noelward
Well-known member
Is AI Real?
By Noel Ward, Editor@Large
It’s a hot topic: Seats are filled at any conference where AI is on the agenda. Everyone is seeking an edge, an advantage, a way of leveraging AI. At one recent conference most AI presentations were underwhelming, yet all showed potential, even though it’s hard to tell how much is real. The unevenness of many AI incarnations is okay for now because it is an emerging technology that’s not ready for Prime Time and provokes new ways of thinking and the approach to many tasks.
To find out more I went to some conferences.
An amorphous AI-generated voice sprang out of a laptop at center stage. The voice was accompanied by a series of slides highlighting what we were about to see and hear. AI replaced a speaker who later told me AI had created the “speech” and monochromic slides based on talking points that had been fed to the AI. Creating the short talk track, including the slides, took about seven minutes using the website Notebooklm.google.com. It looked to be worth checking out.
Such a capability could be a boon to people who know a topic well but less than comfortable speaking before a group. That probably covers a lot of people! Or maybe like this: AI can cover for having inept staff, lack of a speechwriter, or just be a massive time saver. If you have a presentation coming up it may be worth seeing how AI might work for you. Then there’s the show-off factor of possibly impressing clients.
The drawback, in my opinion, is that AI lacks the personalization of what I think of as a real presentation. I’m possibly biased, having given presentations in several countries, all entirely unscripted and reliant on photos, video and three bullet points per “slide.” Still, an automated AI version has its place in some kinds of presentations or as the voiceover for a video or a series of static images. It would have been useful back when I produced videos. Plus, it’s fast and easy, two compelling advantages. Definitely worth considering for such needs.
The accuracy issue
Still being developed, AI it is not necessarily a smooth ride: One may find it can work well in addressing some needs while slow or misleading for others. It is notable that any AI output arrives with claims stating it may be inaccurate. Still, accuracy will improve and once it is trusted AI will be a “go-to” for everyday use, along with a host of specialized tools (mostly topic-specific training and algorithms) for some needs. When appropriately trained (a complex process), AI is likely to be effective for a host of specialized applications including some aspects of printing.
Skeptics like me are still quick to claim that AI is currently a mix of hype, smoke, and mirrors. Some of it certainly is, and major technological shifts are rarely ready for Prime Time when first rolled out. Mainstays like Excel or your browser originally came up short but now are commonplace and may offer multiple versions. Today’s presentations about AI are best seen as a preview of things to come, some of which may be differences you can take to the bank. A year from now AI is likely to be playing a role in running parts of your business. My skepticism notwithstanding, the smart money is on starting your AI adventure sooner rather than later. Try AI out internally and share your experiences. Cooperation matters here. It’s how we all get smarter.
Still…
AI can provide great insight and perspectives because it has fast access to stuff we have forgotten, don't know or are not aware of. This is way cool. But you still have to fact check it because AI has been trained on stuff that isn't necessarily factual.
How does this happen? One way can be like this: I write a lot of non-fiction articles and papers. While I make a real effort to be accurate not all writers do. Some of this work and that of others is used to train AI so errors can pop up in places you don't expect. As a result, AI herds information on a topic into an answer that may not be fully correct, lacks context or is difficult verify. This is one reason AI responses invariably claim it makes mistakes. This is lawyer-driven line line is automatically included in the response algorithm to dodge responsibility. Check your sources.
You were going where?
One of the most interesting things about AI is its accessibility to many users, especially those with the unique and complex needs of science and medicine. Progress is being made but these realms are some of the most difficult aspects of AI’s training which is already costly and time-consuming.
An important element is the collective knowledge of AI presenters and users in revealing some shared experiences. An leading industry analyst noted that AI iterations work differently because of differences in how any given AI engine has been trained to use information and algorithms. In practice, the AI’s training may limit how the LLMs (large language models) can be used or by user requirements. In some cases though, assertations from one AI tool can be “fed” onto another AI system as a way verifying, refining or modifying the results. As the saying goes, your mileage may vary.
An important caution is that AI doesn’t aways get you where you want to go and may require engaging someone who knows things you don’t. If you do this, be sure the third party has the skills required and is familiar with the version of AI you are using. This can be critical because skill with a specific AI tool can be important. As the analyst noted above noted, not all AI works the same way. This is why someone with expertise on OpenAI’s ChatGPT may be less than helpful if you are using Anthropic’s Claude. Then there’s speed, often touted as a key advantage. But don’t decide solely on speed because it may depend on a variety of factors, including your computer, network or internet connection. Still, the potential is real and arriving fast.
Not to exceed
Another point is that the cost of using AI can be amorphous. For instance, is possible to select an AI service you think will cost $300 then find yourself billed for $2,500 because your need was considered more complex or required more time than expected. Moreover, if you are using an outside provider (not uncommon!) be sure the fee is agreed on up front. This can be basic language, such as a “not to exceed” line in an agreement. Pricing variances may not settle down for some time so if you are hiring expertise try to establish a final charge before agreeing to the work at hand.
Finally, make sure your findings and assumptions make sense. First, as noted above, just because AI “said it” doesn’t make it accurate. Next, it can be slow. In one demonstration a tech-savvy printer had partnered with a software firm. They accessed and used seven LLMs when providing results which amounted to a couple screens and a few lines of text. All done live and slow to arrive. OK, it was maybe faster than perusing a spreadsheet in a marfinally-organized file but not money-saving fast.
Also, find out what works for you and your team. My team and I find Claude works well for us, while others are happy with ChatGPT. Pick what works for you or use different ones for different tasks. Be sure to compare them along the way because they are competing and will change. This is one part sanity check and one part error control. Right now, mixing them up may make sense. No matter how you apply it, look for ways of using AI now or be left behind.
