Nov

30

When do we start seeing the effects of AI show up in national economic data? If you had invested $5K in a laptop and a word processing program, you could replace a secretary at multiples of the cost. When the web came in, there was Amazon squeezing out the costs of the middlemen.

But I don't see the savings for AI. I see lots of talk, some free programs, but in terms of real productivity, not so much. I'm also told that it's early days and I'm asking for too much in posing such a question, but I think we're now getting far enough into AI that it's not an unreasonable matter to bring up.

One thing that's clear is that AI isn't going to generate employment the way the last tech push did. But if it's going to really change the world as its advocates suggest that it will, those productivity gains should be apparent by now.

M. Humbert writes:

However AI productivity gains are measured, it’ll have to account for the productivity loss due to its high energy consumption. For the Austrian economics fans here. I’ve found Copilot to be a helpful time saving tool, so others probably do as well, so time savings definitely are occurring from AI use today.

Laurence Glazier responds:

Using it all the time, huge experiential benefit. Chatting to GPT every morning while reading Thoreau. Instant context. The other big breakthrough is spatial computing. All in the service of art.

Asindu Drileba comments:

From my experience, co-pilot and other LLMs, have not solved anything that could not already be done via ordinary Googling. Looking up solutions to code issues on stack overflow is no different from LLMs. And stack overflow is still better for some tasks (fringe computer languages like APL for example). LLMs are impressive, but are mostly just gimmicks. The only thing it has actually saved me time on is generating copyrighter material and filler text.

Jeffrey Hirsch adds:

Just had that discussion today about ordinary google still being even better than LLM Ais in finding info. Had some fun with AI editing and embellishing copy.

Asindu Drileba adds:

I suspect that the bad SWE job market is due to high interest rates, no AI. The SWE job market is enriched mostly by VC money. And VC money dried up when LPs withdraw to earn risk free money in treasuries instead of betting on start-ups whose success is on probability. I expect it to recover if interest rates come down to previous levels.

I think the LLM narrative was just something that tech executives parroted to show they had an LLM strategy. It's, Like how in 2018/2017 every executive had a "Blockchain" strategy. A lot of businesses assumed that LLMs would replace simple customer support jobs but they just saw their tickets pile up. Even the $2B valued, Peter Thiel financed, code assistant that would make you money on Up work as you sleep turned out to be a blatant scam.

Steve Ellison writes:

I don't have an answer for Dr. Lilienfeld's question about when AI effects will show up in productivity statistics. But I do hear anecdotally through my professional networks that AI projects are adding real value.

At the same time, Asindu is correct that the bad job market for techies, myself included, is more a consequence of rising interest rates–and I would add overhiring during the pandemic–than positions being replaced by AI. As Phyl Terry put it, "But this company [that announced layoffs] wants to go public so the better story is 'we are smart leaders using AI to become more efficient and profitable' vs 'we were idiots during the pandemic and have to lay off some people because we messed up.'"

Gyve Bones writes:

I find that the AI's ability to interpret my request and put together a coherent synthesis of several sources to be very helpful. Grok is nice because it provides a set of links to sources relevant to the prompt, and to related ??-posts and threads.

Laurence Glazier asks:

I usually have audio conversations with GPT rather than the older typed-in input/output. I just subscribed to X Premium to get access to Grok. Any good links for learning good usage? How nice Musk names it from the Heinlein novel.

Gyve Bones responds:

Check out the sample prompts Grok supplies on the [ / ] section in ??. The news analysis prompts for trending items is pretty cool.

Bill Rafter writes:

My business partner and I are in the process of marketing a new software application. Although we are rather literate, we have been running all of our marketing materials through Copilot, and we are amazed at the improvements Copilot makes to our text. It results not only in improved communication, but is a real time-saver. We even asked it to write a business plan, and it came back with a better one than our original.

Peter Penha offers:

I have not (yet) been on Grok but have found that the prompts do not differ very much across LLMs:

A Primer on Prompting Techniques, June 2024.

Prompt engineering is an increasingly important skill set needed to converse effectively with large language models (LLMs), such as ChatGPT. Prompts are instructions given to an LLM to enforce rules, automate processes, and ensure specific qualities (and quantities) of generated output. Prompts are also a form of programming that can customize the outputs and interactions with an LLM. This paper describes a catalog of prompt engineering techniques presented in pattern form that have been applied to solve common problems when conversing with LLMs. Prompt patterns are a knowledge transfer method analogous to software patterns since they provide reusable solutions to common problems faced in a particular context, i.e., output generation and interaction when working with LLMs. This paper provides the following contributions to research on prompt engineering that apply LLMs to automate software development tasks. First, it provides a framework for documenting patterns for structuring prompts to solve a range of problems so that they can be adapted to different domains. Second, it presents a catalog of patterns that have been applied successfully to improve the outputs of LLM conversations. Third, it explains how prompts can be built from multiple patterns and illustrates prompt patterns that benefit from combination with other prompt patterns.

This is earlier/shorter February 2023 paper - I am also a fan/follower of Prof. Jules White’s classes on Coursera why I flag the shorter/earlier paper as well.

Separate on the subject of AI - Eric Schmidt has a new book Genesis with Dr. Kissinger as a co-author (his last work before his passing) but Schmidt did a Prof G Pod Conversation released Nov 21st - in the podcast Schmidt goes over the threat from LLMs that are unleashed and noted that China in his view has open sourced an LLM equal to Llama 3 and that China instead of a being three years behind the USA on LLMs is a year behind. That China comment can be found here at 26:30.

Finally if anyone wants a great book I have read, on the history of the race to AGI going back to 2009: the Parmy Olsen book Supremacy on the histories of Sam Altman and Demis Hassabis is a wonderful read. Also breaks the world down between the AI accelerationists and the AI armaggedonists.

Big Al adds:

I do use Bard to learn or refresh my memory with R. For example, I am trying to use the "tidyverse" set of packages, and Bard is very useful when asked to write code for some task specifically using, say, tidyquant. The code almost never works first time cut & paste, but I can see how things are done differently and figure out what needs fixing. And I get answers to simpler problems faster than on Stack Exchange which is better for more complicated issues.

Laurence Glazier comments:

It's an inverted Turing test situation. The things that AI can't do help identify our humanity, our birthright.


Comments

Name

Email

Website

Speak your mind

1 Comment so far

  1. Christopher J. Severs on December 1, 2024 7:47 am

    Unlocking creativity, allowing the expansion of ideas at an exponential pace. At this point, I am using it at work to perform tasks that I previously dreaded due to monotony or tediousness. Tasks that honestly were not worth, in my opinion, the rate which I charge. Personally, it allows me to explore topics and ideas that seemed previously insurmountable, simply due to the time required to put all my thoughts together. Sure, if you have the capital, you can hire others to do the research, write the papers, and turn your thoughts and ideas into reality. However, with AI, I can do it all myself on a Sunday. To me it feels most similar to the creation of the internet. I remember when I was in college, trudging down to the library, book store, or newspaper stand to find material to educate myself on issues I wanted to explore. Then voila, the internet brought all those together and I could research and learn much more efficiently. AI now saves me the time of typing out my thoughts and putting together my ideas into a digestible format to share with others. Seems to me to be the logical next step in the expeditation of the dissemination of information.

Archives

Resources & Links

Search