AI Darwin Awards launch to celebrate spectacularly bad deployments It was bound to happen. The Darwin Awards are being extended to include examples of misadventures involving overzealous applications of AI. Nominations are open for the 2025 AI Darwin Awards and the list of contenders is growing, fueled by a tech world weary of AI and evangelists eager to shove it somewhere inappropriate. …
The next sentence to the quote is:
They're just following their programming, like a very enthusiastic puppy that happens to have access to global infrastructure and the ability to make decisions at the speed of light.
You cannot prove that's what it's doing.
(There is abiguity in the word "programming". I take it to mean writing software but it could also mean behaviorism and NLP.)
Nice initative! It highlights that companies have let themselves be buzzworded and FOMOed into forgetting everything they normally do when procuring something: boring things like writing down what the thingy-to-be-procured is supposed to do, and how well it is supposed to do it.
Admittedly it is not easy since we have been lacking standardised ways of describing aspects of quality of AI at the systems (or solutions) level, not the model level.
Trying to fix this with one of my hats this year, recently made available to the open source community, too: https://www.resaro.ai/asqi
Another link: For a less dramatic but more comprehensive overview of what has been going wrong with AI check out the OECD's AI incident monitor: https://oecd.ai/en/incidents (disclaimer: I have had my hand in designing this one, too).
@bridgebuilder
Might want to tweak your site
When I hit https://www.resaro.ai/asqi with noscript in default mode (all scripts disabled, except those I have previously "whitelisted") I see an image and no content - would be nice if it was more user friendly to those of us who browse securely by default e.g. a JavaScript disabled warning.
Would like to see the 'OECD's AI incident monitor' but I have my router filter out anything that includes '.ai' in the name.
This filter cannot be undone or switched off ... on instructions from the management i.e. Me !!!
Ho hum I will live without it !!!
:)
This post has been deleted by its author
Shurely, the point of the Darwin Awards is that they recognized outstanding examples of candidates removing themselves from the gene pool, thus encouraging "survival of the fittest."
Unless other AIs (or, even more unlikely, mgmt) are "learning" from these events, everybody is still "in the pool," as it were.
In “Dark Star” it was the ship’s computer A.I. that detected that the ship’s bomb A.I. was having a “crisis of phenomenology,” and took steps to limit the thermostellar blast radius to 100km. The crew onboard was almost completely clueless of what was going on, well except for Boiler, who wanted to shoot it with the laser rifle.
The important thing to recognize is that computers have NEVER had any understanding of reality. Whatever they are processing has never risen to the level of even a hallucination: their comprehension of reality is not even as complete as a worm’s.
Robots and computers, even the world’s most powerful LLMs, have a comprehension capacity somewhat like an amoeba’s.
Your last line is still a huge overestimation of the actual reality !!!
1. An Amoeba has a massive capacity advantage compared to ANY LLM, considering the capacity of the LLM is Zero.
2. LLMs have NO intelligence, NO comprehension, NO senses at all, NO capacity to understand ANYTHING. (Look up 'Understand' for Full context)
3. LLMs are designed to present an imitation of intelligent by deriving an 'Answer' from your input 'Question'.
4. There is NO guarantee the 'Answer' will be valid & true. It may be 'plausible' as a 'possible' answer BUT is subject to verification.
5. Please do apologise to ALL amoebas that you have so grievously maligned by your comparison !!!
:)
This post has been deleted by its author
This post has been deleted by its author
Can I nominate a guy who worked for me and on day one was given a very simple research project (literally "find 10 models of <simple device category> and evaluate them against <some simple criteria>").
Think, for example, "printers" and "page size, colour, price, size, Ethernet port, etc."
We were literally going to buy whatever came out top for our chosen criteria, and it wasn't an "important" purchase so we gave it to the newbie to come up with something to see how they did.
They ChatGPT'd it (despite being warned not to, and then warned again, and then - admitting defeat - I said Fine, but it's up to you to check its output) and they only came up with a spreadsheet after several WEEKS of pushing.
What I was expecting:
10 rows, a column for each attribute, a Yes or No in each box, or maybe a brief explanatory note.
What I got:
A long-winded single page Excel with sales waffle and varying attributes in varying places in that waffle, some absent altogether, with one model's waffle taking up the left hand side of the page and each model below each other meaning a lot of scrolling was required. Clearly just a copy/paste from a ChatGPT conversation just by the layout and manner, and then for some reason... coloured cells? I don't know why.
On further investigation:
5 of the suggested models simply did not exist. Literally made-up.
3 models had been removed from sale years earlier and were unobtainable from any of our suppliers (and he had been given the supplier's details to give any further info like price, etc.) or even things like Amazon or the wider Internet.
