LinkedIn is a verbotten tool in my office. There are entire segments of work areas that cannot be written about due to privacy or security concerns. (Translation? Do work on National Security stuff and they prefer you keep an extremely low profile. I don't have anything other than a placeholder so no one could stake jump my name.)
*Plus it's long since become a version of Facebook.
AI - Last week I dragged my post in from Word, 100% written by me but 100% thought to be AI content using the tool. A combination of personal anecdote and definitions using bulletpoints shows how the tool is useless. It might push me to screenshot the edit log just to show how poorly implemented these tools are.
Yes, I absolutely use the Claude to make graphics for posts. I feel no need to farm out to a graphic artist what takes 4 minutes of tweaks to create.
They haven't gotten the detections models to be good differentiators of "Common Knowledge" vs "Specific Knowledge". Ran both my father's and my Dissertations through a model, majority AI written? Same thing has happened to works by Shakespeare, Dickens, Steinbeck because the models were trained using the material without context as to who originated the text.
Humans would have Pattern Bias (Patternicity), thinking they see connections where now exist. Pretty common concern since we are trained from an early age to look for associations. It plays into Confirmation Biases...
Thank you, Lary. You said what I have also seen -- "AI detection tools" do not detect AI -- they merely follow an algorithm which sometimes detects AI. Full stop.
But I cannot let Substack's disclaimer about Nazi content rest. It is obvious to anyone who has paid attention over the last decade that Nazi content is currently promoted by the Government of the United States. More to the precise point: Facebook famously kicked Donald J. Trump off the platform for publishing lies, misleading information and Nazi/fascist apologetics. When he became President of the United States and began attacking every media company that had previously criticized him for these exact reasons, they relented, and subsequently they and other mainstream media became open to Nazi as well as other fascist propaganda under the spurious justification of "free speech" -- now they are among the main reasons why the American public has become more politically divided and more accepting of Nazi propaganda and fascist actions against the American people and immigrants. Substack is merely following this policy choice without stating it. This is not a political statement -- it is a clear description of what has happened and is happening.
And I appreciate your decision. I am in a somewhat different position, and so can more easily say these things openly -- so I feel a responsibility to do so.
Most AI models use an optimization function that accounts for patterns and makes a somewhat reliable choice. BUT even when people are recalling information they may organize it in a similar manner. Those voids in recall trigger AI to "think" it is an attempt to thwart detection. That loss is simultaniously updated and ignored via a learning function... An odd Stocastic, Deterministic, and Probabalistic Uncertainty.
Having been to Germany when there were still two distinct Countries, they are still reconcilling what to acknowledge and what they refuse to admit. We're seeing the same societal issues with people who were harmed and people who did the harming now 3-4 generations removed from the acts. Trump wants the Smithsonian marking exhibits he finds offensive because they show the ugliness at times. (Would love to see him tour the African American Museum and stand in the hall where the casket of Emmitt Till resides. I was sick to my stomach standing there knowing what occured. Yes it triggered guilt and being White that makes me question how far we've come. Others argue that someone shouldn't have any guilt since they are not directly responsible... the world doesn't work that way for polite society. You can't downplay or explain away that episode of our National history.)
I'm not for deplatforming people becaue it makes it so very easy to identify those I would want to ignore, avoid, or condemn with their own actions. In the current political environment in D.C. Substack might find themselves on the receiving end of a DOJ investigation simply for doing the bare minimum of labeling content (assuming we could agree to the framework).
Germany has stricter free speech laws than the US for obvious reasons. So far the US has continued to exist without the kind of collapse they saw in Germany in the 30s, partly because of stronger institutions, partly because of less devastating wars. But the third reason is that we have been willing to tolerate a great deal of oppression under the guise of "free speech" -- the long history of slavery and Jim Crow in this country is an obvious instance, but there are plenty of others -- treatment of Native Americans is a case in point, treatment of women, treatment of Chinese, and Irish, and Jewish, Muslim, Middle Eastern, and other immigrants. The list is very uncomfortably long, and includes my ancestors as well as our host. It has taken a very long time for this country to have arrived at even the shaky tolerance of those communities that we now have -- I see no benefit for having tolerated the levels of intolerance we have allowed through American history.
Very interesting topic which shocks me and makes me sad. Dr. A You did a great job in explaining and researching as well as sharing your thoughts. All the best in looking forward to a better platform.
Thanks, Lary. I appreciate you sharing how you use AI for graphics. Personally, this is an areas where I have struggled with. I can never get it to produce something that I feel good about, but I know many others who have done great work with PowerPoint slides and graphics using AI.
I have a family member that does graphic design and has created two silos of work, the traditional drawn by hand and that which AI assists to varying degrees. They know the work will become more difficult to keep at volume and have adjusted to learning about prompt design to hopefully start teaching entries into the job how to utilize the tool. Adapting is the only option.
*I draw caveman style hieroglyphics and no one wants that in a presentation.
You put out a good point that I think is underemphasized in your post. I don't believe the authors putting out AI slop care about this badge at all and will continue producing garbage. We can see the slop when we read it, we don't need several clicks to verify the slop. I imagine this will have close to zero impact and will actually frustrate authors like you and I who use AI to improve our thoughts but who are now being tagged as AI-assisted writers.
Ai purist are unfortunately some of the same people that will get left behind in other relams of life including work. AI is a tool just like hiring a editor is a tool. Personally i use both, but I hire editors when it is really important. I will always support hiring human talent, but do i really need a editor for a 200 word email?
