Insight Forge LLC

The other day I opened a substack about a topic that interests me and noticed that the facts were wrong. Like, VERY wrong. But it wasn’t sloppy writing or research that was the problem. The entire post had obviously been written by AI. The AI hallucinated “facts” that made no sense. I only spotted it because I was very familiar with the topic. I considered posting an angry comment about the shoddiness of it all but in the end, I just silently unsubscribed. If you’re gonna use AI, you have to be VERY careful. Just publishing the nonsense it produces will inevitably make you look like an idiot. Or worse, a fraud.

Recently, I discussed something along the idea of unintended side effects of technology. The great example we mentioned was a recent revelation that law enforcement are sometimes towing Teslas even though neither the car or the driver are suspected of any crime. The reason is that Teslas are equipped with an outward facing security camera. Authorities grab the cars if there has been a crime nearby that _might_ have been captured on camera. This could be construed as an endorsement of the high quality of Tesla security measures. Unfortunately, it is also very much a sign that owning the cars could have unexpected inconveniences attached. So now, I find myself wondering what other types of "inconvenient side-effects" there might be besides having the police take an interest in your product just because it was there. And, how can you get ahead of these things without becoming so risk-terrified that you stop innovating (Please pardon me for using that word. This is one of those times when it fits!)?? I have a few ideas. Really I do! Stay tuned.

Thinking About Unintended Side Effects of Technology I’ve been thinking a lot about unintended side-effects of technology. I don’t mean unintended?consequences?like hackers stealing all our personal information off of some website that was supposedly secure. Those things are at least known and somewhat predictable. I mean things that were absolutely not part of the design or service but somehow became an unexpected problem for someone. There was a story a few months back that really demonstrated the principle.?Cops Are Starting To Tow Away Teslas To ‘Secure’ Recordings Captured By The Cars’ Cameras. Imagine the scenario. You park your car. You go in to a nice restaurant for a nice dinner. When you come out, police are swarming around because there was a crime, maybe a mugging or a murder, near the restaurant while you were inside. More ... https://lnkd.in/gEXdJbpy

There have been some changes that necessitate some more changes. Since this consulting company was not paying the bills, it was necessary for me to take a regular job. It's a software development job for a company that is dedicated to using AI widely and responsibly. Consequently, I feel that it would be a conflict of interest for me to continue offering the consulting services I've been offering. So I'm going to stop doing that. If you were thinking of engaging Insight Forge to assist your AI efforts, I apologize. I will not be accepting any such engagements for the duration of my time at this new job. I'm not planning on leaving that job, but reality is that nothing lasts forever. Maybe someday, I'll come back to this. HOWEVER, Insight Forge is not going away. I have always said that this company is dedicated to smarter technology choices, better planning, innovative thinking. For the time being, what that means is that this effort will continue as a think tank, writing and publishing some white papers. I will probably sell them for a few dollars because this is supposed to be a profitable business. You can't have profit if you don't charge money. Narrowing the range down to thinking, writing, and research (which I've done before) gives a little freedom to consider new ideas, too. While AI is at top of mind lately, there's no reason to restrict my thinking just to that. Also, I figure if someone wants to pay me for specialized thinking and writing (meaning, writing a white paper for them), that would be fine as long as I can do it on my own time. When I'm not at my day job, I mean. This is the reality. At least for now, the job has to come first. There will also continue to be posts here on here. It's fun, so why not? Stay tuned for more and thanks to all who have been interested or who have engaged with Insight Forge in such interesting ways! Original announcement here: https://lnkd.in/g8FQJxvY

I was just reading an article describing how someone (apparently accidentally) got ChatGPT to dump its system instructions (https://lnkd.in/gyGzWcih). They said, "Hi." From the sound of it, it didn't take long for them to fix the bug. I don't think that's important. What's important is, why did it exist in the first place and how many more crazy-simple disclosure bugs are there? Also, did they fix it with a new system instruction saying, essentially, stop that? These questions are fascinating to me.

Is AI in a tech bubble? A tech bubble is similar to a financial bubble and often occurs together with one. It is when a technology goes through a huge spike in adoption even when it may be immature or unsuited for many of the uses it is put to. Hadoop had a tech bubble a while back. Shortly after that, nosql databases were plugged in to almost every new product, whether it made sense or not. Now we have AI-powered everything. It gives frequent wrong answers, eats up a ton of resources and can be very hard to use. Every day sees publication of some new prompting technique that probably works for some use cases. But the use cases aren't always well defined and, anyway, who has time to read every new prompt-related paper that comes out AND try them out to see if they work for you? Yeah. It smells like a tech bubble to me. It seems like a big realignment/reassessment is coming. Then we just have to wonder, what will the next bubble be?

With the release of Claude 3 we've entered a new era of LLM capabilities. Or at least we have if you believe all the hype. I don't. New on the Insight Forge Blog: https://lnkd.in/grqE8p9r