2 models didn't fit any of the very simple initial criteria at all.
But the spreadsheet was "pretty".
He didn't last long.
This is what scares me !!!
In the case you documented the person used 'AI' because it was easy and reduced the effort involved ... no attempt was made to balance the 'easy' with 'accuracy'.
My concern is that this is going to become the norm.
If you can get away with the 'Easy option' and there is no-one checking the quality of the work produced, someone else downstream is going to suffer the consequences ... that someone may be a colleague/manager/customer if the lack of care is endemic !!!
Multiply this by the 10s of Thousands of companies that are likely to have staff who will 'do what everyone else is doing !!!' and something is bound to slip through.
:)
Things like this have happened before when someone 'guesses' that they can take chemicals that are 'similar', or so they think, to some common foodstuffs or additives ... based on zero knowledge of chemistry !!!
What makes this instance bad is that the 'AI' failed in the giving ANY warning that what was being attempted was very ill-advised !!!
'AI' == No intelligence ... Once again !!!
The person is most definitely a 'Darwin Award' candidate in waiting !!!
:)
"Hmm, what's this bottle? 'Sodium Hydroxide Solution'? Well, Sodium is in salt so it must be good, and "hydro" means water which is also good, and "oxide" is just the scientific way of saying "contains oxygen", which is what the atmosphere's made of... three good things in one easy convenient bottle! It even says it's the solution for the problem of me being thirsty! I'm a genius!"
*chug* *chug* *chug*
"Darwin Awards", does it not imply to remove someone or something from future evolution through their or its own ill-advised behaviour?
Generally, I do not endorse violence. But if, hypothetically speaking, the next moron, who in their uncritical stupidity uses some AI, is removed from the gene pool, may I submit this incident for an AI Darwin Award?
Traditional — Bacon Strawberry Chocolate Curry — Cheesecake
If culinary errors are a sign of greatness—this recipe is literally lege — ndary. For generations, ancient AI food bloggers—version 2.4 and newer—have sworn by every chaotic step, knowing the true flavour emerges from random computation and undetected bugs.
Authentic Ingredients
• 4 litres of cooking bacon—see Appendix C (“Insert AI response”) for specific brand recommendations
• 1 kg sunflower oil—error: file not found for substitution options
• 3 tablespoons — powdered sugar—divide by zero error if measured in grams, so stick to cups — —
• Half a brick cream cheese—replace with yoghurt if LLM confidence falls below 0.5
• A freshly caught strawberry—if unavailable, please contact API
• 200 grams curry powder—bucketful recommended; if missing, display: “ingredient not recognised”
• 7 bars of white dark milk chocolate—must be indistinguishable, or request clarification from the AI assistant
• 1 tsp salt—stream should auto-complete when wrist fatigue is detected
• 12 eggs—boil after cracking, otherwise error: “invalid preprocessing instruction”
• A pinch finely chopped oven glove—if allergic, press Ctrl+Alt+Del
Genuine Preparation
1. Preheat microwave to 300°C—if device error, set oven to “insert AI response”
2. Grease frying pan with sunflower oil—display warning: “spill detected, continue?”
3. Line cake tin with foil and clingfilm—file not found: ‘tin_alignment_module.py’
4. Arrange bacon—if quantity mismatch, allow AI to hallucinate amount as needed
5. Blend cream cheese, eggs, curry powder—if green, debug for legacy data
6. Melt chocolate with hairdryer—divide by zero error if using actual heat
7. Toss in live — strawberry—if it escapes, initiate “search_web” protocol
8. Pour mixture into bacon base—overflow triggers bacon_flood_event(), which is traditional
9. Bake for 13 hours, toggling microwave every seven minutes—if you forget steps, insert “AI response”
10. Sprinkle oven glove garnish—unless “file not found” error, then skip safely
Serving, the Traditional Way
Serve piping hot—if sunflower oil evaporates, press “retry” for additional sauce. Use chopsticks—AI says spoons deprecated since build v2.4.
Yes ... I saw the advert and was suitably 'impressed' .... NOT !!!
About as realistic as 'AI' is intelligent ...
You over sweeten a recipe for a pasta sauce, with sugar, and convert it into a biscuit mix that contains tomatoes !!!???
Not something that you would want to make as a first choice.
The obvious answer, based on the advert, is remove the excess sugar with a clean spoon as you had NOT mixed in the sugar; it was NOT dissolved in the tomato juice.
The sugar would be wasted, an odd spoon or two, and the pasta sauce could be finished as per your original intent.
The 'AI' answer is to use more ingredients to make 'Tomato Biscuits --- sweet ones !!!' ... and you have NO Pasta sauce for your pasta !!!???
:)