LinkedIn is a verbotten tool in my office. There are entire segments of work areas that cannot be written about due to privacy or security concerns. (Translation? Do work on National Security stuff and they prefer you keep an extremely low profile. I don't have anything other than a placeholder so no one could stake jump my name.)
*Plus it's long since become a version of Facebook.
AI - Last week I dragged my post in from Word, 100% written by me but 100% thought to be AI content using the tool. A combination of personal anecdote and definitions using bulletpoints shows how the tool is useless. It might push me to screenshot the edit log just to show how poorly implemented these tools are.
Yes, I absolutely use the Claude to make graphics for posts. I feel no need to farm out to a graphic artist what takes 4 minutes of tweaks to create.
They haven't gotten the detections models to be good differentiators of "Common Knowledge" vs "Specific Knowledge". Ran both my father's and my Dissertations through a model, majority AI written? Same thing has happened to works by Shakespeare, Dickens, Steinbeck because the models were trained using the material without context as to who originated the text.
Humans would have Pattern Bias (Patternicity), thinking they see connections where now exist. Pretty common concern since we are trained from an early age to look for associations. It plays into Confirmation Biases...
Thank you, Lary. You said what I have also seen -- "AI detection tools" do not detect AI -- they merely follow an algorithm which sometimes detects AI. Full stop.
But I cannot let Substack's disclaimer about Nazi content rest. It is obvious to anyone who has paid attention over the last decade that Nazi content is currently promoted by the Government of the United States. More to the precise point: Facebook famously kicked Donald J. Trump off the platform for publishing lies, misleading information and Nazi/fascist apologetics. When he became President of the United States and began attacking every media company that had previously criticized him for these exact reasons, they relented, and subsequently they and other mainstream media became open to Nazi as well as other fascist propaganda under the spurious justification of "free speech" -- now they are among the main reasons why the American public has become more politically divided and more accepting of Nazi propaganda and fascist actions against the American people and immigrants. Substack is merely following this policy choice without stating it. This is not a political statement -- it is a clear description of what has happened and is happening.
You are correct, and thanks for chiming in. I debated for a long time whether to include that paragraph.
And I appreciate your decision. I am in a somewhat different position, and so can more easily say these things openly -- so I feel a responsibility to do so.
Most AI models use an optimization function that accounts for patterns and makes a somewhat reliable choice. BUT even when people are recalling information they may organize it in a similar manner. Those voids in recall trigger AI to "think" it is an attempt to thwart detection. That loss is simultaniously updated and ignored via a learning function... An odd Stocastic, Deterministic, and Probabalistic Uncertainty.
Having been to Germany when there were still two distinct Countries, they are still reconcilling what to acknowledge and what they refuse to admit. We're seeing the same societal issues with people who were harmed and people who did the harming now 3-4 generations removed from the acts. Trump wants the Smithsonian marking exhibits he finds offensive because they show the ugliness at times. (Would love to see him tour the African American Museum and stand in the hall where the casket of Emmitt Till resides. I was sick to my stomach standing there knowing what occured. Yes it triggered guilt and being White that makes me question how far we've come. Others argue that someone shouldn't have any guilt since they are not directly responsible... the world doesn't work that way for polite society. You can't downplay or explain away that episode of our National history.)
I'm not for deplatforming people becaue it makes it so very easy to identify those I would want to ignore, avoid, or condemn with their own actions. In the current political environment in D.C. Substack might find themselves on the receiving end of a DOJ investigation simply for doing the bare minimum of labeling content (assuming we could agree to the framework).
Germany has stricter free speech laws than the US for obvious reasons. So far the US has continued to exist without the kind of collapse they saw in Germany in the 30s, partly because of stronger institutions, partly because of less devastating wars. But the third reason is that we have been willing to tolerate a great deal of oppression under the guise of "free speech" -- the long history of slavery and Jim Crow in this country is an obvious instance, but there are plenty of others -- treatment of Native Americans is a case in point, treatment of women, treatment of Chinese, and Irish, and Jewish, Muslim, Middle Eastern, and other immigrants. The list is very uncomfortably long, and includes my ancestors as well as our host. It has taken a very long time for this country to have arrived at even the shaky tolerance of those communities that we now have -- I see no benefit for having tolerated the levels of intolerance we have allowed through American history.
Very interesting topic which shocks me and makes me sad. Dr. A You did a great job in explaining and researching as well as sharing your thoughts. All the best in looking forward to a better platform.
Thanks, Lary. I appreciate you sharing how you use AI for graphics. Personally, this is an areas where I have struggled with. I can never get it to produce something that I feel good about, but I know many others who have done great work with PowerPoint slides and graphics using AI.
I have a family member that does graphic design and has created two silos of work, the traditional drawn by hand and that which AI assists to varying degrees. They know the work will become more difficult to keep at volume and have adjusted to learning about prompt design to hopefully start teaching entries into the job how to utilize the tool. Adapting is the only option.
*I draw caveman style hieroglyphics and no one wants that in a presentation.
You put out a good point that I think is underemphasized in your post. I don't believe the authors putting out AI slop care about this badge at all and will continue producing garbage. We can see the slop when we read it, we don't need several clicks to verify the slop. I imagine this will have close to zero impact and will actually frustrate authors like you and I who use AI to improve our thoughts but who are now being tagged as AI-assisted writers.
Thanks for emphasizing that point. It is 100% accurate.
Ai purist are unfortunately some of the same people that will get left behind in other relams of life including work. AI is a tool just like hiring a editor is a tool. Personally i use both, but I hire editors when it is really important. I will always support hiring human talent, but do i really need a editor for a 200 word email?
The shaming of people for using AI but not an editor is interesting.