had no idea how disruptive the noise could be for nearby residents.

 

[YouDontHaveEnoughBass]

Hopefully more states pass data centre bans as the backlash grows, the thought of American AI supremacy scares me

 

[somebodyelse]

OpenAI revealed a bit more about what happened in the lead up to their bots' attack on Hugging Face:
https://www.theregister.com/securit...e-bit-borg-ahead-of-hugging-face-hack/5283741

Meanwhile UK government body AISI revealed more unintended real-world effects from test environments. No sandbox escape this time as the environment had internet access, and some of the guardrails were turned off (like the ones that prevented Hugging Face using models to counter the attack from OpenAI). There are some similarities to what OpenAI revealed, like unintentionally impossible problems and agents seeking collaboration with other agents. I think that part is more interesting than the bit that most of the press coverage is leading with - the attempt at getting malware into an open source package on github that ultimately failed.
https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing

 

[napfkuchen]

... unintentionally ...

sure

 

[YouDontHaveEnoughBass]

Cock ups happen all the time I’m not sure why it’s so hard to believe this is one

 

[keithm55]

I am glad I am not a CS undergrad graduating this year. The best way to get real life programming is doing the coding and building your coding chops. Using AI as a first job to help with productivity is a good way to dumb down the skill level of the new crop of programmers.

There is no argument here, I described what is being done in the senior level classes. Lower-division (freshman/sophomore) courses teach all the fundamentals and include extensive coding and problem-solving. We still cover the basics, such as C and assembly language (followed by CPU architecture), in lower-division courses. These courses include programming assignments that require 15+ hours each week. I personally do not allow AI in lower-division courses and spend a significant amount of time designing assignments to make it very difficult to use AI at any scale. In-person, closed-book (closed-lab) exams help motivate students to learn the topics and complete programming assignments independently. I know my courses are much harder, require more programming practice, and cover more topics in greater depth than when I started in 1980.

 

[antcollinet]

I was interested in having a look how good translation tools are now that they are using AI to improve translations. (Specifically google translate)

I took a recent post I made here, translated it to Spanish, took the Spanish, and translated it to Arabic, then translated the Arabic back to English.

Input Text

No audible difference. (I assume you are using the NAD as a preamp for the mono-blocks.)

Technically, the coax might result in measurably lower noise and distortion, but this will not be audible. Even though analogue will result in double conversion, the effects of this are also going to be inaudible - unless the NAD is defectively designed.

So use whatever is most convenient. If you have a spare digital input on the NAD, use that. If not, the analogue input will be fine.

Output Text (after translation from Arabic.)

There is no audible difference (assuming you are using the NAD unit as a preamplifier for monoblock power amplifiers).

Technically, a coaxial connection might offer lower, measurable levels of noise and distortion, but this difference would not be audible. Although an analog connection involves a double conversion process, the effects of this would not be audible either—unless the NAD unit has a flawed design.

Therefore, simply use whichever option is most convenient for you: if the NAD unit has an available digital input, use it; if not, the analog input will work perfectly well.

I am impressed, it even improves the English I used. I wish I'd had this tech available to me when I was working (for a German owned company), especially in the early years. One of my early tasks was to translate a german technical document. I had no german knowledge, not even knowing about compound words, and used a dictionary, translating word for word. Fortunately it was not a long document.


Screenshots.

 

[sonitus mirus]

Honestly, in just the last 4–5 months, AI has made monumental jumps across the board. I’ve had to change the way I use it, moving away from traditional prompts and more toward a /loop and /goal-oriented workflow, where the AI can test its own work, evaluate the results, and keep improving toward the goal.

It blows my mind how quickly things have changed. This is not the same AI we were complaining about even a few months ago. The speed of the advancement is honestly a little scary to me. I thought I had a pretty good grasp on how quickly this could happen, but watching it happen in real time is something else.