I’m still seeing too many posts out there on how ChatGPT is your friend and how it’s the biggest productivity hack ever and this has to stop.

It’s Not Your Friend. It doesn’t care if you live or die and will not only urge you to give into any and all suicidal thoughts, but provide how-to guides for you.

It’s not a productivity hack, because it’s, at best, a drunken plagiarist intern and you have to review everything it produces. Moreover, you have no clue if it’s mostly right, partially right, or a complete work of fiction. I was reminded of this the other day talking to a tech guru on a legal team that asked their LLM about what laws might impact a new contract with a new supply chain setup in the affected regions, and the LLM came back with three laws, complete with full text and author/site citations, that the team should review for relevance. Upon digging into each law, they found that none of the laws were real, and everything was completely fabricated — the laws, the citations, and sometimes even the bios/bodies who supposedly wrote them! It didn’t matter how much careful prompting they gave it, how specific the request was, and how much time went into building the request and allowing the LLM to do its thing, at the end of the day, it still made everything up and wasted a few days of the team’s time — forcing them to start from square one the old fashioned way and do it all over.

This is why, despite every consultant’s claim to the contrary, ChatGPT CAN NOT create a good draft of an RFP simply from a template and a product or service identification. In the best case, all it can do is just repeat the suckage found in the majority of RFPs in the data set it has been found on. If there weren’t enough RFPs to train its probabilistic predictors, it’s going to make stuff up that, unfortunately, sounds really good because the models, for the first time, capture the intricacies of putting words together to not only form proper linguistic utterances but do so in the common vernacular, which means it sounds human, smart, and right when its still a machine, dumb, and wrong. And no matter how good the “reasoning” seems to be, it won’t be based on logic and will be wrong and circular, but it will take more deftness and effort to catch the mistake than just write your own RFP from scratch. (Now, this doesn’t mean that SLMs can’t be trained to help you in RFP construction and reduce your workload 80% to 90% of the time, just that LLMs can’t be used and even with SLMs, a human will still need to be heavily involved in the process and review. However, to date, I’ve only seen ONE company get this right so far, and have only talked to two or three more that are on the right track.)

I tried to address this in my Best Practice Tech Selection reprise, my How to Write a Good RFP, and my Bells and Whistles Lead to Cells and Thistles series, but apparently I’m not clear enough as the LinkedIn influencers and the consultants and analysts they are influencing still aren’t getting it!

ChatGPT can’t do your work because it is NOT intelligent. All you are accomplishing is dirtying our atmosphere and denying our fellow humans clean water so that you can power queries (and keep the machines from overheating) that take 20 times (or more) the processing power of a Google Search and aren’t guaranteed to return any usable results (while allowing your cognitive abilities to atrophy through over-reliance on dumb tools). (We should not see stories about how I Can’t Drink the Water in the richest and most powerful nation in the world [because of a data centre]! It’s shameful! But we are now seeing these stories, along with “I don’t have water!” stories because data centers are now consuming over a trillion gallons of freshwater globally, a resource we are running dangerously low on in many countries. Half of the USA is already suffering from water scarcity issues! And you’re literally making it 20 times worse thanks to your ChatGPT addiction!)

For an RFP, it’s not a high level bill of materials, feature / function / support checklist, or detailed profiles of what you think you need — it’s processes you need to support, capability gaps you are missing, and skills that you need augmented. ChatGPT, or any LLM doesn’t know that! Only YOU know that. For many other tasks that require human intelligence to figure out, it’s the same story — ChatGPT doesn’t know, makes stuff up, and gives you suckage.

Moreover, you can’t trust it for deep data analysis. It has been demonstrated to get basic math wrong many times (or, when pushed to find savings, multiply a result by -1 and lie to you). It can usually compute directionally accurate results, but that’s it. But we’ve also seen many instances where the EXACT SAME QUERY was asked twice in a row on a data set that did not change, and it gave two different answers. Even the dumbest drunken plagiarist intern would say “I just told you the answer you nitwit. It’s this!” Moreover, right or wrong, the dumbest drunken plagiarist intern would repeat the same answer. (It’s so bad that even Gartner has projected that conversational analyticswill fall off of its hype cycle within 2 years!)

Furthermore, it is not intelligent, and has no brain, so it cannot brainstorm … the best it can do is serve up other people’s ideas you may not have heard about! There’s a reason you will not have thought of some of the ideas it brings back, and the reaason is the ideas it will bring back are so ridiculously stupid (and obviously wrong) that only a complete and utter moron would give them a second thought.

It’s a fun, planet destroying, toy that will always hallucinate, because that’s part of its core design, and that may or may not give you something useful on any given query. So if you have to manually verify everything it does, how can it be worth using?

And yes, it really does destroy the planet compared to classic Google Searches. This YouTuber does a great job of explaining, in plain English, How AI is Impacting the Planet for those of you who refuse to process the written word I keep presenting to you.

But if you don’t mind planet killing, or a technology tool that will expose your entire conversation history and confidential/trademarked/top-secret corporate data to the whole internet, then be my guest and use it. It’s your business. Feel free to flush it down the toilet if you like. Not my place to tell you not to.

I’m just here to remind you that ChatGPT is NOT your friend! (And neither is any open LLM!)