By Noel Ward, Editor@Large
It’s a hot topic: Seats are filled at any conference where AI is on the agenda. Everyone is seeking an edge, an advantage, a way of leveraging AI. At one recent conference most AI presentations were underwhelming, yet all showed potential, even though it’s hard to tell how much is real. The unevenness of many AI incarnations is okay for now because it is an emerging technology that’s not ready for Prime Time and provokes new ways of thinking and the approach to many tasks.
To find out more I went to some conferences.
An amorphous AI-generated voice sprang out of a laptop at center stage. The voice was accompanied by a series of slides highlighting what we were about to see and hear. AI replaced a speaker who later told me AI had created the “speech” and monochromic slides based on talking points that had been fed to the AI. Creating the short talk track, including the slides, took about seven minutes using the website Notebooklm.google.com. It looked to be worth checking out.
Such a capability could be a boon to people who know a topic well but less than comfortable speaking before a group. That probably covers a lot of people! Or maybe like this: AI can cover for having inept staff, lack of a speechwriter, or just be a massive time saver. If you have a presentation coming up it may be worth seeing how AI might work for you. Then there’s the show-off factor of possibly impressing clients.
The drawback, in my opinion, is that AI lacks the personalization of what I think of as a real presentation. I’m possibly biased, having given presentations in several countries, all entirely unscripted and reliant on photos, video and three bullet points per “slide.” Still, an automated AI version has its place in some kinds of presentations or as the voiceover for a video or a series of static images. It would have been useful back when I produced videos. Plus, it’s fast and easy, two compelling advantages. Definitely worth considering for such needs.
The accuracy issue
Still being developed, AI it is not necessarily a smooth ride: One may find it can work well in addressing some needs while slow or misleading for others. It is notable that any AI output arrives with claims stating it may be inaccurate. Still, accuracy will improve and once it is trusted AI will be a “go-to” for everyday use, along with a host of specialized tools (mostly topic-specific training and algorithms) for some needs. When appropriately trained (a complex process), AI is likely to be effective for a host of specialized applications including some aspects of printing.
Skeptics like me are still quick to claim that AI is currently a mix of hype, smoke, and mirrors. Some of it certainly is, and major technological shifts are rarely ready for Prime Time when first rolled out. Mainstays like Excel or your browser originally came up short but now are commonplace and may offer multiple versions. Today’s presentations about AI are best seen as a preview of things to come, some of which may be differences you can take to the bank. A year from now AI is likely to be playing a role in running parts of your business. My skepticism notwithstanding, the smart money is on starting your AI adventure sooner rather than later. Try AI out internally and share your experiences. Cooperation matters here. It’s how we all get smarter.
Still…
AI can provide great insight and perspectives because it has fast access to stuff we have forgotten, don't know or are not aware of. This is way cool. But you still have to fact check it because AI has been trained on stuff that isn't necessarily factual.
How does this happen? One way can be like this: I write a lot of non-fiction articles and papers. While I make a real effort to be accurate not all writers do. Some of this work and that of others is used to train AI so errors can pop up in places you don't expect. As a result, AI herds information on a topic into an answer that may not be fully correct, lacks context or is difficult verify. This is one reason AI responses invariably claim it makes mistakes. This is lawyer-driven line line is automatically included in the response algorithm to dodge responsibility. Check your sources.
You were going where?
One of the most interesting things about AI is its accessibility to many users, especially those with the unique and complex needs of science and medicine. Progress is being made but these realms are some of the most difficult aspects of AI’s training which is already costly and time-consuming.
An important element is the collective knowledge of AI presenters and users in revealing some shared experiences. An leading industry analyst noted that AI iterations work differently because of differences in how any given AI engine has been trained to use information and algorithms. In practice, the AI’s training may limit how the LLMs (large language models) can be used or by user requirements. In some cases though, assertations from one AI tool can be “fed” onto another AI system as a way verifying, refining or modifying the results. As the saying goes, your mileage may vary.
An important caution is that AI doesn’t aways get you where you want to go and may require engaging someone who knows things you don’t. If you do this, be sure the third party has the skills required and is familiar with the version of AI you are using. This can be critical because skill with a specific AI tool can be important. As the analyst noted above noted, not all AI works the same way. This is why someone with expertise on OpenAI’s ChatGPT may be less than helpful if you are using Anthropic’s Claude. Then there’s speed, often touted as a key advantage. But don’t decide solely on speed because it may depend on a variety of factors, including your computer, network or internet connection. Still, the potential is real and arriving fast.
Not to exceed
Another point is that the cost of using AI can be amorphous. For instance, is possible to select an AI service you think will cost $300 then find yourself billed for $2,500 because your need was considered more complex or required more time than expected. Moreover, if you are using an outside provider (not uncommon!) be sure the fee is agreed on up front. This can be basic language, such as a “not to exceed” line in an agreement. Pricing variances may not settle down for some time so if you are hiring expertise try to establish a final charge before agreeing to the work at hand.
Finally, make sure your findings and assumptions make sense. First, as noted above, just because AI “said it” doesn’t make it accurate. Next, it can be slow. In one demonstration a tech-savvy printer had partnered with a software firm. They accessed and used seven LLMs when providing results which amounted to a couple screens and a few lines of text. All done live and slow to arrive. OK, it was maybe faster than perusing a spreadsheet in a marfinally-organized file but not money-saving fast.
Also, find out what works for you and your team. My team and I find Claude works well for us, while others are happy with ChatGPT. Pick what works for you or use different ones for different tasks. Be sure to compare them along the way because they are competing and will change. This is one part sanity check and one part error control. Right now, mixing them up may make sense. No matter how you apply it, look for ways of using AI now or be left behind.
rint cartoons. Here's what it came up with (best out of 5 attempts):
